paper symbol (put here the unique symbol of your …epubs.surrey.ac.uk/812574/1/manuscript -...
TRANSCRIPT
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
Soft-Sensing Method for Optimizing Combustion Efficiency of Reheating Furnaces
Jian-Guo Wang1 Tiao Shen1 Jing-Hui Zhao1 Shi-Wei Ma1 Xiao-Fei Wang2 Yuan Yao3 Tao
Chen4
1 School of Mechatronical Engineering and Automation Shanghai University Shanghai Key Lab
of Power Station Automation Technology Shanghai 200072 China
2 Shanghai Zhenhua Heavy Industries Company Limited Shanghai 200125 China
3 Department of Chemical Engineering National Tsing Hua University Hsinchu 30013 Taiwan
4 Department of Chemical and Process Engineering University of Surrey Guildford GU2 7XH
United Kingdom
Abstract
Rolling mill reheating furnaces are widely used in large-scale iron and steel plants
the efficient operation of which has been hampered by the complexity of the
combustion mechanism In this paper a soft-sensing method is developed for modeling
and predicting combustion efficiency since it cannot be measured directly Statistical
methods are utilized to ascertain the significance of the proposed derived variables for
the combustion efficiency modeling By employing the nonnegative garrote variable
selection procedure an adaptive scheme for combustion efficiency modeling and
adjustment is proposed and virtually implemented on a rolling mill reheating furnace
The results show that significant energy saving can be achieved when the furnace is
operated with the proposed model-based optimization strategy
Keywords soft-sensing variable selection data-driven statistical analysis reheating
furnace combustion efficiency
1
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
Correspondence information
J G Wang Tel +86-21-56331278 Email jgwangshueducn
Y Yao Tel +886-3-5713690 Email yyaomxnthuedutw
2
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
1 Introduction
The reheating furnace for a rolling mill is the most energy consumption equipment in
a large-scale iron and steel plant thus it is of great significance to improve the
combustion efficiency and reduce gas consumption [1 2] Since combustion efficiency
cannot be measured directly the adjustment of oxygen content in the exhaust gas is
often used to indirectly control the efficiency Another method is to estimate
combustion efficiency based on oxygen content in the exhaust gas and then implement
control actions [3] However the performance of these methods depends on the
precision and stability of oxygen analyzers which are susceptible to corrosion and wear
of high-temperature gases and difficult to maintain in full operational status for a long
period of time
When quality variables cannot be easily obtained a soft sensor model that can predict
these quality characteristics (as response variables) using readily available sensor
variables (as candidate predictors) will be most desirable A variety of soft sensor
methods and applications have been studied in different fields [4-7]
For a reheating furnace a number of soft sensors have been investigated The feature
of continuous prediction of temperature and composition of the combustion atmosphere
has the potential of acting as a soft sensor thereby leading to a reduced number of
temperature measurements and sampling for chemical analysis [8] The secure
economic and stable control of the combustion process is realized by the cooperation
work of a cascade fuzzy control system for furnace temperature a ratio control system
for air flow with a soft-sensing model plus a fault diagnosis model [9] A data-driven
soft sensor modeling technique for furnace temperature of the Opposition Multi-Burner
(OMB) gasifier is proposed and the selection of secondary variables and model structure
3
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
of a back propagation (BP) neural network is studied which indicates that the furnace
temperature predictive model integrating the principal component analysis (PCA) and
the BP neural network has a promising performance with good predictive precision
[10] A soft sensor modeling method is proposed to predict the billet temperature of the
reheating furnace based on a relevance vector machine (RVM) which has a higher
prediction accuracy and a certain practical significance to the on-site production of a
reheating furnace [11] The least square support vector machine (LSSVM) inductance
model optimized by the particle swarm optimization method with a compression factor
(PSO-CF) algorithm is presented for the difficulty of time prediction which can
improve PSO convergence accuracy and effectively avoid falling into a local optimum
[12] However the soft sensor developed for combustion efficiency was not
investigated in these research efforts which is significant for energy conservation
On the premise of the model prediction accuracy the model-based control makes
optimal operation feasible which can then be successfully employed to operate a
reheating furnace in an efficient way The potential of the nonlinear model predictive
control techniques is explored to improve the temperature control for the metal slabs in
a hot mill reheating furnace and particularly whether or not these control techniques
can be exploited to reduce energy consumption [13] Steinboeck et al developed a
mathematical model of the reheating process of steel slabs in industrial fuel-fired
furnaces in 2010 They exploited a dynamic optimization method for temperature
control of the steel slabs in a continuous reheating furnace and a temperature control
method for reheating steel slabs in an industrial furnace in 2011 They also designed a
nonlinear model predictive controller for a reheating furnace for steel slabs in 2013 [2
14 15]
4
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
Obviously the research on the numerical model for the heating performance of
reheating furnace can be done based on basic combustion theory and heat transmission
characteristics Many scholars devote themselves to the simulations of the heat flow
phenomenon in the reheating furnace Zhang et al attempted to apply a computational
fluid dynamic (CFD) simulation to predict the combustion performance for a reheating
furnace by simplifying the furnace to a cuboid and assuming that the slab possesses
infinite length and enters the reheating furnace at a fixed speed [16] The CFD method
has been applied to the study of reaction turbulence radiation heat transmission and the
calculation for the steady state heat transmission rate of the slab under the given
temperature [17] In the works of references [16-20] the authors used the given
temperature data of slabs to compute the steady flow and temperature field However
owing to the changing operating conditions the actual implementation of these
numerical model methods still bristle with difficulties although the methods mentioned
above are feasible for the prediction Thus for an online application it is necessary to
adopt a real time data-driven model to resolve the time varying characteristics
Proper variable selection is an important step in model building for a large-scale
combustion system A well-trimmed variable dimension ensures the acquired model is
transparent comprehensible and robust Some studies reported that the combustion
model built by a selected subset of input variables provide more accurate predictions of
combustion efficiency than the entire set of variables [21-23] Recently shrinkage
methods which conduct variable selection by shrinking or setting some coefficients of a
ldquogreedyrdquo model to zero have received significant attention A popular form of these
methods is the non-negative garrote (NNG) [23 24]
5
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
Against this background this paper aims to propose a combustion efficiency index
for the reheating furnace and investigate for room in improvement regarding energy
conservation The primary contribution is a practical combustion efficiency index the
incorporation of the derived variables and soft-sensing method for the optimization of
combustion efficiency of reheating furnaces The derived variables are found more
physically meaningful than the plain variables when constructing the model of
combustion efficiency By employing a NNG variable selection procedure an adaptive
scheme for combustion efficiency modeling and adjustment is proposed and virtually
implemented for a rolling reheating furnace The results show that there is significant
room for energy conservation
The remainder of the paper is organized as follows In the next section the reheating
furnace and the data preprocessing is described In Section 3 the statistics analysis for
different variables and the formation of derived variables are presented In Section 4
the framework of an adaptive model based on NNG variable selection is presented and
two models developed for the temperature and temperature-gas (TG) ratio are
compared according to the model prediction precision A model-based optimization
scheme is provided and applied to the combustion efficiency improvement for an actual
case of a reheating furnace presented in Section 5 Several remarks and a summary
conclude the last section
2 Plant description and data preprocessing
The schematic of the heating process in the rolling mill reheating furnace is shown in
Fig 1 There are four zones in the reheating furnace including the preheating zone (P)
the first heating zone (1) the second heating zone (2) and the soaking zone (S) The
6
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
steel slab moves through the four zones in turn and is heated to the demanded state
using a specific temperature increase curve As is shown in Table 1 the soaking zone
has two areas that are defined as up and down and both of the areas possess the same
five variables including two manipulated variables the air flow (A) and the gas flow
(G) and three temperatures in left center and right sections of the area (T-l T-c and T-
r) The other zones have the same variables as the soaking zone hence there are 40
variables in total for the reheating furnace
Fig 1 The schematic of the heating process in the reheating furnace
Table 1 Variables and descriptions in the soaking zone
Variable Description unit
AS-u Air flow in the lsquouprsquo area Nm3h
AS-d Air flow in the lsquodownrsquo area Nm3h
GS-u Gas flow in the lsquouprsquo area Nm3h
GS-d Gas flow in the lsquodownrsquo area Nm3h
TS-ul Temperature in the left part of the lsquouprsquo area
TS-uc Temperature in the center part of the lsquouprsquo area
TS-ur Temperature in the right part of the lsquouprsquo area
7
143
144
145
146
147
148
149
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
TS-dl Temperature in the left part of the lsquodownrsquo area
TS-dc Temperature in the center part of the lsquodownrsquo area
TS-dr Temperature in the right part of the lsquodownrsquo area
A data set of 20000 samples was used in this study The samples were collected from
an actual reheating furnace in a large iron and steel plant located in Shanghai from
September 14 to September 27 2014 The operational data is taken on a per minute
basis
1
092
093
042
041
046
092
1
091
04
039
045
093
091
1
038
038
045
042
04
038
1
096
096
041
039
038
096
1
096
046
045
045
096
096
1
TS-ul
TS-uc
TS-ur
TS-dl
TS-dc
TS-dr
TS-ul TS-uc TS-ur TS-dl TS-dc TS-dr
04
05
06
07
08
09
1
Fig 2 Correlation between each temperature in the soaking zone
In order to investigate the relation among different temperatures in each zone
correlation analysis is conducted for the soaking zone as illustrated in Fig 2 It can be
seen that the temperature in three parts of the lsquouprsquo area is highly correlated with a
correlation coefficient greater than 09 The similar results exist for the lsquodownrsquo area as
well On the contrary the correlation coefficient between temperatures in one part of
8
150
151
152
153
154
155
156
157
158
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
lsquouprsquo area and any part of lsquodownrsquo area does not exceed 05 Therefore for the reduction
of the data dimension the three temperatures in the lsquouprsquo area or the lsquodownrsquo area can be
treated as only one variable which can be taken as the mean value or the first principal
component acquired from the PCA analysis Considering the reservation of the variable
physical meaning the former is preferred and used
3 Statistics analysis and incorporation of derived variables
In this section in order to uncover the physical knowledge for the actual operation
guidance and confirmation statistical analysis is performed for the 16 input manipulated
variables and the eight output variables (ie the temperatures in the four zones) of a
reheating furnace system For the combustion efficiency evaluation and modeling two
types of derived ratio variables are introduced which is helpful to reveal the
information included in the data
31 Correlation analysis
Correlation analysis between the temperatures (T) in each area and all of the air flows
(A) and the gas flows (G) is performed and shown in Fig 3 It can be seen that only the
temperatures in both areas of the soaking zone are the most highly related to the air flow
and the gas flow in its own zone However this phenomenon does not occur in the other
three zones The temperatures in the second heating zone mainly depend on the air flow
and the gas flow in its own zone as well as the nearby first heating zone As for the
preheating zone and first heating zone no apparent correlation can be observed
Obviously these analysis results could not tell us the explicit information about how to
evaluate the efficiency levels and key manipulated variables
9
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
-101
T P-u
AP GP A1 G1 A2 G2 AS GS
AP GP A1 G1 A2 G2 AS GS
AP GP A1 G1 A2 G2 AS GS
-101
T P-d
-101
T 1-u
-101
T 1-d
AP GP A1 G1 A2 G2 AS GS
-101
T 2-u
AP GP A1 G1 A2 G2 AS GS
-101
T 2-d
-101
T S-u
-101
T S-d
AP GP A1 G1 A2 G2 AS GS
AP GP A1 G1 A2 G2 AS GS
AP GP A1 G1 A2 G2 AS GS
UPDOWN
Fig 3 Correlation between air flows gas flows and temperatures
32 Incorporation of derived variables
During the stable heating stage the quantity of heat absorbed and removed from the
slabs from each furnace zone is relatively constant Hence the derived variable TG
ratio can be treated as an index for the combustion efficiency level This is because a
higher TG ratio signifies more combustion heat generated from unit gas ie higher
combustion efficiency
Moreover it is known that the appropriate air and fuel ratio is vital for the
combustion efficiency so the air-gas ratio (AG) is utilized as another derived variable
for the research Again the correlation analysis is performed for two types of derived
variables The correlations between different variables including the AG ratio G and
the TG ratio are shown in Fig 4 It can be clearly seen from the four red rectangle
blocks that TG of each zone is only remarkably related to AG and G in the same zone
10
183
184
185
186
187
188
189
190
191
192
193
194
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
Furthermore as related to the lsquouprsquo or lsquodownrsquo areas in one zone TG in each area has the
highest correlation with AG and G in the same area while AG and G in the opposite
area of the same zone is secondary This can be easily seen from the red and blue color
markings in each red rectangle block Thus it is of great significance to introduce these
derived ratio variables
-101
TG
P-u
AGP GP AG1G1 AG2 G2 AGS GS
-101
TG
P-d
AGP GP AG1 G1 AG2 G2 AGS GS
-101
TG
1-u
AGP GP AG1 G1 AG2 G2 AGS GS
-101
TG
1-d
AGP GP AG1 G1 AG2 G2 AGS GS
-101
TG
2-u
AGP GP AG1 G1 AG2 G2 AGS GS
-101
TG
2-d
AGP GP AG1 G1 AG2 G2 AGS GS
-101
TG
S-u
AGP GP AG1 G1 AG2 G2 AGS GS
-101
TG
S-d
AGP GP AG1 G1 AG2 G2 AGS GS
UPDOWN
Fig 4 Correlation between the air-gas ratio gas flow and TG ratio
33 LDA analysis
Linear discriminant analysis (LDA) aims to finding a projection direction that
maximizes the separation of class means and minimizes the within-class variance [25]
In this section LDA is utilized to identify the discriminating variables that play an
important role in determining combustion efficiency levels All the data are partitioned
into five groups according to their efficiency levels
11
195
196
197
198
199
200
201
202
203
204
205
206
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
Based on a descending order of TG ratios five efficiency levels for the lsquouprsquo area of
the soaking zone are denoted as HH H M L and LL LDA is conducted on the three
groups of the data with levels of HH M and LL
Fig 5 shows the scattering of the LDA projections of the process observations
collected at the three efficiency levels where y1 and y2 correspond to the first two LDA
components that contain most discriminant information The weighting factors ( and
) of the 16 input variables composing the projections y1 and y2 are shown in Fig 6
where and From left to right the 16 input variables are defined as
the eight AG ratio variables and the eight gas flow variables with the order of variables
of each kind P-u P-d 1-u 1-d 2-u 2-d S-u and S-d
-8 -6 -4 -2 0 2 4 6-4
-3
-2
-1
0
1
2
3
4
5
6
Projection y1
Pro
ject
ion
y2
HHMLL
Fig 5 Scattering of LDA projections y1 and y2 for three efficiency levels
12
207
208
209
210
211
212
213
214
215
216
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
AG_S-u-4
-2
0
2
Wei
ghtin
g fa
ctor
for y
1
AG_S-u
-101234
Wei
ghtin
g fa
ctor
for y
2
Fig 6 Weight factors of various input variables for projections y1 and y2
The two figures reveal that the three groups are clearly discriminated by the LDA
projection and the most significant variables for the different efficiency levels are the
derived variables ie the AG in the lsquouprsquo area of the soaking zone A similar result can
be obtained in the lsquodownrsquo area or any area of the other zones Therefore AG in each
heating area is the key manipulated variable that determines the different combustion
efficiency
4 Modeling and prediction of temperature and temperature-gas ratio
For the model-based operation optimization the models for the temperature and TG
ratio based on the NNG algorithm will be developed and compared
41 NNG variable selection algorithm
The NNG method can be generalized into a two-stage shrinkage method In the first
stage the sign for each variable is determined using the ordinary least square procedure
13
217
218
219
220
221
222
223
224
225
226
227
228
229
230
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
and in the second stage the corresponding magnitudes are computed by solving a series
of constrained quadratic programming
A set of observation data is provided where is the input matrix whose
columns represent the measured candidate variables and is the corresponding
vector of the response data The following expression is given with the number of the
response variable being equal to 1 but a similar procedure can be generalized to any
number of variables Let and be normalized to the zero-mean and unit standard
deviation Additionally let be a set of the ordinary least square estimates of the
coefficients of the following linear model then
(1)
The second stage shrinkage can be formulated as the following optimization problem
subject to
(2)
As decreased and the NNG is tightened more of the become zero and the
remaining nonzero coefficients are shrunk A solution path exists with on which
the appropriate shrinkage can be selected Conventionally the v-fold cross-validation is
used to estimate the prediction error and to select the best solution in the solution path
so as to minimize the prediction or model error
42 Modeling and prediction of temperature and TG ratio
The input-output relations change with time as the reheating process develops An
adaptive modeling strategy is often used to resolve time-varying characteristics of
14
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
industrial processes In this paper the NNG-based regression modeling prediction and
optimization is implemented in a moving window manner where the size of the
window and the length of the moving step are selected as 1000 and 100 respectively In
addition the prediction horizon is also set as 100 In each step the NNG regression
model is built based on the data in the current window which is then used for the
prediction and optimization in the subsequent prediction horizon Next the window
moves forward by replacing the oldest 100 samples for model training by the
observations collected in the previous prediction horizon This moving window strategy
is workable because in the reheating process the input-output relation is slowly time-
varying and the model is still valid for the prediction and optimization in the subsequent
short time
The modeling of the temperature T is based on the air flow rates and gas flow rates
while the modeling for the TG ratio is based on the AG ratios and gas flow rates
Taking the lsquouprsquo area of the soaking zone as an example the prediction result for T and
the TG ratio in the lsquouprsquo area of the soaking zone is shown in Figs 7 and 8 respectively
0 02 04 06 08 1 12 14 16 18 2
x 104
1140
1160
1180
1200
1220
1240
1260
1280
1300
1320
Sampling intervals
T S-u (
)
OriginalPredicted
Fig 7 Prediction of temperature in the lsquouprsquo area of the soaking zone
15
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
0 02 04 06 08 1 12 14 16 18 2
x 104
02
04
06
08
1
12
14
16
18
2
Sampling intervals
TG
S-u
OriginalPredicted
Fig 8 Prediction of TG ratio in the lsquouprsquo area of the soaking zone
A comparison between Figs 7 and 8 show that the prediction of the TG ratio is more
accurate than the prediction of temperature The average R2 in 190 NNG regressions is
0935 for the TG ratio while the average R2 is only 0814 for temperature This is
consistent with the statistical analysis which reveals that the correlation between AG
and TG is higher than that between A and T These results indicate that derived
variables are more meaningful for the purpose of prediction and the modeling of the
TG ratio is more appropriate for implementing optimization
The selected frequency for each variable in 190 NNG regressions and the coefficients
of each variable in 20 NNG regressions for the TG ratio modeling of the lsquouprsquo area of
the soaking zone are shown in Figs 9 and 10 respectively Fig 9 shows that the selected
frequency of the variables in the lsquouprsquo area of the soaking zone is much higher than
variables in other zones Similarly as is shown in Fig 10 the NNG regression
coefficients of the variables in the lsquouprsquo area of the soaking zone are much larger than the
variables in the other areas These results indicated that variables in the lsquouprsquo area of
16
269
270
271
272
273
274
275
276
277
278
279
280
281
282
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
soaking zone are mostly contributed to the modeling of the TG ratio in same area
which is quite reasonable Similar results can be obtained for the other areas
0
20
40
60
80
100
120
140
160
180
200
Sel
ecte
d fre
quen
cy
AGP-u AGP-d AG1-u AG1-d AG2-u AG2-d AGS-u AGS-d GP-u GP-d G1-u G1-d G2-u G2-d GS-u GS-d
Fig 9 Variable selected frequency of over 190 runs for the TG ratio model of the lsquouprsquo
area of the soaking zone
0 2 4 6 8 10 12 14 16 18 20-05
05
AG
P-u
0 2 4 6 8 10 12 14 16 18 20-05
05
GP
-u
0 2 4 6 8 10 12 14 16 18 20-1
0
AG
P-d
0 2 4 6 8 10 12 14 16 18 20-1
0
GP
-d
0 2 4 6 8 10 12 14 16 18 20-05
05
AG
1-u
0 2 4 6 8 10 12 14 16 18 20-05
05
G1-
u
0 2 4 6 8 10 12 14 16 18 20-05
05
AG
1-d
0 2 4 6 8 10 12 14 16 18 20-05
05
G1-
d
0 2 4 6 8 10 12 14 16 18 20-1
0
AG
2-u
0 2 4 6 8 10 12 14 16 18 200
1
G2-
u
0 2 4 6 8 10 12 14 16 18 20-1
0
AG
2-d
0 2 4 6 8 10 12 14 16 18 20-1
0
G2-
d
0 2 4 6 8 10 12 14 16 18 200
1
AG
S-u
0 2 4 6 8 10 12 14 16 18 20-1
0
GS
-u
0 2 4 6 8 10 12 14 16 18 20-05
05
AG
S-d
0 2 4 6 8 10 12 14 16 18 20-05
05
GS
-d
Fig 10 Part of the NNG regression coefficients of each variable over 190 runs for the
TG ratio model of the lsquouprsquo area of the soaking zone
It is remarkable that the NNG regression coefficients corresponding to the AG ratio
in the lsquouprsquo area of the soaking zone are consistently positive On the contrary the
coefficients of the gas flow rate in the same area are consistently negative This
indicates that under this condition the rise of the AG leads to the increase of the TG
17
283
284
285
286
287
288
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
while the rise of the G leads to the drop of the TG This property is helpful for
performing optimization and improve combustion efficiency
For comparison the modeling results for two other algorithms artificial neural
network (ANN) and partial least squares (PLS) [26] are compared with the NNG
algorithm in the following three aspects [27]
(1) Model Size the number of variables selected for modeling
(2) Model Magnitude the mean of the L1 norm of the regression coefficients
(3) Prediction Precision the mean squared prediction error (MSPE)
Summary of the algorithm comparison is shown in Table 2 It can be seen that the
superiorities of the NNG regression in model size model magnitude and model
precision are remarkable
Table 2 Summary of algorithm comparison
Index NNG ANN PLS
Model Size 849 16 16
Model Magnitude 10075 -- 12150
MSPE 00093 00108 00140
5 Model-based optimization
51 Implemention of model-based optimization operation
The goal of optimization is to seek an optimal combination of AG and G in the lsquouprsquo
area of the soaking zone so as to minimize the gas consumption while keeping the
temperature at the target value According to the heating schedule the most expected
temperature in the soaking zone is 1200 Therefore the target temperature
is set at this value to achieve the desired heating effect As shown in Fig 11 in the real
18
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
operation the temperature fluctuates around the target value because of the imperfect
control performance However in order to simplify the analysis and compute the
maximum possible energy saving perfect temperature control is assumed when
adopting the model-based optimization strategy In other words it is assumed that the
actual temperature in the lsquouprsquo area of the soaking zone can be adjusted to the expected
temperature ie 1200
0 02 04 06 08 1 12 14 16 18 2
x 104
1150
1200
1250
1300
Sampling intervals
T S-u (
)
OriginalTarget
Fig 11 Original and target temperature in the lsquouprsquo area of the soaking zone
The adjustment scheme takes the maximum value and minimum value of the original
operation data as the upper and lower bounds for the adjustment Moreover in order to
assure the validity of the linear model the increment or decrement of AG and G should
not beyond of the original value (considered as 10 for the purposes of this paper)
The adjustment strategy can be formulated as
19
307
308
309
310
311
312
313
314
315
316
317
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
(3)
where NNG() denotes the NNG regression model of the TG ratio and are the
NNG regression coefficients of the AG and G in the lsquouprsquo area of the soaking zone for
modeling TG in same area
With the model-based strategy and adaptive model of the TG ratio given in Section
42 the optimization adjustment results can be obtained as follows
The adjustment of gas flow in the lsquouprsquo area of the soaking zone shown in Fig 12
illustrates that the implementation of the model-based optimization operation can reduce
the consumption of the gas flow Compared with the original case 938 of the gas can
be saved on average by utilizing the model-based adjustment
0 02 04 06 08 1 12 14 16 18 2
x 104
-800
-700
-600
-500
-400
-300
-200
-100
0
Sampling intervals
Adj
ustm
ent o
f gas
flow
m(
3 h)
20
318
319
320
321
322
323
324
325
326
327
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
(a) Adjustment amount of gas flow
0 02 04 06 08 1 12 14 16 18 2
x 104
500
1000
1500
2000
2500
3000
3500
4000
4500
5000G
as fl
owm
(3 h
)
Sampling intervals
OriginalAdjusted
(b) Comparison of original and adjusted gas flow
095 096 097 098 099 1 101 102 103 104 105
x 104
500
1000
1500
2000
2500
3000
3500
4000
4500
5000
Gas
flow
m(
3 h)
Sampling intervals
OriginalAdjusted
(c) Comparison of original and adjusted gas flow in interval [9501-10500]
Fig 12 Adjustment of gas flow in the lsquouprsquo area of the soaking zone
The adjustment of the air flow in the lsquouprsquo area of the soaking zone given in Fig 13
shows that the air flow is reduced in most cases and only increased for a few cases In
21
328
329
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
general 681 of the air flow is reduced after the implementation of the model-based
optimization
0 02 04 06 08 1 12 14 16 18 2
x 104
-2000
-1500
-1000
-500
0
500
1000
Sampling intervals
Adj
ustm
ent o
f air
flow
(m3 h
)
(a) Adjustment amount of air flow
0 02 04 06 08 1 12 14 16 18 2
x 104
4000
5000
6000
7000
8000
9000
10000
Air
flow
m(
3 h)
Sampling intervals
OriginalAdjusted
(b) Comparison of original and adjusted air flow
22
330
331
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
095 096 097 098 099 1 101 102 103 104 105
x 104
4000
5000
6000
7000
8000
9000
10000
Air
flow
m(
3 h)
Sampling intervals
OriginalAdjusted
(c) Comparison of original and adjusted air flow in interval [9501-10500]
Fig 13 Adjustment of air flow in the lsquouprsquo area of the soaking zone
52 Discussions
In this section detailed analysis for the optimization operation results is provided
The statistics of optimization at the boundary conditions shown in Table 3 illustrates
that the adjusted value for G reaches its lower limit in most cases (8045 for
and 687 for ) which is the most energy-efficient point In a number of cases
(1079 for and 001 for ) the adjusted AG reaches its upper
limit These results indicate that the optimization operations maximize the combustion
efficiency by decreasing G and increasing the AG ratio up to the boundary conditions
However the lower limit is also unexpectedly attained in a small number of cases
(110 for and 078 for ) for which a more detailed analysis is
given below
23
332
333
334
335
336
337
338
339
340
341
342
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
Table 3 Statistics of optimization at boundary conditions
Total
Amount 0 1374 0 16090 17464
Percentag
e 0 687 0 8045
8732
Total
Amount 2 220 2158 156 2536
Percentag
e 001 110 1079 078
1268
Table 4 Statistics for different optimization operations
Adjustment
Gdarr
Guarr TotalAGuarr AGdarr
Tdarr Tuarr Tdarr Tuarr
Amount 9548 3744 6328 380 0 20000
Percentage 4774 1872 3164 190 0 100
The statistics for different optimization operations is performed and shown in Table
4 It can be seen that all the adjusted operations result in the reduction of gas
consumption In over half the cases (4774 with decreased temperature and 1872
with increased temperature) AG is adjusted to higher levels Meanwhile a number of
the adjustments (3164) lower the temperature by reducing the G and AG at the same
time This result indicates that the original temperature cannot be adjusted to the target
by only reducing G in the constraint conditions The remaining cases (19) are
24
343
344
345
346
347
348
349
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
relatively special in which an excess of air is supplied in the original operation and the
adjusted operation thereby increasing the temperature by a smaller AG with less gas
consumption
6 Conclusion
Based on the actual operation data this paper aims to explore the improvement of the
combustion efficiency and the room for energy conservation Correlation analysis and
LDA show that it is of great significance to introduce two derived ratio variables which
are the AG ratio and the TG ratio A type of combustion efficiency utilizing an on-line
soft sensor is put forward by employing a NNG variable selection algorithm which
provides a good solution to the combustion efficiency real-time measurement problem
of a reheating furnace The implementation of the model-based optimization is studied
based on the actual operational data Detailed analysis for the optimization results is
given for the different cases The results show that significant energy conservation can
be achieved when the furnace operation is optimized by using the developed soft sensor
model
Acknowledgement
The authors would like to thank the financial support provided by the National Nature
Science Foundation of China under Grant 61171145 Y Yao was supported by Ministry
of Science amp Technology ROC under Grant number MOST 104-2221-E-007-129
References
25
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[1] Z J Wang Q D Wu and T Y Chai Optimal-setting control for complicated
industrial processes and its application studyControl Engineering Practice vol
12 pp 65-74 2004
[2] A Steinboeck K Graichen and A Kugi Dynamic Optimization of a Slab
Reheating Furnace With Consistent Approximation of Control VariablesIEEE
Transactions on Control Systems Technology vol 19 pp 1444-1456 2011
[3] B T Zhang C Y Wang Q Qin and L Li Influence of Boiler Combustion
Adjustment on NOxEmission and Boiler EfficiencyAdvanced Materials
Research vol 732-733 pp 234-237 2013
[4] C K Yoo and IB Lee Soft Sensor and Adaptive Model-Based Dissolved
Oxygen Control for Biological Wastewater Treatment ProcessesEnvironmental
Engineering Science vol 21 pp 331-340 2004
[5] S A Bhat D N Saraf S Gupta and S K Gupta Use of Agitator Power as a
Soft Sensor for Bulk Free-Radical Polymerization of Methyl Methacrylate in
Batch ReactorsIndustrial amp Engineering Chemistry Research vol 45 pp 4243-
4255 2006
[6] K Desai Y Badhe S S Tambe and B D Kulkarni Soft-sensor development
for fed-batch bioreactors using support vector regressionBiochemical
Engineering Journal vol 27 pp 225-239 2006
[7] Y P Badhe Lonari J Tambe S S amp Kulkarni B D Improve polyethylene
process control and product qualityHydrocarbon Processing vol 86 pp 53-60
2007
26
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[8] N K Nath K Mandal A K Singh B Basu C Bhanu S Kumar et al Ladle
furnace on-line reckoner for prediction and control of steel temperature and
compositionIronmaking amp Steelmaking vol 33 pp 140-150 2006
[9] A J Yan T Y Chai F H Wu and P Wang Hybrid intelligent control of
combustion process for ore-roasting furnaceJournal of Control Theory and
Applications vol 6 pp 80-85 2008
[10] J Li W M Zhong H Cheng X D Kong and F Qian A data-driven soft
sensor modeling for furnace temperature of Opposed Multi-Burner gasifier pp
705-710 2011
[11] Y H Yang Y H Liu X Z Liu and S K Qin Billet temperature soft sensor
model of reheating furnace based on RVM method pp 4003-4006 2011
[12] J H Wang C Wang X F Zhu and X K Fang Application of soft sensor in
welding seam tracking prediction based on LSSVM and PSO with compression
factor pp 2441-2446 2013
[13] L Balbis J Balderud and M J Grimble Nonlinear predictive control of steel
slab reheating furnace pp 1679-1684 2008
[14] A Steinboeck D Wild T Kiefer and A Kugi A mathematical model of a slab
reheating furnace with radiative heat transfer and non-participating gaseous
media International Journal Of Heat And Mass Transfer vol 53 pp 5933-
5946 Dec 2010
[15] A Steinboeck D Wild and A Kugi Nonlinear model predictive control of a
continuous slab reheating furnace Control Engineering Practice vol 21 pp
495-508 2013
27
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[16] C Zhang T Ishii and S Sugiyama Numerical Modeling Of the Thermal
Performance Of Regenerative Slab Reheat Furnaces Numerical Heat Transfer
Part A Applications vol 32 pp 613-631 1997
[17] J G Kim and K Y Huh Prediction of Transient Slab Temperature Distribution
in the Re-heating Furnace of a Walking-beam Type for Rolling of Steel Slabs
ISIJ International vol 40 pp 1115-1123 2000
[18] J G Kim K Y and H I T K Three-Dimensional Analysis Of the Walking-
Beam-Type Slab Reheating Furnace In Hot Strip Mills Numerical Heat
Transfer Part A Applications vol 38 pp 589-609 2000
[19] CT Hsieh MJ Huang ST Lee and CH Wang Numerical Modeling of a
Walking-Beam-Type Slab Reheating Furnace Numerical Heat Transfer Part A
Applications vol 53 pp 966-981 2008
[20] MJ Huang CT Hsieh ST Lee and CH Wang A Coupled Numerical Study
of Slab Temperature and Gas Temperature in the Walking-Beam-Type Slab
Reheating Furnace Numerical Heat Transfer Part A Applications vol 54 pp
625-646 2008
[21] Z Song and A Kusiak Constraint-Based Control of Boiler Efficiency A Data-
Mining Approach IEEE Transactions on Industrial Informatics vol 3 pp 73-
83 2007
[22] J Q Li J J Gu and C L Niu The Operation Optimization based on
Correlation Analysis of Operation Parameters in Power Plant pp 138-141
2008
28
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[23] J G Wang S S Shieh S S Jang D S H Wong and C W Wu A two-tier
approach to the data-driven modeling on thermal efficiency of a BFGcoal co-
firing boiler Fuel vol 111 pp 528-534 Sep 2013
[24] L Breiman Better Subset Regression Using the Nonnegative Garrote
Technometrics vol 37 pp 373-384 1995
[25] R A Fisher The use of multiple measurements in taxonomic problems
AnnHum Genet vol 7 pp 179-188 1936
[26] J Liu Developing a soft sensor based on sparse partial least squares with
variable selection Journal of Process Control vol 24 pp 1046-1056 2014
[27] K Sun J Liu JL Kang SS Jang D SH Wong and DS Chen
Development of a variable selection method for soft sensor using artificial
neural network and nonnegative garrote Journal of Process Control vol 24 pp
1068-1075 2014
29
439
440
441
442
443
444
445
446
447
448
449
450
451
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
Correspondence information
J G Wang Tel +86-21-56331278 Email jgwangshueducn
Y Yao Tel +886-3-5713690 Email yyaomxnthuedutw
2
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
1 Introduction
The reheating furnace for a rolling mill is the most energy consumption equipment in
a large-scale iron and steel plant thus it is of great significance to improve the
combustion efficiency and reduce gas consumption [1 2] Since combustion efficiency
cannot be measured directly the adjustment of oxygen content in the exhaust gas is
often used to indirectly control the efficiency Another method is to estimate
combustion efficiency based on oxygen content in the exhaust gas and then implement
control actions [3] However the performance of these methods depends on the
precision and stability of oxygen analyzers which are susceptible to corrosion and wear
of high-temperature gases and difficult to maintain in full operational status for a long
period of time
When quality variables cannot be easily obtained a soft sensor model that can predict
these quality characteristics (as response variables) using readily available sensor
variables (as candidate predictors) will be most desirable A variety of soft sensor
methods and applications have been studied in different fields [4-7]
For a reheating furnace a number of soft sensors have been investigated The feature
of continuous prediction of temperature and composition of the combustion atmosphere
has the potential of acting as a soft sensor thereby leading to a reduced number of
temperature measurements and sampling for chemical analysis [8] The secure
economic and stable control of the combustion process is realized by the cooperation
work of a cascade fuzzy control system for furnace temperature a ratio control system
for air flow with a soft-sensing model plus a fault diagnosis model [9] A data-driven
soft sensor modeling technique for furnace temperature of the Opposition Multi-Burner
(OMB) gasifier is proposed and the selection of secondary variables and model structure
3
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
of a back propagation (BP) neural network is studied which indicates that the furnace
temperature predictive model integrating the principal component analysis (PCA) and
the BP neural network has a promising performance with good predictive precision
[10] A soft sensor modeling method is proposed to predict the billet temperature of the
reheating furnace based on a relevance vector machine (RVM) which has a higher
prediction accuracy and a certain practical significance to the on-site production of a
reheating furnace [11] The least square support vector machine (LSSVM) inductance
model optimized by the particle swarm optimization method with a compression factor
(PSO-CF) algorithm is presented for the difficulty of time prediction which can
improve PSO convergence accuracy and effectively avoid falling into a local optimum
[12] However the soft sensor developed for combustion efficiency was not
investigated in these research efforts which is significant for energy conservation
On the premise of the model prediction accuracy the model-based control makes
optimal operation feasible which can then be successfully employed to operate a
reheating furnace in an efficient way The potential of the nonlinear model predictive
control techniques is explored to improve the temperature control for the metal slabs in
a hot mill reheating furnace and particularly whether or not these control techniques
can be exploited to reduce energy consumption [13] Steinboeck et al developed a
mathematical model of the reheating process of steel slabs in industrial fuel-fired
furnaces in 2010 They exploited a dynamic optimization method for temperature
control of the steel slabs in a continuous reheating furnace and a temperature control
method for reheating steel slabs in an industrial furnace in 2011 They also designed a
nonlinear model predictive controller for a reheating furnace for steel slabs in 2013 [2
14 15]
4
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
Obviously the research on the numerical model for the heating performance of
reheating furnace can be done based on basic combustion theory and heat transmission
characteristics Many scholars devote themselves to the simulations of the heat flow
phenomenon in the reheating furnace Zhang et al attempted to apply a computational
fluid dynamic (CFD) simulation to predict the combustion performance for a reheating
furnace by simplifying the furnace to a cuboid and assuming that the slab possesses
infinite length and enters the reheating furnace at a fixed speed [16] The CFD method
has been applied to the study of reaction turbulence radiation heat transmission and the
calculation for the steady state heat transmission rate of the slab under the given
temperature [17] In the works of references [16-20] the authors used the given
temperature data of slabs to compute the steady flow and temperature field However
owing to the changing operating conditions the actual implementation of these
numerical model methods still bristle with difficulties although the methods mentioned
above are feasible for the prediction Thus for an online application it is necessary to
adopt a real time data-driven model to resolve the time varying characteristics
Proper variable selection is an important step in model building for a large-scale
combustion system A well-trimmed variable dimension ensures the acquired model is
transparent comprehensible and robust Some studies reported that the combustion
model built by a selected subset of input variables provide more accurate predictions of
combustion efficiency than the entire set of variables [21-23] Recently shrinkage
methods which conduct variable selection by shrinking or setting some coefficients of a
ldquogreedyrdquo model to zero have received significant attention A popular form of these
methods is the non-negative garrote (NNG) [23 24]
5
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
Against this background this paper aims to propose a combustion efficiency index
for the reheating furnace and investigate for room in improvement regarding energy
conservation The primary contribution is a practical combustion efficiency index the
incorporation of the derived variables and soft-sensing method for the optimization of
combustion efficiency of reheating furnaces The derived variables are found more
physically meaningful than the plain variables when constructing the model of
combustion efficiency By employing a NNG variable selection procedure an adaptive
scheme for combustion efficiency modeling and adjustment is proposed and virtually
implemented for a rolling reheating furnace The results show that there is significant
room for energy conservation
The remainder of the paper is organized as follows In the next section the reheating
furnace and the data preprocessing is described In Section 3 the statistics analysis for
different variables and the formation of derived variables are presented In Section 4
the framework of an adaptive model based on NNG variable selection is presented and
two models developed for the temperature and temperature-gas (TG) ratio are
compared according to the model prediction precision A model-based optimization
scheme is provided and applied to the combustion efficiency improvement for an actual
case of a reheating furnace presented in Section 5 Several remarks and a summary
conclude the last section
2 Plant description and data preprocessing
The schematic of the heating process in the rolling mill reheating furnace is shown in
Fig 1 There are four zones in the reheating furnace including the preheating zone (P)
the first heating zone (1) the second heating zone (2) and the soaking zone (S) The
6
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
steel slab moves through the four zones in turn and is heated to the demanded state
using a specific temperature increase curve As is shown in Table 1 the soaking zone
has two areas that are defined as up and down and both of the areas possess the same
five variables including two manipulated variables the air flow (A) and the gas flow
(G) and three temperatures in left center and right sections of the area (T-l T-c and T-
r) The other zones have the same variables as the soaking zone hence there are 40
variables in total for the reheating furnace
Fig 1 The schematic of the heating process in the reheating furnace
Table 1 Variables and descriptions in the soaking zone
Variable Description unit
AS-u Air flow in the lsquouprsquo area Nm3h
AS-d Air flow in the lsquodownrsquo area Nm3h
GS-u Gas flow in the lsquouprsquo area Nm3h
GS-d Gas flow in the lsquodownrsquo area Nm3h
TS-ul Temperature in the left part of the lsquouprsquo area
TS-uc Temperature in the center part of the lsquouprsquo area
TS-ur Temperature in the right part of the lsquouprsquo area
7
143
144
145
146
147
148
149
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
TS-dl Temperature in the left part of the lsquodownrsquo area
TS-dc Temperature in the center part of the lsquodownrsquo area
TS-dr Temperature in the right part of the lsquodownrsquo area
A data set of 20000 samples was used in this study The samples were collected from
an actual reheating furnace in a large iron and steel plant located in Shanghai from
September 14 to September 27 2014 The operational data is taken on a per minute
basis
1
092
093
042
041
046
092
1
091
04
039
045
093
091
1
038
038
045
042
04
038
1
096
096
041
039
038
096
1
096
046
045
045
096
096
1
TS-ul
TS-uc
TS-ur
TS-dl
TS-dc
TS-dr
TS-ul TS-uc TS-ur TS-dl TS-dc TS-dr
04
05
06
07
08
09
1
Fig 2 Correlation between each temperature in the soaking zone
In order to investigate the relation among different temperatures in each zone
correlation analysis is conducted for the soaking zone as illustrated in Fig 2 It can be
seen that the temperature in three parts of the lsquouprsquo area is highly correlated with a
correlation coefficient greater than 09 The similar results exist for the lsquodownrsquo area as
well On the contrary the correlation coefficient between temperatures in one part of
8
150
151
152
153
154
155
156
157
158
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
lsquouprsquo area and any part of lsquodownrsquo area does not exceed 05 Therefore for the reduction
of the data dimension the three temperatures in the lsquouprsquo area or the lsquodownrsquo area can be
treated as only one variable which can be taken as the mean value or the first principal
component acquired from the PCA analysis Considering the reservation of the variable
physical meaning the former is preferred and used
3 Statistics analysis and incorporation of derived variables
In this section in order to uncover the physical knowledge for the actual operation
guidance and confirmation statistical analysis is performed for the 16 input manipulated
variables and the eight output variables (ie the temperatures in the four zones) of a
reheating furnace system For the combustion efficiency evaluation and modeling two
types of derived ratio variables are introduced which is helpful to reveal the
information included in the data
31 Correlation analysis
Correlation analysis between the temperatures (T) in each area and all of the air flows
(A) and the gas flows (G) is performed and shown in Fig 3 It can be seen that only the
temperatures in both areas of the soaking zone are the most highly related to the air flow
and the gas flow in its own zone However this phenomenon does not occur in the other
three zones The temperatures in the second heating zone mainly depend on the air flow
and the gas flow in its own zone as well as the nearby first heating zone As for the
preheating zone and first heating zone no apparent correlation can be observed
Obviously these analysis results could not tell us the explicit information about how to
evaluate the efficiency levels and key manipulated variables
9
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
-101
T P-u
AP GP A1 G1 A2 G2 AS GS
AP GP A1 G1 A2 G2 AS GS
AP GP A1 G1 A2 G2 AS GS
-101
T P-d
-101
T 1-u
-101
T 1-d
AP GP A1 G1 A2 G2 AS GS
-101
T 2-u
AP GP A1 G1 A2 G2 AS GS
-101
T 2-d
-101
T S-u
-101
T S-d
AP GP A1 G1 A2 G2 AS GS
AP GP A1 G1 A2 G2 AS GS
AP GP A1 G1 A2 G2 AS GS
UPDOWN
Fig 3 Correlation between air flows gas flows and temperatures
32 Incorporation of derived variables
During the stable heating stage the quantity of heat absorbed and removed from the
slabs from each furnace zone is relatively constant Hence the derived variable TG
ratio can be treated as an index for the combustion efficiency level This is because a
higher TG ratio signifies more combustion heat generated from unit gas ie higher
combustion efficiency
Moreover it is known that the appropriate air and fuel ratio is vital for the
combustion efficiency so the air-gas ratio (AG) is utilized as another derived variable
for the research Again the correlation analysis is performed for two types of derived
variables The correlations between different variables including the AG ratio G and
the TG ratio are shown in Fig 4 It can be clearly seen from the four red rectangle
blocks that TG of each zone is only remarkably related to AG and G in the same zone
10
183
184
185
186
187
188
189
190
191
192
193
194
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
Furthermore as related to the lsquouprsquo or lsquodownrsquo areas in one zone TG in each area has the
highest correlation with AG and G in the same area while AG and G in the opposite
area of the same zone is secondary This can be easily seen from the red and blue color
markings in each red rectangle block Thus it is of great significance to introduce these
derived ratio variables
-101
TG
P-u
AGP GP AG1G1 AG2 G2 AGS GS
-101
TG
P-d
AGP GP AG1 G1 AG2 G2 AGS GS
-101
TG
1-u
AGP GP AG1 G1 AG2 G2 AGS GS
-101
TG
1-d
AGP GP AG1 G1 AG2 G2 AGS GS
-101
TG
2-u
AGP GP AG1 G1 AG2 G2 AGS GS
-101
TG
2-d
AGP GP AG1 G1 AG2 G2 AGS GS
-101
TG
S-u
AGP GP AG1 G1 AG2 G2 AGS GS
-101
TG
S-d
AGP GP AG1 G1 AG2 G2 AGS GS
UPDOWN
Fig 4 Correlation between the air-gas ratio gas flow and TG ratio
33 LDA analysis
Linear discriminant analysis (LDA) aims to finding a projection direction that
maximizes the separation of class means and minimizes the within-class variance [25]
In this section LDA is utilized to identify the discriminating variables that play an
important role in determining combustion efficiency levels All the data are partitioned
into five groups according to their efficiency levels
11
195
196
197
198
199
200
201
202
203
204
205
206
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
Based on a descending order of TG ratios five efficiency levels for the lsquouprsquo area of
the soaking zone are denoted as HH H M L and LL LDA is conducted on the three
groups of the data with levels of HH M and LL
Fig 5 shows the scattering of the LDA projections of the process observations
collected at the three efficiency levels where y1 and y2 correspond to the first two LDA
components that contain most discriminant information The weighting factors ( and
) of the 16 input variables composing the projections y1 and y2 are shown in Fig 6
where and From left to right the 16 input variables are defined as
the eight AG ratio variables and the eight gas flow variables with the order of variables
of each kind P-u P-d 1-u 1-d 2-u 2-d S-u and S-d
-8 -6 -4 -2 0 2 4 6-4
-3
-2
-1
0
1
2
3
4
5
6
Projection y1
Pro
ject
ion
y2
HHMLL
Fig 5 Scattering of LDA projections y1 and y2 for three efficiency levels
12
207
208
209
210
211
212
213
214
215
216
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
AG_S-u-4
-2
0
2
Wei
ghtin
g fa
ctor
for y
1
AG_S-u
-101234
Wei
ghtin
g fa
ctor
for y
2
Fig 6 Weight factors of various input variables for projections y1 and y2
The two figures reveal that the three groups are clearly discriminated by the LDA
projection and the most significant variables for the different efficiency levels are the
derived variables ie the AG in the lsquouprsquo area of the soaking zone A similar result can
be obtained in the lsquodownrsquo area or any area of the other zones Therefore AG in each
heating area is the key manipulated variable that determines the different combustion
efficiency
4 Modeling and prediction of temperature and temperature-gas ratio
For the model-based operation optimization the models for the temperature and TG
ratio based on the NNG algorithm will be developed and compared
41 NNG variable selection algorithm
The NNG method can be generalized into a two-stage shrinkage method In the first
stage the sign for each variable is determined using the ordinary least square procedure
13
217
218
219
220
221
222
223
224
225
226
227
228
229
230
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
and in the second stage the corresponding magnitudes are computed by solving a series
of constrained quadratic programming
A set of observation data is provided where is the input matrix whose
columns represent the measured candidate variables and is the corresponding
vector of the response data The following expression is given with the number of the
response variable being equal to 1 but a similar procedure can be generalized to any
number of variables Let and be normalized to the zero-mean and unit standard
deviation Additionally let be a set of the ordinary least square estimates of the
coefficients of the following linear model then
(1)
The second stage shrinkage can be formulated as the following optimization problem
subject to
(2)
As decreased and the NNG is tightened more of the become zero and the
remaining nonzero coefficients are shrunk A solution path exists with on which
the appropriate shrinkage can be selected Conventionally the v-fold cross-validation is
used to estimate the prediction error and to select the best solution in the solution path
so as to minimize the prediction or model error
42 Modeling and prediction of temperature and TG ratio
The input-output relations change with time as the reheating process develops An
adaptive modeling strategy is often used to resolve time-varying characteristics of
14
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
industrial processes In this paper the NNG-based regression modeling prediction and
optimization is implemented in a moving window manner where the size of the
window and the length of the moving step are selected as 1000 and 100 respectively In
addition the prediction horizon is also set as 100 In each step the NNG regression
model is built based on the data in the current window which is then used for the
prediction and optimization in the subsequent prediction horizon Next the window
moves forward by replacing the oldest 100 samples for model training by the
observations collected in the previous prediction horizon This moving window strategy
is workable because in the reheating process the input-output relation is slowly time-
varying and the model is still valid for the prediction and optimization in the subsequent
short time
The modeling of the temperature T is based on the air flow rates and gas flow rates
while the modeling for the TG ratio is based on the AG ratios and gas flow rates
Taking the lsquouprsquo area of the soaking zone as an example the prediction result for T and
the TG ratio in the lsquouprsquo area of the soaking zone is shown in Figs 7 and 8 respectively
0 02 04 06 08 1 12 14 16 18 2
x 104
1140
1160
1180
1200
1220
1240
1260
1280
1300
1320
Sampling intervals
T S-u (
)
OriginalPredicted
Fig 7 Prediction of temperature in the lsquouprsquo area of the soaking zone
15
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
0 02 04 06 08 1 12 14 16 18 2
x 104
02
04
06
08
1
12
14
16
18
2
Sampling intervals
TG
S-u
OriginalPredicted
Fig 8 Prediction of TG ratio in the lsquouprsquo area of the soaking zone
A comparison between Figs 7 and 8 show that the prediction of the TG ratio is more
accurate than the prediction of temperature The average R2 in 190 NNG regressions is
0935 for the TG ratio while the average R2 is only 0814 for temperature This is
consistent with the statistical analysis which reveals that the correlation between AG
and TG is higher than that between A and T These results indicate that derived
variables are more meaningful for the purpose of prediction and the modeling of the
TG ratio is more appropriate for implementing optimization
The selected frequency for each variable in 190 NNG regressions and the coefficients
of each variable in 20 NNG regressions for the TG ratio modeling of the lsquouprsquo area of
the soaking zone are shown in Figs 9 and 10 respectively Fig 9 shows that the selected
frequency of the variables in the lsquouprsquo area of the soaking zone is much higher than
variables in other zones Similarly as is shown in Fig 10 the NNG regression
coefficients of the variables in the lsquouprsquo area of the soaking zone are much larger than the
variables in the other areas These results indicated that variables in the lsquouprsquo area of
16
269
270
271
272
273
274
275
276
277
278
279
280
281
282
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
soaking zone are mostly contributed to the modeling of the TG ratio in same area
which is quite reasonable Similar results can be obtained for the other areas
0
20
40
60
80
100
120
140
160
180
200
Sel
ecte
d fre
quen
cy
AGP-u AGP-d AG1-u AG1-d AG2-u AG2-d AGS-u AGS-d GP-u GP-d G1-u G1-d G2-u G2-d GS-u GS-d
Fig 9 Variable selected frequency of over 190 runs for the TG ratio model of the lsquouprsquo
area of the soaking zone
0 2 4 6 8 10 12 14 16 18 20-05
05
AG
P-u
0 2 4 6 8 10 12 14 16 18 20-05
05
GP
-u
0 2 4 6 8 10 12 14 16 18 20-1
0
AG
P-d
0 2 4 6 8 10 12 14 16 18 20-1
0
GP
-d
0 2 4 6 8 10 12 14 16 18 20-05
05
AG
1-u
0 2 4 6 8 10 12 14 16 18 20-05
05
G1-
u
0 2 4 6 8 10 12 14 16 18 20-05
05
AG
1-d
0 2 4 6 8 10 12 14 16 18 20-05
05
G1-
d
0 2 4 6 8 10 12 14 16 18 20-1
0
AG
2-u
0 2 4 6 8 10 12 14 16 18 200
1
G2-
u
0 2 4 6 8 10 12 14 16 18 20-1
0
AG
2-d
0 2 4 6 8 10 12 14 16 18 20-1
0
G2-
d
0 2 4 6 8 10 12 14 16 18 200
1
AG
S-u
0 2 4 6 8 10 12 14 16 18 20-1
0
GS
-u
0 2 4 6 8 10 12 14 16 18 20-05
05
AG
S-d
0 2 4 6 8 10 12 14 16 18 20-05
05
GS
-d
Fig 10 Part of the NNG regression coefficients of each variable over 190 runs for the
TG ratio model of the lsquouprsquo area of the soaking zone
It is remarkable that the NNG regression coefficients corresponding to the AG ratio
in the lsquouprsquo area of the soaking zone are consistently positive On the contrary the
coefficients of the gas flow rate in the same area are consistently negative This
indicates that under this condition the rise of the AG leads to the increase of the TG
17
283
284
285
286
287
288
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
while the rise of the G leads to the drop of the TG This property is helpful for
performing optimization and improve combustion efficiency
For comparison the modeling results for two other algorithms artificial neural
network (ANN) and partial least squares (PLS) [26] are compared with the NNG
algorithm in the following three aspects [27]
(1) Model Size the number of variables selected for modeling
(2) Model Magnitude the mean of the L1 norm of the regression coefficients
(3) Prediction Precision the mean squared prediction error (MSPE)
Summary of the algorithm comparison is shown in Table 2 It can be seen that the
superiorities of the NNG regression in model size model magnitude and model
precision are remarkable
Table 2 Summary of algorithm comparison
Index NNG ANN PLS
Model Size 849 16 16
Model Magnitude 10075 -- 12150
MSPE 00093 00108 00140
5 Model-based optimization
51 Implemention of model-based optimization operation
The goal of optimization is to seek an optimal combination of AG and G in the lsquouprsquo
area of the soaking zone so as to minimize the gas consumption while keeping the
temperature at the target value According to the heating schedule the most expected
temperature in the soaking zone is 1200 Therefore the target temperature
is set at this value to achieve the desired heating effect As shown in Fig 11 in the real
18
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
operation the temperature fluctuates around the target value because of the imperfect
control performance However in order to simplify the analysis and compute the
maximum possible energy saving perfect temperature control is assumed when
adopting the model-based optimization strategy In other words it is assumed that the
actual temperature in the lsquouprsquo area of the soaking zone can be adjusted to the expected
temperature ie 1200
0 02 04 06 08 1 12 14 16 18 2
x 104
1150
1200
1250
1300
Sampling intervals
T S-u (
)
OriginalTarget
Fig 11 Original and target temperature in the lsquouprsquo area of the soaking zone
The adjustment scheme takes the maximum value and minimum value of the original
operation data as the upper and lower bounds for the adjustment Moreover in order to
assure the validity of the linear model the increment or decrement of AG and G should
not beyond of the original value (considered as 10 for the purposes of this paper)
The adjustment strategy can be formulated as
19
307
308
309
310
311
312
313
314
315
316
317
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
(3)
where NNG() denotes the NNG regression model of the TG ratio and are the
NNG regression coefficients of the AG and G in the lsquouprsquo area of the soaking zone for
modeling TG in same area
With the model-based strategy and adaptive model of the TG ratio given in Section
42 the optimization adjustment results can be obtained as follows
The adjustment of gas flow in the lsquouprsquo area of the soaking zone shown in Fig 12
illustrates that the implementation of the model-based optimization operation can reduce
the consumption of the gas flow Compared with the original case 938 of the gas can
be saved on average by utilizing the model-based adjustment
0 02 04 06 08 1 12 14 16 18 2
x 104
-800
-700
-600
-500
-400
-300
-200
-100
0
Sampling intervals
Adj
ustm
ent o
f gas
flow
m(
3 h)
20
318
319
320
321
322
323
324
325
326
327
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
(a) Adjustment amount of gas flow
0 02 04 06 08 1 12 14 16 18 2
x 104
500
1000
1500
2000
2500
3000
3500
4000
4500
5000G
as fl
owm
(3 h
)
Sampling intervals
OriginalAdjusted
(b) Comparison of original and adjusted gas flow
095 096 097 098 099 1 101 102 103 104 105
x 104
500
1000
1500
2000
2500
3000
3500
4000
4500
5000
Gas
flow
m(
3 h)
Sampling intervals
OriginalAdjusted
(c) Comparison of original and adjusted gas flow in interval [9501-10500]
Fig 12 Adjustment of gas flow in the lsquouprsquo area of the soaking zone
The adjustment of the air flow in the lsquouprsquo area of the soaking zone given in Fig 13
shows that the air flow is reduced in most cases and only increased for a few cases In
21
328
329
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
general 681 of the air flow is reduced after the implementation of the model-based
optimization
0 02 04 06 08 1 12 14 16 18 2
x 104
-2000
-1500
-1000
-500
0
500
1000
Sampling intervals
Adj
ustm
ent o
f air
flow
(m3 h
)
(a) Adjustment amount of air flow
0 02 04 06 08 1 12 14 16 18 2
x 104
4000
5000
6000
7000
8000
9000
10000
Air
flow
m(
3 h)
Sampling intervals
OriginalAdjusted
(b) Comparison of original and adjusted air flow
22
330
331
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
095 096 097 098 099 1 101 102 103 104 105
x 104
4000
5000
6000
7000
8000
9000
10000
Air
flow
m(
3 h)
Sampling intervals
OriginalAdjusted
(c) Comparison of original and adjusted air flow in interval [9501-10500]
Fig 13 Adjustment of air flow in the lsquouprsquo area of the soaking zone
52 Discussions
In this section detailed analysis for the optimization operation results is provided
The statistics of optimization at the boundary conditions shown in Table 3 illustrates
that the adjusted value for G reaches its lower limit in most cases (8045 for
and 687 for ) which is the most energy-efficient point In a number of cases
(1079 for and 001 for ) the adjusted AG reaches its upper
limit These results indicate that the optimization operations maximize the combustion
efficiency by decreasing G and increasing the AG ratio up to the boundary conditions
However the lower limit is also unexpectedly attained in a small number of cases
(110 for and 078 for ) for which a more detailed analysis is
given below
23
332
333
334
335
336
337
338
339
340
341
342
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
Table 3 Statistics of optimization at boundary conditions
Total
Amount 0 1374 0 16090 17464
Percentag
e 0 687 0 8045
8732
Total
Amount 2 220 2158 156 2536
Percentag
e 001 110 1079 078
1268
Table 4 Statistics for different optimization operations
Adjustment
Gdarr
Guarr TotalAGuarr AGdarr
Tdarr Tuarr Tdarr Tuarr
Amount 9548 3744 6328 380 0 20000
Percentage 4774 1872 3164 190 0 100
The statistics for different optimization operations is performed and shown in Table
4 It can be seen that all the adjusted operations result in the reduction of gas
consumption In over half the cases (4774 with decreased temperature and 1872
with increased temperature) AG is adjusted to higher levels Meanwhile a number of
the adjustments (3164) lower the temperature by reducing the G and AG at the same
time This result indicates that the original temperature cannot be adjusted to the target
by only reducing G in the constraint conditions The remaining cases (19) are
24
343
344
345
346
347
348
349
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
relatively special in which an excess of air is supplied in the original operation and the
adjusted operation thereby increasing the temperature by a smaller AG with less gas
consumption
6 Conclusion
Based on the actual operation data this paper aims to explore the improvement of the
combustion efficiency and the room for energy conservation Correlation analysis and
LDA show that it is of great significance to introduce two derived ratio variables which
are the AG ratio and the TG ratio A type of combustion efficiency utilizing an on-line
soft sensor is put forward by employing a NNG variable selection algorithm which
provides a good solution to the combustion efficiency real-time measurement problem
of a reheating furnace The implementation of the model-based optimization is studied
based on the actual operational data Detailed analysis for the optimization results is
given for the different cases The results show that significant energy conservation can
be achieved when the furnace operation is optimized by using the developed soft sensor
model
Acknowledgement
The authors would like to thank the financial support provided by the National Nature
Science Foundation of China under Grant 61171145 Y Yao was supported by Ministry
of Science amp Technology ROC under Grant number MOST 104-2221-E-007-129
References
25
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[1] Z J Wang Q D Wu and T Y Chai Optimal-setting control for complicated
industrial processes and its application studyControl Engineering Practice vol
12 pp 65-74 2004
[2] A Steinboeck K Graichen and A Kugi Dynamic Optimization of a Slab
Reheating Furnace With Consistent Approximation of Control VariablesIEEE
Transactions on Control Systems Technology vol 19 pp 1444-1456 2011
[3] B T Zhang C Y Wang Q Qin and L Li Influence of Boiler Combustion
Adjustment on NOxEmission and Boiler EfficiencyAdvanced Materials
Research vol 732-733 pp 234-237 2013
[4] C K Yoo and IB Lee Soft Sensor and Adaptive Model-Based Dissolved
Oxygen Control for Biological Wastewater Treatment ProcessesEnvironmental
Engineering Science vol 21 pp 331-340 2004
[5] S A Bhat D N Saraf S Gupta and S K Gupta Use of Agitator Power as a
Soft Sensor for Bulk Free-Radical Polymerization of Methyl Methacrylate in
Batch ReactorsIndustrial amp Engineering Chemistry Research vol 45 pp 4243-
4255 2006
[6] K Desai Y Badhe S S Tambe and B D Kulkarni Soft-sensor development
for fed-batch bioreactors using support vector regressionBiochemical
Engineering Journal vol 27 pp 225-239 2006
[7] Y P Badhe Lonari J Tambe S S amp Kulkarni B D Improve polyethylene
process control and product qualityHydrocarbon Processing vol 86 pp 53-60
2007
26
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[8] N K Nath K Mandal A K Singh B Basu C Bhanu S Kumar et al Ladle
furnace on-line reckoner for prediction and control of steel temperature and
compositionIronmaking amp Steelmaking vol 33 pp 140-150 2006
[9] A J Yan T Y Chai F H Wu and P Wang Hybrid intelligent control of
combustion process for ore-roasting furnaceJournal of Control Theory and
Applications vol 6 pp 80-85 2008
[10] J Li W M Zhong H Cheng X D Kong and F Qian A data-driven soft
sensor modeling for furnace temperature of Opposed Multi-Burner gasifier pp
705-710 2011
[11] Y H Yang Y H Liu X Z Liu and S K Qin Billet temperature soft sensor
model of reheating furnace based on RVM method pp 4003-4006 2011
[12] J H Wang C Wang X F Zhu and X K Fang Application of soft sensor in
welding seam tracking prediction based on LSSVM and PSO with compression
factor pp 2441-2446 2013
[13] L Balbis J Balderud and M J Grimble Nonlinear predictive control of steel
slab reheating furnace pp 1679-1684 2008
[14] A Steinboeck D Wild T Kiefer and A Kugi A mathematical model of a slab
reheating furnace with radiative heat transfer and non-participating gaseous
media International Journal Of Heat And Mass Transfer vol 53 pp 5933-
5946 Dec 2010
[15] A Steinboeck D Wild and A Kugi Nonlinear model predictive control of a
continuous slab reheating furnace Control Engineering Practice vol 21 pp
495-508 2013
27
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[16] C Zhang T Ishii and S Sugiyama Numerical Modeling Of the Thermal
Performance Of Regenerative Slab Reheat Furnaces Numerical Heat Transfer
Part A Applications vol 32 pp 613-631 1997
[17] J G Kim and K Y Huh Prediction of Transient Slab Temperature Distribution
in the Re-heating Furnace of a Walking-beam Type for Rolling of Steel Slabs
ISIJ International vol 40 pp 1115-1123 2000
[18] J G Kim K Y and H I T K Three-Dimensional Analysis Of the Walking-
Beam-Type Slab Reheating Furnace In Hot Strip Mills Numerical Heat
Transfer Part A Applications vol 38 pp 589-609 2000
[19] CT Hsieh MJ Huang ST Lee and CH Wang Numerical Modeling of a
Walking-Beam-Type Slab Reheating Furnace Numerical Heat Transfer Part A
Applications vol 53 pp 966-981 2008
[20] MJ Huang CT Hsieh ST Lee and CH Wang A Coupled Numerical Study
of Slab Temperature and Gas Temperature in the Walking-Beam-Type Slab
Reheating Furnace Numerical Heat Transfer Part A Applications vol 54 pp
625-646 2008
[21] Z Song and A Kusiak Constraint-Based Control of Boiler Efficiency A Data-
Mining Approach IEEE Transactions on Industrial Informatics vol 3 pp 73-
83 2007
[22] J Q Li J J Gu and C L Niu The Operation Optimization based on
Correlation Analysis of Operation Parameters in Power Plant pp 138-141
2008
28
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[23] J G Wang S S Shieh S S Jang D S H Wong and C W Wu A two-tier
approach to the data-driven modeling on thermal efficiency of a BFGcoal co-
firing boiler Fuel vol 111 pp 528-534 Sep 2013
[24] L Breiman Better Subset Regression Using the Nonnegative Garrote
Technometrics vol 37 pp 373-384 1995
[25] R A Fisher The use of multiple measurements in taxonomic problems
AnnHum Genet vol 7 pp 179-188 1936
[26] J Liu Developing a soft sensor based on sparse partial least squares with
variable selection Journal of Process Control vol 24 pp 1046-1056 2014
[27] K Sun J Liu JL Kang SS Jang D SH Wong and DS Chen
Development of a variable selection method for soft sensor using artificial
neural network and nonnegative garrote Journal of Process Control vol 24 pp
1068-1075 2014
29
439
440
441
442
443
444
445
446
447
448
449
450
451
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
1 Introduction
The reheating furnace for a rolling mill is the most energy consumption equipment in
a large-scale iron and steel plant thus it is of great significance to improve the
combustion efficiency and reduce gas consumption [1 2] Since combustion efficiency
cannot be measured directly the adjustment of oxygen content in the exhaust gas is
often used to indirectly control the efficiency Another method is to estimate
combustion efficiency based on oxygen content in the exhaust gas and then implement
control actions [3] However the performance of these methods depends on the
precision and stability of oxygen analyzers which are susceptible to corrosion and wear
of high-temperature gases and difficult to maintain in full operational status for a long
period of time
When quality variables cannot be easily obtained a soft sensor model that can predict
these quality characteristics (as response variables) using readily available sensor
variables (as candidate predictors) will be most desirable A variety of soft sensor
methods and applications have been studied in different fields [4-7]
For a reheating furnace a number of soft sensors have been investigated The feature
of continuous prediction of temperature and composition of the combustion atmosphere
has the potential of acting as a soft sensor thereby leading to a reduced number of
temperature measurements and sampling for chemical analysis [8] The secure
economic and stable control of the combustion process is realized by the cooperation
work of a cascade fuzzy control system for furnace temperature a ratio control system
for air flow with a soft-sensing model plus a fault diagnosis model [9] A data-driven
soft sensor modeling technique for furnace temperature of the Opposition Multi-Burner
(OMB) gasifier is proposed and the selection of secondary variables and model structure
3
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
of a back propagation (BP) neural network is studied which indicates that the furnace
temperature predictive model integrating the principal component analysis (PCA) and
the BP neural network has a promising performance with good predictive precision
[10] A soft sensor modeling method is proposed to predict the billet temperature of the
reheating furnace based on a relevance vector machine (RVM) which has a higher
prediction accuracy and a certain practical significance to the on-site production of a
reheating furnace [11] The least square support vector machine (LSSVM) inductance
model optimized by the particle swarm optimization method with a compression factor
(PSO-CF) algorithm is presented for the difficulty of time prediction which can
improve PSO convergence accuracy and effectively avoid falling into a local optimum
[12] However the soft sensor developed for combustion efficiency was not
investigated in these research efforts which is significant for energy conservation
On the premise of the model prediction accuracy the model-based control makes
optimal operation feasible which can then be successfully employed to operate a
reheating furnace in an efficient way The potential of the nonlinear model predictive
control techniques is explored to improve the temperature control for the metal slabs in
a hot mill reheating furnace and particularly whether or not these control techniques
can be exploited to reduce energy consumption [13] Steinboeck et al developed a
mathematical model of the reheating process of steel slabs in industrial fuel-fired
furnaces in 2010 They exploited a dynamic optimization method for temperature
control of the steel slabs in a continuous reheating furnace and a temperature control
method for reheating steel slabs in an industrial furnace in 2011 They also designed a
nonlinear model predictive controller for a reheating furnace for steel slabs in 2013 [2
14 15]
4
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
Obviously the research on the numerical model for the heating performance of
reheating furnace can be done based on basic combustion theory and heat transmission
characteristics Many scholars devote themselves to the simulations of the heat flow
phenomenon in the reheating furnace Zhang et al attempted to apply a computational
fluid dynamic (CFD) simulation to predict the combustion performance for a reheating
furnace by simplifying the furnace to a cuboid and assuming that the slab possesses
infinite length and enters the reheating furnace at a fixed speed [16] The CFD method
has been applied to the study of reaction turbulence radiation heat transmission and the
calculation for the steady state heat transmission rate of the slab under the given
temperature [17] In the works of references [16-20] the authors used the given
temperature data of slabs to compute the steady flow and temperature field However
owing to the changing operating conditions the actual implementation of these
numerical model methods still bristle with difficulties although the methods mentioned
above are feasible for the prediction Thus for an online application it is necessary to
adopt a real time data-driven model to resolve the time varying characteristics
Proper variable selection is an important step in model building for a large-scale
combustion system A well-trimmed variable dimension ensures the acquired model is
transparent comprehensible and robust Some studies reported that the combustion
model built by a selected subset of input variables provide more accurate predictions of
combustion efficiency than the entire set of variables [21-23] Recently shrinkage
methods which conduct variable selection by shrinking or setting some coefficients of a
ldquogreedyrdquo model to zero have received significant attention A popular form of these
methods is the non-negative garrote (NNG) [23 24]
5
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
Against this background this paper aims to propose a combustion efficiency index
for the reheating furnace and investigate for room in improvement regarding energy
conservation The primary contribution is a practical combustion efficiency index the
incorporation of the derived variables and soft-sensing method for the optimization of
combustion efficiency of reheating furnaces The derived variables are found more
physically meaningful than the plain variables when constructing the model of
combustion efficiency By employing a NNG variable selection procedure an adaptive
scheme for combustion efficiency modeling and adjustment is proposed and virtually
implemented for a rolling reheating furnace The results show that there is significant
room for energy conservation
The remainder of the paper is organized as follows In the next section the reheating
furnace and the data preprocessing is described In Section 3 the statistics analysis for
different variables and the formation of derived variables are presented In Section 4
the framework of an adaptive model based on NNG variable selection is presented and
two models developed for the temperature and temperature-gas (TG) ratio are
compared according to the model prediction precision A model-based optimization
scheme is provided and applied to the combustion efficiency improvement for an actual
case of a reheating furnace presented in Section 5 Several remarks and a summary
conclude the last section
2 Plant description and data preprocessing
The schematic of the heating process in the rolling mill reheating furnace is shown in
Fig 1 There are four zones in the reheating furnace including the preheating zone (P)
the first heating zone (1) the second heating zone (2) and the soaking zone (S) The
6
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
steel slab moves through the four zones in turn and is heated to the demanded state
using a specific temperature increase curve As is shown in Table 1 the soaking zone
has two areas that are defined as up and down and both of the areas possess the same
five variables including two manipulated variables the air flow (A) and the gas flow
(G) and three temperatures in left center and right sections of the area (T-l T-c and T-
r) The other zones have the same variables as the soaking zone hence there are 40
variables in total for the reheating furnace
Fig 1 The schematic of the heating process in the reheating furnace
Table 1 Variables and descriptions in the soaking zone
Variable Description unit
AS-u Air flow in the lsquouprsquo area Nm3h
AS-d Air flow in the lsquodownrsquo area Nm3h
GS-u Gas flow in the lsquouprsquo area Nm3h
GS-d Gas flow in the lsquodownrsquo area Nm3h
TS-ul Temperature in the left part of the lsquouprsquo area
TS-uc Temperature in the center part of the lsquouprsquo area
TS-ur Temperature in the right part of the lsquouprsquo area
7
143
144
145
146
147
148
149
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
TS-dl Temperature in the left part of the lsquodownrsquo area
TS-dc Temperature in the center part of the lsquodownrsquo area
TS-dr Temperature in the right part of the lsquodownrsquo area
A data set of 20000 samples was used in this study The samples were collected from
an actual reheating furnace in a large iron and steel plant located in Shanghai from
September 14 to September 27 2014 The operational data is taken on a per minute
basis
1
092
093
042
041
046
092
1
091
04
039
045
093
091
1
038
038
045
042
04
038
1
096
096
041
039
038
096
1
096
046
045
045
096
096
1
TS-ul
TS-uc
TS-ur
TS-dl
TS-dc
TS-dr
TS-ul TS-uc TS-ur TS-dl TS-dc TS-dr
04
05
06
07
08
09
1
Fig 2 Correlation between each temperature in the soaking zone
In order to investigate the relation among different temperatures in each zone
correlation analysis is conducted for the soaking zone as illustrated in Fig 2 It can be
seen that the temperature in three parts of the lsquouprsquo area is highly correlated with a
correlation coefficient greater than 09 The similar results exist for the lsquodownrsquo area as
well On the contrary the correlation coefficient between temperatures in one part of
8
150
151
152
153
154
155
156
157
158
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
lsquouprsquo area and any part of lsquodownrsquo area does not exceed 05 Therefore for the reduction
of the data dimension the three temperatures in the lsquouprsquo area or the lsquodownrsquo area can be
treated as only one variable which can be taken as the mean value or the first principal
component acquired from the PCA analysis Considering the reservation of the variable
physical meaning the former is preferred and used
3 Statistics analysis and incorporation of derived variables
In this section in order to uncover the physical knowledge for the actual operation
guidance and confirmation statistical analysis is performed for the 16 input manipulated
variables and the eight output variables (ie the temperatures in the four zones) of a
reheating furnace system For the combustion efficiency evaluation and modeling two
types of derived ratio variables are introduced which is helpful to reveal the
information included in the data
31 Correlation analysis
Correlation analysis between the temperatures (T) in each area and all of the air flows
(A) and the gas flows (G) is performed and shown in Fig 3 It can be seen that only the
temperatures in both areas of the soaking zone are the most highly related to the air flow
and the gas flow in its own zone However this phenomenon does not occur in the other
three zones The temperatures in the second heating zone mainly depend on the air flow
and the gas flow in its own zone as well as the nearby first heating zone As for the
preheating zone and first heating zone no apparent correlation can be observed
Obviously these analysis results could not tell us the explicit information about how to
evaluate the efficiency levels and key manipulated variables
9
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
-101
T P-u
AP GP A1 G1 A2 G2 AS GS
AP GP A1 G1 A2 G2 AS GS
AP GP A1 G1 A2 G2 AS GS
-101
T P-d
-101
T 1-u
-101
T 1-d
AP GP A1 G1 A2 G2 AS GS
-101
T 2-u
AP GP A1 G1 A2 G2 AS GS
-101
T 2-d
-101
T S-u
-101
T S-d
AP GP A1 G1 A2 G2 AS GS
AP GP A1 G1 A2 G2 AS GS
AP GP A1 G1 A2 G2 AS GS
UPDOWN
Fig 3 Correlation between air flows gas flows and temperatures
32 Incorporation of derived variables
During the stable heating stage the quantity of heat absorbed and removed from the
slabs from each furnace zone is relatively constant Hence the derived variable TG
ratio can be treated as an index for the combustion efficiency level This is because a
higher TG ratio signifies more combustion heat generated from unit gas ie higher
combustion efficiency
Moreover it is known that the appropriate air and fuel ratio is vital for the
combustion efficiency so the air-gas ratio (AG) is utilized as another derived variable
for the research Again the correlation analysis is performed for two types of derived
variables The correlations between different variables including the AG ratio G and
the TG ratio are shown in Fig 4 It can be clearly seen from the four red rectangle
blocks that TG of each zone is only remarkably related to AG and G in the same zone
10
183
184
185
186
187
188
189
190
191
192
193
194
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
Furthermore as related to the lsquouprsquo or lsquodownrsquo areas in one zone TG in each area has the
highest correlation with AG and G in the same area while AG and G in the opposite
area of the same zone is secondary This can be easily seen from the red and blue color
markings in each red rectangle block Thus it is of great significance to introduce these
derived ratio variables
-101
TG
P-u
AGP GP AG1G1 AG2 G2 AGS GS
-101
TG
P-d
AGP GP AG1 G1 AG2 G2 AGS GS
-101
TG
1-u
AGP GP AG1 G1 AG2 G2 AGS GS
-101
TG
1-d
AGP GP AG1 G1 AG2 G2 AGS GS
-101
TG
2-u
AGP GP AG1 G1 AG2 G2 AGS GS
-101
TG
2-d
AGP GP AG1 G1 AG2 G2 AGS GS
-101
TG
S-u
AGP GP AG1 G1 AG2 G2 AGS GS
-101
TG
S-d
AGP GP AG1 G1 AG2 G2 AGS GS
UPDOWN
Fig 4 Correlation between the air-gas ratio gas flow and TG ratio
33 LDA analysis
Linear discriminant analysis (LDA) aims to finding a projection direction that
maximizes the separation of class means and minimizes the within-class variance [25]
In this section LDA is utilized to identify the discriminating variables that play an
important role in determining combustion efficiency levels All the data are partitioned
into five groups according to their efficiency levels
11
195
196
197
198
199
200
201
202
203
204
205
206
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
Based on a descending order of TG ratios five efficiency levels for the lsquouprsquo area of
the soaking zone are denoted as HH H M L and LL LDA is conducted on the three
groups of the data with levels of HH M and LL
Fig 5 shows the scattering of the LDA projections of the process observations
collected at the three efficiency levels where y1 and y2 correspond to the first two LDA
components that contain most discriminant information The weighting factors ( and
) of the 16 input variables composing the projections y1 and y2 are shown in Fig 6
where and From left to right the 16 input variables are defined as
the eight AG ratio variables and the eight gas flow variables with the order of variables
of each kind P-u P-d 1-u 1-d 2-u 2-d S-u and S-d
-8 -6 -4 -2 0 2 4 6-4
-3
-2
-1
0
1
2
3
4
5
6
Projection y1
Pro
ject
ion
y2
HHMLL
Fig 5 Scattering of LDA projections y1 and y2 for three efficiency levels
12
207
208
209
210
211
212
213
214
215
216
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
AG_S-u-4
-2
0
2
Wei
ghtin
g fa
ctor
for y
1
AG_S-u
-101234
Wei
ghtin
g fa
ctor
for y
2
Fig 6 Weight factors of various input variables for projections y1 and y2
The two figures reveal that the three groups are clearly discriminated by the LDA
projection and the most significant variables for the different efficiency levels are the
derived variables ie the AG in the lsquouprsquo area of the soaking zone A similar result can
be obtained in the lsquodownrsquo area or any area of the other zones Therefore AG in each
heating area is the key manipulated variable that determines the different combustion
efficiency
4 Modeling and prediction of temperature and temperature-gas ratio
For the model-based operation optimization the models for the temperature and TG
ratio based on the NNG algorithm will be developed and compared
41 NNG variable selection algorithm
The NNG method can be generalized into a two-stage shrinkage method In the first
stage the sign for each variable is determined using the ordinary least square procedure
13
217
218
219
220
221
222
223
224
225
226
227
228
229
230
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
and in the second stage the corresponding magnitudes are computed by solving a series
of constrained quadratic programming
A set of observation data is provided where is the input matrix whose
columns represent the measured candidate variables and is the corresponding
vector of the response data The following expression is given with the number of the
response variable being equal to 1 but a similar procedure can be generalized to any
number of variables Let and be normalized to the zero-mean and unit standard
deviation Additionally let be a set of the ordinary least square estimates of the
coefficients of the following linear model then
(1)
The second stage shrinkage can be formulated as the following optimization problem
subject to
(2)
As decreased and the NNG is tightened more of the become zero and the
remaining nonzero coefficients are shrunk A solution path exists with on which
the appropriate shrinkage can be selected Conventionally the v-fold cross-validation is
used to estimate the prediction error and to select the best solution in the solution path
so as to minimize the prediction or model error
42 Modeling and prediction of temperature and TG ratio
The input-output relations change with time as the reheating process develops An
adaptive modeling strategy is often used to resolve time-varying characteristics of
14
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
industrial processes In this paper the NNG-based regression modeling prediction and
optimization is implemented in a moving window manner where the size of the
window and the length of the moving step are selected as 1000 and 100 respectively In
addition the prediction horizon is also set as 100 In each step the NNG regression
model is built based on the data in the current window which is then used for the
prediction and optimization in the subsequent prediction horizon Next the window
moves forward by replacing the oldest 100 samples for model training by the
observations collected in the previous prediction horizon This moving window strategy
is workable because in the reheating process the input-output relation is slowly time-
varying and the model is still valid for the prediction and optimization in the subsequent
short time
The modeling of the temperature T is based on the air flow rates and gas flow rates
while the modeling for the TG ratio is based on the AG ratios and gas flow rates
Taking the lsquouprsquo area of the soaking zone as an example the prediction result for T and
the TG ratio in the lsquouprsquo area of the soaking zone is shown in Figs 7 and 8 respectively
0 02 04 06 08 1 12 14 16 18 2
x 104
1140
1160
1180
1200
1220
1240
1260
1280
1300
1320
Sampling intervals
T S-u (
)
OriginalPredicted
Fig 7 Prediction of temperature in the lsquouprsquo area of the soaking zone
15
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
0 02 04 06 08 1 12 14 16 18 2
x 104
02
04
06
08
1
12
14
16
18
2
Sampling intervals
TG
S-u
OriginalPredicted
Fig 8 Prediction of TG ratio in the lsquouprsquo area of the soaking zone
A comparison between Figs 7 and 8 show that the prediction of the TG ratio is more
accurate than the prediction of temperature The average R2 in 190 NNG regressions is
0935 for the TG ratio while the average R2 is only 0814 for temperature This is
consistent with the statistical analysis which reveals that the correlation between AG
and TG is higher than that between A and T These results indicate that derived
variables are more meaningful for the purpose of prediction and the modeling of the
TG ratio is more appropriate for implementing optimization
The selected frequency for each variable in 190 NNG regressions and the coefficients
of each variable in 20 NNG regressions for the TG ratio modeling of the lsquouprsquo area of
the soaking zone are shown in Figs 9 and 10 respectively Fig 9 shows that the selected
frequency of the variables in the lsquouprsquo area of the soaking zone is much higher than
variables in other zones Similarly as is shown in Fig 10 the NNG regression
coefficients of the variables in the lsquouprsquo area of the soaking zone are much larger than the
variables in the other areas These results indicated that variables in the lsquouprsquo area of
16
269
270
271
272
273
274
275
276
277
278
279
280
281
282
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
soaking zone are mostly contributed to the modeling of the TG ratio in same area
which is quite reasonable Similar results can be obtained for the other areas
0
20
40
60
80
100
120
140
160
180
200
Sel
ecte
d fre
quen
cy
AGP-u AGP-d AG1-u AG1-d AG2-u AG2-d AGS-u AGS-d GP-u GP-d G1-u G1-d G2-u G2-d GS-u GS-d
Fig 9 Variable selected frequency of over 190 runs for the TG ratio model of the lsquouprsquo
area of the soaking zone
0 2 4 6 8 10 12 14 16 18 20-05
05
AG
P-u
0 2 4 6 8 10 12 14 16 18 20-05
05
GP
-u
0 2 4 6 8 10 12 14 16 18 20-1
0
AG
P-d
0 2 4 6 8 10 12 14 16 18 20-1
0
GP
-d
0 2 4 6 8 10 12 14 16 18 20-05
05
AG
1-u
0 2 4 6 8 10 12 14 16 18 20-05
05
G1-
u
0 2 4 6 8 10 12 14 16 18 20-05
05
AG
1-d
0 2 4 6 8 10 12 14 16 18 20-05
05
G1-
d
0 2 4 6 8 10 12 14 16 18 20-1
0
AG
2-u
0 2 4 6 8 10 12 14 16 18 200
1
G2-
u
0 2 4 6 8 10 12 14 16 18 20-1
0
AG
2-d
0 2 4 6 8 10 12 14 16 18 20-1
0
G2-
d
0 2 4 6 8 10 12 14 16 18 200
1
AG
S-u
0 2 4 6 8 10 12 14 16 18 20-1
0
GS
-u
0 2 4 6 8 10 12 14 16 18 20-05
05
AG
S-d
0 2 4 6 8 10 12 14 16 18 20-05
05
GS
-d
Fig 10 Part of the NNG regression coefficients of each variable over 190 runs for the
TG ratio model of the lsquouprsquo area of the soaking zone
It is remarkable that the NNG regression coefficients corresponding to the AG ratio
in the lsquouprsquo area of the soaking zone are consistently positive On the contrary the
coefficients of the gas flow rate in the same area are consistently negative This
indicates that under this condition the rise of the AG leads to the increase of the TG
17
283
284
285
286
287
288
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
while the rise of the G leads to the drop of the TG This property is helpful for
performing optimization and improve combustion efficiency
For comparison the modeling results for two other algorithms artificial neural
network (ANN) and partial least squares (PLS) [26] are compared with the NNG
algorithm in the following three aspects [27]
(1) Model Size the number of variables selected for modeling
(2) Model Magnitude the mean of the L1 norm of the regression coefficients
(3) Prediction Precision the mean squared prediction error (MSPE)
Summary of the algorithm comparison is shown in Table 2 It can be seen that the
superiorities of the NNG regression in model size model magnitude and model
precision are remarkable
Table 2 Summary of algorithm comparison
Index NNG ANN PLS
Model Size 849 16 16
Model Magnitude 10075 -- 12150
MSPE 00093 00108 00140
5 Model-based optimization
51 Implemention of model-based optimization operation
The goal of optimization is to seek an optimal combination of AG and G in the lsquouprsquo
area of the soaking zone so as to minimize the gas consumption while keeping the
temperature at the target value According to the heating schedule the most expected
temperature in the soaking zone is 1200 Therefore the target temperature
is set at this value to achieve the desired heating effect As shown in Fig 11 in the real
18
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
operation the temperature fluctuates around the target value because of the imperfect
control performance However in order to simplify the analysis and compute the
maximum possible energy saving perfect temperature control is assumed when
adopting the model-based optimization strategy In other words it is assumed that the
actual temperature in the lsquouprsquo area of the soaking zone can be adjusted to the expected
temperature ie 1200
0 02 04 06 08 1 12 14 16 18 2
x 104
1150
1200
1250
1300
Sampling intervals
T S-u (
)
OriginalTarget
Fig 11 Original and target temperature in the lsquouprsquo area of the soaking zone
The adjustment scheme takes the maximum value and minimum value of the original
operation data as the upper and lower bounds for the adjustment Moreover in order to
assure the validity of the linear model the increment or decrement of AG and G should
not beyond of the original value (considered as 10 for the purposes of this paper)
The adjustment strategy can be formulated as
19
307
308
309
310
311
312
313
314
315
316
317
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
(3)
where NNG() denotes the NNG regression model of the TG ratio and are the
NNG regression coefficients of the AG and G in the lsquouprsquo area of the soaking zone for
modeling TG in same area
With the model-based strategy and adaptive model of the TG ratio given in Section
42 the optimization adjustment results can be obtained as follows
The adjustment of gas flow in the lsquouprsquo area of the soaking zone shown in Fig 12
illustrates that the implementation of the model-based optimization operation can reduce
the consumption of the gas flow Compared with the original case 938 of the gas can
be saved on average by utilizing the model-based adjustment
0 02 04 06 08 1 12 14 16 18 2
x 104
-800
-700
-600
-500
-400
-300
-200
-100
0
Sampling intervals
Adj
ustm
ent o
f gas
flow
m(
3 h)
20
318
319
320
321
322
323
324
325
326
327
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
(a) Adjustment amount of gas flow
0 02 04 06 08 1 12 14 16 18 2
x 104
500
1000
1500
2000
2500
3000
3500
4000
4500
5000G
as fl
owm
(3 h
)
Sampling intervals
OriginalAdjusted
(b) Comparison of original and adjusted gas flow
095 096 097 098 099 1 101 102 103 104 105
x 104
500
1000
1500
2000
2500
3000
3500
4000
4500
5000
Gas
flow
m(
3 h)
Sampling intervals
OriginalAdjusted
(c) Comparison of original and adjusted gas flow in interval [9501-10500]
Fig 12 Adjustment of gas flow in the lsquouprsquo area of the soaking zone
The adjustment of the air flow in the lsquouprsquo area of the soaking zone given in Fig 13
shows that the air flow is reduced in most cases and only increased for a few cases In
21
328
329
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
general 681 of the air flow is reduced after the implementation of the model-based
optimization
0 02 04 06 08 1 12 14 16 18 2
x 104
-2000
-1500
-1000
-500
0
500
1000
Sampling intervals
Adj
ustm
ent o
f air
flow
(m3 h
)
(a) Adjustment amount of air flow
0 02 04 06 08 1 12 14 16 18 2
x 104
4000
5000
6000
7000
8000
9000
10000
Air
flow
m(
3 h)
Sampling intervals
OriginalAdjusted
(b) Comparison of original and adjusted air flow
22
330
331
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
095 096 097 098 099 1 101 102 103 104 105
x 104
4000
5000
6000
7000
8000
9000
10000
Air
flow
m(
3 h)
Sampling intervals
OriginalAdjusted
(c) Comparison of original and adjusted air flow in interval [9501-10500]
Fig 13 Adjustment of air flow in the lsquouprsquo area of the soaking zone
52 Discussions
In this section detailed analysis for the optimization operation results is provided
The statistics of optimization at the boundary conditions shown in Table 3 illustrates
that the adjusted value for G reaches its lower limit in most cases (8045 for
and 687 for ) which is the most energy-efficient point In a number of cases
(1079 for and 001 for ) the adjusted AG reaches its upper
limit These results indicate that the optimization operations maximize the combustion
efficiency by decreasing G and increasing the AG ratio up to the boundary conditions
However the lower limit is also unexpectedly attained in a small number of cases
(110 for and 078 for ) for which a more detailed analysis is
given below
23
332
333
334
335
336
337
338
339
340
341
342
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
Table 3 Statistics of optimization at boundary conditions
Total
Amount 0 1374 0 16090 17464
Percentag
e 0 687 0 8045
8732
Total
Amount 2 220 2158 156 2536
Percentag
e 001 110 1079 078
1268
Table 4 Statistics for different optimization operations
Adjustment
Gdarr
Guarr TotalAGuarr AGdarr
Tdarr Tuarr Tdarr Tuarr
Amount 9548 3744 6328 380 0 20000
Percentage 4774 1872 3164 190 0 100
The statistics for different optimization operations is performed and shown in Table
4 It can be seen that all the adjusted operations result in the reduction of gas
consumption In over half the cases (4774 with decreased temperature and 1872
with increased temperature) AG is adjusted to higher levels Meanwhile a number of
the adjustments (3164) lower the temperature by reducing the G and AG at the same
time This result indicates that the original temperature cannot be adjusted to the target
by only reducing G in the constraint conditions The remaining cases (19) are
24
343
344
345
346
347
348
349
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
relatively special in which an excess of air is supplied in the original operation and the
adjusted operation thereby increasing the temperature by a smaller AG with less gas
consumption
6 Conclusion
Based on the actual operation data this paper aims to explore the improvement of the
combustion efficiency and the room for energy conservation Correlation analysis and
LDA show that it is of great significance to introduce two derived ratio variables which
are the AG ratio and the TG ratio A type of combustion efficiency utilizing an on-line
soft sensor is put forward by employing a NNG variable selection algorithm which
provides a good solution to the combustion efficiency real-time measurement problem
of a reheating furnace The implementation of the model-based optimization is studied
based on the actual operational data Detailed analysis for the optimization results is
given for the different cases The results show that significant energy conservation can
be achieved when the furnace operation is optimized by using the developed soft sensor
model
Acknowledgement
The authors would like to thank the financial support provided by the National Nature
Science Foundation of China under Grant 61171145 Y Yao was supported by Ministry
of Science amp Technology ROC under Grant number MOST 104-2221-E-007-129
References
25
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[1] Z J Wang Q D Wu and T Y Chai Optimal-setting control for complicated
industrial processes and its application studyControl Engineering Practice vol
12 pp 65-74 2004
[2] A Steinboeck K Graichen and A Kugi Dynamic Optimization of a Slab
Reheating Furnace With Consistent Approximation of Control VariablesIEEE
Transactions on Control Systems Technology vol 19 pp 1444-1456 2011
[3] B T Zhang C Y Wang Q Qin and L Li Influence of Boiler Combustion
Adjustment on NOxEmission and Boiler EfficiencyAdvanced Materials
Research vol 732-733 pp 234-237 2013
[4] C K Yoo and IB Lee Soft Sensor and Adaptive Model-Based Dissolved
Oxygen Control for Biological Wastewater Treatment ProcessesEnvironmental
Engineering Science vol 21 pp 331-340 2004
[5] S A Bhat D N Saraf S Gupta and S K Gupta Use of Agitator Power as a
Soft Sensor for Bulk Free-Radical Polymerization of Methyl Methacrylate in
Batch ReactorsIndustrial amp Engineering Chemistry Research vol 45 pp 4243-
4255 2006
[6] K Desai Y Badhe S S Tambe and B D Kulkarni Soft-sensor development
for fed-batch bioreactors using support vector regressionBiochemical
Engineering Journal vol 27 pp 225-239 2006
[7] Y P Badhe Lonari J Tambe S S amp Kulkarni B D Improve polyethylene
process control and product qualityHydrocarbon Processing vol 86 pp 53-60
2007
26
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[8] N K Nath K Mandal A K Singh B Basu C Bhanu S Kumar et al Ladle
furnace on-line reckoner for prediction and control of steel temperature and
compositionIronmaking amp Steelmaking vol 33 pp 140-150 2006
[9] A J Yan T Y Chai F H Wu and P Wang Hybrid intelligent control of
combustion process for ore-roasting furnaceJournal of Control Theory and
Applications vol 6 pp 80-85 2008
[10] J Li W M Zhong H Cheng X D Kong and F Qian A data-driven soft
sensor modeling for furnace temperature of Opposed Multi-Burner gasifier pp
705-710 2011
[11] Y H Yang Y H Liu X Z Liu and S K Qin Billet temperature soft sensor
model of reheating furnace based on RVM method pp 4003-4006 2011
[12] J H Wang C Wang X F Zhu and X K Fang Application of soft sensor in
welding seam tracking prediction based on LSSVM and PSO with compression
factor pp 2441-2446 2013
[13] L Balbis J Balderud and M J Grimble Nonlinear predictive control of steel
slab reheating furnace pp 1679-1684 2008
[14] A Steinboeck D Wild T Kiefer and A Kugi A mathematical model of a slab
reheating furnace with radiative heat transfer and non-participating gaseous
media International Journal Of Heat And Mass Transfer vol 53 pp 5933-
5946 Dec 2010
[15] A Steinboeck D Wild and A Kugi Nonlinear model predictive control of a
continuous slab reheating furnace Control Engineering Practice vol 21 pp
495-508 2013
27
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[16] C Zhang T Ishii and S Sugiyama Numerical Modeling Of the Thermal
Performance Of Regenerative Slab Reheat Furnaces Numerical Heat Transfer
Part A Applications vol 32 pp 613-631 1997
[17] J G Kim and K Y Huh Prediction of Transient Slab Temperature Distribution
in the Re-heating Furnace of a Walking-beam Type for Rolling of Steel Slabs
ISIJ International vol 40 pp 1115-1123 2000
[18] J G Kim K Y and H I T K Three-Dimensional Analysis Of the Walking-
Beam-Type Slab Reheating Furnace In Hot Strip Mills Numerical Heat
Transfer Part A Applications vol 38 pp 589-609 2000
[19] CT Hsieh MJ Huang ST Lee and CH Wang Numerical Modeling of a
Walking-Beam-Type Slab Reheating Furnace Numerical Heat Transfer Part A
Applications vol 53 pp 966-981 2008
[20] MJ Huang CT Hsieh ST Lee and CH Wang A Coupled Numerical Study
of Slab Temperature and Gas Temperature in the Walking-Beam-Type Slab
Reheating Furnace Numerical Heat Transfer Part A Applications vol 54 pp
625-646 2008
[21] Z Song and A Kusiak Constraint-Based Control of Boiler Efficiency A Data-
Mining Approach IEEE Transactions on Industrial Informatics vol 3 pp 73-
83 2007
[22] J Q Li J J Gu and C L Niu The Operation Optimization based on
Correlation Analysis of Operation Parameters in Power Plant pp 138-141
2008
28
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[23] J G Wang S S Shieh S S Jang D S H Wong and C W Wu A two-tier
approach to the data-driven modeling on thermal efficiency of a BFGcoal co-
firing boiler Fuel vol 111 pp 528-534 Sep 2013
[24] L Breiman Better Subset Regression Using the Nonnegative Garrote
Technometrics vol 37 pp 373-384 1995
[25] R A Fisher The use of multiple measurements in taxonomic problems
AnnHum Genet vol 7 pp 179-188 1936
[26] J Liu Developing a soft sensor based on sparse partial least squares with
variable selection Journal of Process Control vol 24 pp 1046-1056 2014
[27] K Sun J Liu JL Kang SS Jang D SH Wong and DS Chen
Development of a variable selection method for soft sensor using artificial
neural network and nonnegative garrote Journal of Process Control vol 24 pp
1068-1075 2014
29
439
440
441
442
443
444
445
446
447
448
449
450
451
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
of a back propagation (BP) neural network is studied which indicates that the furnace
temperature predictive model integrating the principal component analysis (PCA) and
the BP neural network has a promising performance with good predictive precision
[10] A soft sensor modeling method is proposed to predict the billet temperature of the
reheating furnace based on a relevance vector machine (RVM) which has a higher
prediction accuracy and a certain practical significance to the on-site production of a
reheating furnace [11] The least square support vector machine (LSSVM) inductance
model optimized by the particle swarm optimization method with a compression factor
(PSO-CF) algorithm is presented for the difficulty of time prediction which can
improve PSO convergence accuracy and effectively avoid falling into a local optimum
[12] However the soft sensor developed for combustion efficiency was not
investigated in these research efforts which is significant for energy conservation
On the premise of the model prediction accuracy the model-based control makes
optimal operation feasible which can then be successfully employed to operate a
reheating furnace in an efficient way The potential of the nonlinear model predictive
control techniques is explored to improve the temperature control for the metal slabs in
a hot mill reheating furnace and particularly whether or not these control techniques
can be exploited to reduce energy consumption [13] Steinboeck et al developed a
mathematical model of the reheating process of steel slabs in industrial fuel-fired
furnaces in 2010 They exploited a dynamic optimization method for temperature
control of the steel slabs in a continuous reheating furnace and a temperature control
method for reheating steel slabs in an industrial furnace in 2011 They also designed a
nonlinear model predictive controller for a reheating furnace for steel slabs in 2013 [2
14 15]
4
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
Obviously the research on the numerical model for the heating performance of
reheating furnace can be done based on basic combustion theory and heat transmission
characteristics Many scholars devote themselves to the simulations of the heat flow
phenomenon in the reheating furnace Zhang et al attempted to apply a computational
fluid dynamic (CFD) simulation to predict the combustion performance for a reheating
furnace by simplifying the furnace to a cuboid and assuming that the slab possesses
infinite length and enters the reheating furnace at a fixed speed [16] The CFD method
has been applied to the study of reaction turbulence radiation heat transmission and the
calculation for the steady state heat transmission rate of the slab under the given
temperature [17] In the works of references [16-20] the authors used the given
temperature data of slabs to compute the steady flow and temperature field However
owing to the changing operating conditions the actual implementation of these
numerical model methods still bristle with difficulties although the methods mentioned
above are feasible for the prediction Thus for an online application it is necessary to
adopt a real time data-driven model to resolve the time varying characteristics
Proper variable selection is an important step in model building for a large-scale
combustion system A well-trimmed variable dimension ensures the acquired model is
transparent comprehensible and robust Some studies reported that the combustion
model built by a selected subset of input variables provide more accurate predictions of
combustion efficiency than the entire set of variables [21-23] Recently shrinkage
methods which conduct variable selection by shrinking or setting some coefficients of a
ldquogreedyrdquo model to zero have received significant attention A popular form of these
methods is the non-negative garrote (NNG) [23 24]
5
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
Against this background this paper aims to propose a combustion efficiency index
for the reheating furnace and investigate for room in improvement regarding energy
conservation The primary contribution is a practical combustion efficiency index the
incorporation of the derived variables and soft-sensing method for the optimization of
combustion efficiency of reheating furnaces The derived variables are found more
physically meaningful than the plain variables when constructing the model of
combustion efficiency By employing a NNG variable selection procedure an adaptive
scheme for combustion efficiency modeling and adjustment is proposed and virtually
implemented for a rolling reheating furnace The results show that there is significant
room for energy conservation
The remainder of the paper is organized as follows In the next section the reheating
furnace and the data preprocessing is described In Section 3 the statistics analysis for
different variables and the formation of derived variables are presented In Section 4
the framework of an adaptive model based on NNG variable selection is presented and
two models developed for the temperature and temperature-gas (TG) ratio are
compared according to the model prediction precision A model-based optimization
scheme is provided and applied to the combustion efficiency improvement for an actual
case of a reheating furnace presented in Section 5 Several remarks and a summary
conclude the last section
2 Plant description and data preprocessing
The schematic of the heating process in the rolling mill reheating furnace is shown in
Fig 1 There are four zones in the reheating furnace including the preheating zone (P)
the first heating zone (1) the second heating zone (2) and the soaking zone (S) The
6
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
steel slab moves through the four zones in turn and is heated to the demanded state
using a specific temperature increase curve As is shown in Table 1 the soaking zone
has two areas that are defined as up and down and both of the areas possess the same
five variables including two manipulated variables the air flow (A) and the gas flow
(G) and three temperatures in left center and right sections of the area (T-l T-c and T-
r) The other zones have the same variables as the soaking zone hence there are 40
variables in total for the reheating furnace
Fig 1 The schematic of the heating process in the reheating furnace
Table 1 Variables and descriptions in the soaking zone
Variable Description unit
AS-u Air flow in the lsquouprsquo area Nm3h
AS-d Air flow in the lsquodownrsquo area Nm3h
GS-u Gas flow in the lsquouprsquo area Nm3h
GS-d Gas flow in the lsquodownrsquo area Nm3h
TS-ul Temperature in the left part of the lsquouprsquo area
TS-uc Temperature in the center part of the lsquouprsquo area
TS-ur Temperature in the right part of the lsquouprsquo area
7
143
144
145
146
147
148
149
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
TS-dl Temperature in the left part of the lsquodownrsquo area
TS-dc Temperature in the center part of the lsquodownrsquo area
TS-dr Temperature in the right part of the lsquodownrsquo area
A data set of 20000 samples was used in this study The samples were collected from
an actual reheating furnace in a large iron and steel plant located in Shanghai from
September 14 to September 27 2014 The operational data is taken on a per minute
basis
1
092
093
042
041
046
092
1
091
04
039
045
093
091
1
038
038
045
042
04
038
1
096
096
041
039
038
096
1
096
046
045
045
096
096
1
TS-ul
TS-uc
TS-ur
TS-dl
TS-dc
TS-dr
TS-ul TS-uc TS-ur TS-dl TS-dc TS-dr
04
05
06
07
08
09
1
Fig 2 Correlation between each temperature in the soaking zone
In order to investigate the relation among different temperatures in each zone
correlation analysis is conducted for the soaking zone as illustrated in Fig 2 It can be
seen that the temperature in three parts of the lsquouprsquo area is highly correlated with a
correlation coefficient greater than 09 The similar results exist for the lsquodownrsquo area as
well On the contrary the correlation coefficient between temperatures in one part of
8
150
151
152
153
154
155
156
157
158
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
lsquouprsquo area and any part of lsquodownrsquo area does not exceed 05 Therefore for the reduction
of the data dimension the three temperatures in the lsquouprsquo area or the lsquodownrsquo area can be
treated as only one variable which can be taken as the mean value or the first principal
component acquired from the PCA analysis Considering the reservation of the variable
physical meaning the former is preferred and used
3 Statistics analysis and incorporation of derived variables
In this section in order to uncover the physical knowledge for the actual operation
guidance and confirmation statistical analysis is performed for the 16 input manipulated
variables and the eight output variables (ie the temperatures in the four zones) of a
reheating furnace system For the combustion efficiency evaluation and modeling two
types of derived ratio variables are introduced which is helpful to reveal the
information included in the data
31 Correlation analysis
Correlation analysis between the temperatures (T) in each area and all of the air flows
(A) and the gas flows (G) is performed and shown in Fig 3 It can be seen that only the
temperatures in both areas of the soaking zone are the most highly related to the air flow
and the gas flow in its own zone However this phenomenon does not occur in the other
three zones The temperatures in the second heating zone mainly depend on the air flow
and the gas flow in its own zone as well as the nearby first heating zone As for the
preheating zone and first heating zone no apparent correlation can be observed
Obviously these analysis results could not tell us the explicit information about how to
evaluate the efficiency levels and key manipulated variables
9
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
-101
T P-u
AP GP A1 G1 A2 G2 AS GS
AP GP A1 G1 A2 G2 AS GS
AP GP A1 G1 A2 G2 AS GS
-101
T P-d
-101
T 1-u
-101
T 1-d
AP GP A1 G1 A2 G2 AS GS
-101
T 2-u
AP GP A1 G1 A2 G2 AS GS
-101
T 2-d
-101
T S-u
-101
T S-d
AP GP A1 G1 A2 G2 AS GS
AP GP A1 G1 A2 G2 AS GS
AP GP A1 G1 A2 G2 AS GS
UPDOWN
Fig 3 Correlation between air flows gas flows and temperatures
32 Incorporation of derived variables
During the stable heating stage the quantity of heat absorbed and removed from the
slabs from each furnace zone is relatively constant Hence the derived variable TG
ratio can be treated as an index for the combustion efficiency level This is because a
higher TG ratio signifies more combustion heat generated from unit gas ie higher
combustion efficiency
Moreover it is known that the appropriate air and fuel ratio is vital for the
combustion efficiency so the air-gas ratio (AG) is utilized as another derived variable
for the research Again the correlation analysis is performed for two types of derived
variables The correlations between different variables including the AG ratio G and
the TG ratio are shown in Fig 4 It can be clearly seen from the four red rectangle
blocks that TG of each zone is only remarkably related to AG and G in the same zone
10
183
184
185
186
187
188
189
190
191
192
193
194
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
Furthermore as related to the lsquouprsquo or lsquodownrsquo areas in one zone TG in each area has the
highest correlation with AG and G in the same area while AG and G in the opposite
area of the same zone is secondary This can be easily seen from the red and blue color
markings in each red rectangle block Thus it is of great significance to introduce these
derived ratio variables
-101
TG
P-u
AGP GP AG1G1 AG2 G2 AGS GS
-101
TG
P-d
AGP GP AG1 G1 AG2 G2 AGS GS
-101
TG
1-u
AGP GP AG1 G1 AG2 G2 AGS GS
-101
TG
1-d
AGP GP AG1 G1 AG2 G2 AGS GS
-101
TG
2-u
AGP GP AG1 G1 AG2 G2 AGS GS
-101
TG
2-d
AGP GP AG1 G1 AG2 G2 AGS GS
-101
TG
S-u
AGP GP AG1 G1 AG2 G2 AGS GS
-101
TG
S-d
AGP GP AG1 G1 AG2 G2 AGS GS
UPDOWN
Fig 4 Correlation between the air-gas ratio gas flow and TG ratio
33 LDA analysis
Linear discriminant analysis (LDA) aims to finding a projection direction that
maximizes the separation of class means and minimizes the within-class variance [25]
In this section LDA is utilized to identify the discriminating variables that play an
important role in determining combustion efficiency levels All the data are partitioned
into five groups according to their efficiency levels
11
195
196
197
198
199
200
201
202
203
204
205
206
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
Based on a descending order of TG ratios five efficiency levels for the lsquouprsquo area of
the soaking zone are denoted as HH H M L and LL LDA is conducted on the three
groups of the data with levels of HH M and LL
Fig 5 shows the scattering of the LDA projections of the process observations
collected at the three efficiency levels where y1 and y2 correspond to the first two LDA
components that contain most discriminant information The weighting factors ( and
) of the 16 input variables composing the projections y1 and y2 are shown in Fig 6
where and From left to right the 16 input variables are defined as
the eight AG ratio variables and the eight gas flow variables with the order of variables
of each kind P-u P-d 1-u 1-d 2-u 2-d S-u and S-d
-8 -6 -4 -2 0 2 4 6-4
-3
-2
-1
0
1
2
3
4
5
6
Projection y1
Pro
ject
ion
y2
HHMLL
Fig 5 Scattering of LDA projections y1 and y2 for three efficiency levels
12
207
208
209
210
211
212
213
214
215
216
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
AG_S-u-4
-2
0
2
Wei
ghtin
g fa
ctor
for y
1
AG_S-u
-101234
Wei
ghtin
g fa
ctor
for y
2
Fig 6 Weight factors of various input variables for projections y1 and y2
The two figures reveal that the three groups are clearly discriminated by the LDA
projection and the most significant variables for the different efficiency levels are the
derived variables ie the AG in the lsquouprsquo area of the soaking zone A similar result can
be obtained in the lsquodownrsquo area or any area of the other zones Therefore AG in each
heating area is the key manipulated variable that determines the different combustion
efficiency
4 Modeling and prediction of temperature and temperature-gas ratio
For the model-based operation optimization the models for the temperature and TG
ratio based on the NNG algorithm will be developed and compared
41 NNG variable selection algorithm
The NNG method can be generalized into a two-stage shrinkage method In the first
stage the sign for each variable is determined using the ordinary least square procedure
13
217
218
219
220
221
222
223
224
225
226
227
228
229
230
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
and in the second stage the corresponding magnitudes are computed by solving a series
of constrained quadratic programming
A set of observation data is provided where is the input matrix whose
columns represent the measured candidate variables and is the corresponding
vector of the response data The following expression is given with the number of the
response variable being equal to 1 but a similar procedure can be generalized to any
number of variables Let and be normalized to the zero-mean and unit standard
deviation Additionally let be a set of the ordinary least square estimates of the
coefficients of the following linear model then
(1)
The second stage shrinkage can be formulated as the following optimization problem
subject to
(2)
As decreased and the NNG is tightened more of the become zero and the
remaining nonzero coefficients are shrunk A solution path exists with on which
the appropriate shrinkage can be selected Conventionally the v-fold cross-validation is
used to estimate the prediction error and to select the best solution in the solution path
so as to minimize the prediction or model error
42 Modeling and prediction of temperature and TG ratio
The input-output relations change with time as the reheating process develops An
adaptive modeling strategy is often used to resolve time-varying characteristics of
14
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
industrial processes In this paper the NNG-based regression modeling prediction and
optimization is implemented in a moving window manner where the size of the
window and the length of the moving step are selected as 1000 and 100 respectively In
addition the prediction horizon is also set as 100 In each step the NNG regression
model is built based on the data in the current window which is then used for the
prediction and optimization in the subsequent prediction horizon Next the window
moves forward by replacing the oldest 100 samples for model training by the
observations collected in the previous prediction horizon This moving window strategy
is workable because in the reheating process the input-output relation is slowly time-
varying and the model is still valid for the prediction and optimization in the subsequent
short time
The modeling of the temperature T is based on the air flow rates and gas flow rates
while the modeling for the TG ratio is based on the AG ratios and gas flow rates
Taking the lsquouprsquo area of the soaking zone as an example the prediction result for T and
the TG ratio in the lsquouprsquo area of the soaking zone is shown in Figs 7 and 8 respectively
0 02 04 06 08 1 12 14 16 18 2
x 104
1140
1160
1180
1200
1220
1240
1260
1280
1300
1320
Sampling intervals
T S-u (
)
OriginalPredicted
Fig 7 Prediction of temperature in the lsquouprsquo area of the soaking zone
15
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
0 02 04 06 08 1 12 14 16 18 2
x 104
02
04
06
08
1
12
14
16
18
2
Sampling intervals
TG
S-u
OriginalPredicted
Fig 8 Prediction of TG ratio in the lsquouprsquo area of the soaking zone
A comparison between Figs 7 and 8 show that the prediction of the TG ratio is more
accurate than the prediction of temperature The average R2 in 190 NNG regressions is
0935 for the TG ratio while the average R2 is only 0814 for temperature This is
consistent with the statistical analysis which reveals that the correlation between AG
and TG is higher than that between A and T These results indicate that derived
variables are more meaningful for the purpose of prediction and the modeling of the
TG ratio is more appropriate for implementing optimization
The selected frequency for each variable in 190 NNG regressions and the coefficients
of each variable in 20 NNG regressions for the TG ratio modeling of the lsquouprsquo area of
the soaking zone are shown in Figs 9 and 10 respectively Fig 9 shows that the selected
frequency of the variables in the lsquouprsquo area of the soaking zone is much higher than
variables in other zones Similarly as is shown in Fig 10 the NNG regression
coefficients of the variables in the lsquouprsquo area of the soaking zone are much larger than the
variables in the other areas These results indicated that variables in the lsquouprsquo area of
16
269
270
271
272
273
274
275
276
277
278
279
280
281
282
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
soaking zone are mostly contributed to the modeling of the TG ratio in same area
which is quite reasonable Similar results can be obtained for the other areas
0
20
40
60
80
100
120
140
160
180
200
Sel
ecte
d fre
quen
cy
AGP-u AGP-d AG1-u AG1-d AG2-u AG2-d AGS-u AGS-d GP-u GP-d G1-u G1-d G2-u G2-d GS-u GS-d
Fig 9 Variable selected frequency of over 190 runs for the TG ratio model of the lsquouprsquo
area of the soaking zone
0 2 4 6 8 10 12 14 16 18 20-05
05
AG
P-u
0 2 4 6 8 10 12 14 16 18 20-05
05
GP
-u
0 2 4 6 8 10 12 14 16 18 20-1
0
AG
P-d
0 2 4 6 8 10 12 14 16 18 20-1
0
GP
-d
0 2 4 6 8 10 12 14 16 18 20-05
05
AG
1-u
0 2 4 6 8 10 12 14 16 18 20-05
05
G1-
u
0 2 4 6 8 10 12 14 16 18 20-05
05
AG
1-d
0 2 4 6 8 10 12 14 16 18 20-05
05
G1-
d
0 2 4 6 8 10 12 14 16 18 20-1
0
AG
2-u
0 2 4 6 8 10 12 14 16 18 200
1
G2-
u
0 2 4 6 8 10 12 14 16 18 20-1
0
AG
2-d
0 2 4 6 8 10 12 14 16 18 20-1
0
G2-
d
0 2 4 6 8 10 12 14 16 18 200
1
AG
S-u
0 2 4 6 8 10 12 14 16 18 20-1
0
GS
-u
0 2 4 6 8 10 12 14 16 18 20-05
05
AG
S-d
0 2 4 6 8 10 12 14 16 18 20-05
05
GS
-d
Fig 10 Part of the NNG regression coefficients of each variable over 190 runs for the
TG ratio model of the lsquouprsquo area of the soaking zone
It is remarkable that the NNG regression coefficients corresponding to the AG ratio
in the lsquouprsquo area of the soaking zone are consistently positive On the contrary the
coefficients of the gas flow rate in the same area are consistently negative This
indicates that under this condition the rise of the AG leads to the increase of the TG
17
283
284
285
286
287
288
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
while the rise of the G leads to the drop of the TG This property is helpful for
performing optimization and improve combustion efficiency
For comparison the modeling results for two other algorithms artificial neural
network (ANN) and partial least squares (PLS) [26] are compared with the NNG
algorithm in the following three aspects [27]
(1) Model Size the number of variables selected for modeling
(2) Model Magnitude the mean of the L1 norm of the regression coefficients
(3) Prediction Precision the mean squared prediction error (MSPE)
Summary of the algorithm comparison is shown in Table 2 It can be seen that the
superiorities of the NNG regression in model size model magnitude and model
precision are remarkable
Table 2 Summary of algorithm comparison
Index NNG ANN PLS
Model Size 849 16 16
Model Magnitude 10075 -- 12150
MSPE 00093 00108 00140
5 Model-based optimization
51 Implemention of model-based optimization operation
The goal of optimization is to seek an optimal combination of AG and G in the lsquouprsquo
area of the soaking zone so as to minimize the gas consumption while keeping the
temperature at the target value According to the heating schedule the most expected
temperature in the soaking zone is 1200 Therefore the target temperature
is set at this value to achieve the desired heating effect As shown in Fig 11 in the real
18
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
operation the temperature fluctuates around the target value because of the imperfect
control performance However in order to simplify the analysis and compute the
maximum possible energy saving perfect temperature control is assumed when
adopting the model-based optimization strategy In other words it is assumed that the
actual temperature in the lsquouprsquo area of the soaking zone can be adjusted to the expected
temperature ie 1200
0 02 04 06 08 1 12 14 16 18 2
x 104
1150
1200
1250
1300
Sampling intervals
T S-u (
)
OriginalTarget
Fig 11 Original and target temperature in the lsquouprsquo area of the soaking zone
The adjustment scheme takes the maximum value and minimum value of the original
operation data as the upper and lower bounds for the adjustment Moreover in order to
assure the validity of the linear model the increment or decrement of AG and G should
not beyond of the original value (considered as 10 for the purposes of this paper)
The adjustment strategy can be formulated as
19
307
308
309
310
311
312
313
314
315
316
317
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
(3)
where NNG() denotes the NNG regression model of the TG ratio and are the
NNG regression coefficients of the AG and G in the lsquouprsquo area of the soaking zone for
modeling TG in same area
With the model-based strategy and adaptive model of the TG ratio given in Section
42 the optimization adjustment results can be obtained as follows
The adjustment of gas flow in the lsquouprsquo area of the soaking zone shown in Fig 12
illustrates that the implementation of the model-based optimization operation can reduce
the consumption of the gas flow Compared with the original case 938 of the gas can
be saved on average by utilizing the model-based adjustment
0 02 04 06 08 1 12 14 16 18 2
x 104
-800
-700
-600
-500
-400
-300
-200
-100
0
Sampling intervals
Adj
ustm
ent o
f gas
flow
m(
3 h)
20
318
319
320
321
322
323
324
325
326
327
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
(a) Adjustment amount of gas flow
0 02 04 06 08 1 12 14 16 18 2
x 104
500
1000
1500
2000
2500
3000
3500
4000
4500
5000G
as fl
owm
(3 h
)
Sampling intervals
OriginalAdjusted
(b) Comparison of original and adjusted gas flow
095 096 097 098 099 1 101 102 103 104 105
x 104
500
1000
1500
2000
2500
3000
3500
4000
4500
5000
Gas
flow
m(
3 h)
Sampling intervals
OriginalAdjusted
(c) Comparison of original and adjusted gas flow in interval [9501-10500]
Fig 12 Adjustment of gas flow in the lsquouprsquo area of the soaking zone
The adjustment of the air flow in the lsquouprsquo area of the soaking zone given in Fig 13
shows that the air flow is reduced in most cases and only increased for a few cases In
21
328
329
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
general 681 of the air flow is reduced after the implementation of the model-based
optimization
0 02 04 06 08 1 12 14 16 18 2
x 104
-2000
-1500
-1000
-500
0
500
1000
Sampling intervals
Adj
ustm
ent o
f air
flow
(m3 h
)
(a) Adjustment amount of air flow
0 02 04 06 08 1 12 14 16 18 2
x 104
4000
5000
6000
7000
8000
9000
10000
Air
flow
m(
3 h)
Sampling intervals
OriginalAdjusted
(b) Comparison of original and adjusted air flow
22
330
331
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
095 096 097 098 099 1 101 102 103 104 105
x 104
4000
5000
6000
7000
8000
9000
10000
Air
flow
m(
3 h)
Sampling intervals
OriginalAdjusted
(c) Comparison of original and adjusted air flow in interval [9501-10500]
Fig 13 Adjustment of air flow in the lsquouprsquo area of the soaking zone
52 Discussions
In this section detailed analysis for the optimization operation results is provided
The statistics of optimization at the boundary conditions shown in Table 3 illustrates
that the adjusted value for G reaches its lower limit in most cases (8045 for
and 687 for ) which is the most energy-efficient point In a number of cases
(1079 for and 001 for ) the adjusted AG reaches its upper
limit These results indicate that the optimization operations maximize the combustion
efficiency by decreasing G and increasing the AG ratio up to the boundary conditions
However the lower limit is also unexpectedly attained in a small number of cases
(110 for and 078 for ) for which a more detailed analysis is
given below
23
332
333
334
335
336
337
338
339
340
341
342
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
Table 3 Statistics of optimization at boundary conditions
Total
Amount 0 1374 0 16090 17464
Percentag
e 0 687 0 8045
8732
Total
Amount 2 220 2158 156 2536
Percentag
e 001 110 1079 078
1268
Table 4 Statistics for different optimization operations
Adjustment
Gdarr
Guarr TotalAGuarr AGdarr
Tdarr Tuarr Tdarr Tuarr
Amount 9548 3744 6328 380 0 20000
Percentage 4774 1872 3164 190 0 100
The statistics for different optimization operations is performed and shown in Table
4 It can be seen that all the adjusted operations result in the reduction of gas
consumption In over half the cases (4774 with decreased temperature and 1872
with increased temperature) AG is adjusted to higher levels Meanwhile a number of
the adjustments (3164) lower the temperature by reducing the G and AG at the same
time This result indicates that the original temperature cannot be adjusted to the target
by only reducing G in the constraint conditions The remaining cases (19) are
24
343
344
345
346
347
348
349
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
relatively special in which an excess of air is supplied in the original operation and the
adjusted operation thereby increasing the temperature by a smaller AG with less gas
consumption
6 Conclusion
Based on the actual operation data this paper aims to explore the improvement of the
combustion efficiency and the room for energy conservation Correlation analysis and
LDA show that it is of great significance to introduce two derived ratio variables which
are the AG ratio and the TG ratio A type of combustion efficiency utilizing an on-line
soft sensor is put forward by employing a NNG variable selection algorithm which
provides a good solution to the combustion efficiency real-time measurement problem
of a reheating furnace The implementation of the model-based optimization is studied
based on the actual operational data Detailed analysis for the optimization results is
given for the different cases The results show that significant energy conservation can
be achieved when the furnace operation is optimized by using the developed soft sensor
model
Acknowledgement
The authors would like to thank the financial support provided by the National Nature
Science Foundation of China under Grant 61171145 Y Yao was supported by Ministry
of Science amp Technology ROC under Grant number MOST 104-2221-E-007-129
References
25
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[1] Z J Wang Q D Wu and T Y Chai Optimal-setting control for complicated
industrial processes and its application studyControl Engineering Practice vol
12 pp 65-74 2004
[2] A Steinboeck K Graichen and A Kugi Dynamic Optimization of a Slab
Reheating Furnace With Consistent Approximation of Control VariablesIEEE
Transactions on Control Systems Technology vol 19 pp 1444-1456 2011
[3] B T Zhang C Y Wang Q Qin and L Li Influence of Boiler Combustion
Adjustment on NOxEmission and Boiler EfficiencyAdvanced Materials
Research vol 732-733 pp 234-237 2013
[4] C K Yoo and IB Lee Soft Sensor and Adaptive Model-Based Dissolved
Oxygen Control for Biological Wastewater Treatment ProcessesEnvironmental
Engineering Science vol 21 pp 331-340 2004
[5] S A Bhat D N Saraf S Gupta and S K Gupta Use of Agitator Power as a
Soft Sensor for Bulk Free-Radical Polymerization of Methyl Methacrylate in
Batch ReactorsIndustrial amp Engineering Chemistry Research vol 45 pp 4243-
4255 2006
[6] K Desai Y Badhe S S Tambe and B D Kulkarni Soft-sensor development
for fed-batch bioreactors using support vector regressionBiochemical
Engineering Journal vol 27 pp 225-239 2006
[7] Y P Badhe Lonari J Tambe S S amp Kulkarni B D Improve polyethylene
process control and product qualityHydrocarbon Processing vol 86 pp 53-60
2007
26
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[8] N K Nath K Mandal A K Singh B Basu C Bhanu S Kumar et al Ladle
furnace on-line reckoner for prediction and control of steel temperature and
compositionIronmaking amp Steelmaking vol 33 pp 140-150 2006
[9] A J Yan T Y Chai F H Wu and P Wang Hybrid intelligent control of
combustion process for ore-roasting furnaceJournal of Control Theory and
Applications vol 6 pp 80-85 2008
[10] J Li W M Zhong H Cheng X D Kong and F Qian A data-driven soft
sensor modeling for furnace temperature of Opposed Multi-Burner gasifier pp
705-710 2011
[11] Y H Yang Y H Liu X Z Liu and S K Qin Billet temperature soft sensor
model of reheating furnace based on RVM method pp 4003-4006 2011
[12] J H Wang C Wang X F Zhu and X K Fang Application of soft sensor in
welding seam tracking prediction based on LSSVM and PSO with compression
factor pp 2441-2446 2013
[13] L Balbis J Balderud and M J Grimble Nonlinear predictive control of steel
slab reheating furnace pp 1679-1684 2008
[14] A Steinboeck D Wild T Kiefer and A Kugi A mathematical model of a slab
reheating furnace with radiative heat transfer and non-participating gaseous
media International Journal Of Heat And Mass Transfer vol 53 pp 5933-
5946 Dec 2010
[15] A Steinboeck D Wild and A Kugi Nonlinear model predictive control of a
continuous slab reheating furnace Control Engineering Practice vol 21 pp
495-508 2013
27
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[16] C Zhang T Ishii and S Sugiyama Numerical Modeling Of the Thermal
Performance Of Regenerative Slab Reheat Furnaces Numerical Heat Transfer
Part A Applications vol 32 pp 613-631 1997
[17] J G Kim and K Y Huh Prediction of Transient Slab Temperature Distribution
in the Re-heating Furnace of a Walking-beam Type for Rolling of Steel Slabs
ISIJ International vol 40 pp 1115-1123 2000
[18] J G Kim K Y and H I T K Three-Dimensional Analysis Of the Walking-
Beam-Type Slab Reheating Furnace In Hot Strip Mills Numerical Heat
Transfer Part A Applications vol 38 pp 589-609 2000
[19] CT Hsieh MJ Huang ST Lee and CH Wang Numerical Modeling of a
Walking-Beam-Type Slab Reheating Furnace Numerical Heat Transfer Part A
Applications vol 53 pp 966-981 2008
[20] MJ Huang CT Hsieh ST Lee and CH Wang A Coupled Numerical Study
of Slab Temperature and Gas Temperature in the Walking-Beam-Type Slab
Reheating Furnace Numerical Heat Transfer Part A Applications vol 54 pp
625-646 2008
[21] Z Song and A Kusiak Constraint-Based Control of Boiler Efficiency A Data-
Mining Approach IEEE Transactions on Industrial Informatics vol 3 pp 73-
83 2007
[22] J Q Li J J Gu and C L Niu The Operation Optimization based on
Correlation Analysis of Operation Parameters in Power Plant pp 138-141
2008
28
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[23] J G Wang S S Shieh S S Jang D S H Wong and C W Wu A two-tier
approach to the data-driven modeling on thermal efficiency of a BFGcoal co-
firing boiler Fuel vol 111 pp 528-534 Sep 2013
[24] L Breiman Better Subset Regression Using the Nonnegative Garrote
Technometrics vol 37 pp 373-384 1995
[25] R A Fisher The use of multiple measurements in taxonomic problems
AnnHum Genet vol 7 pp 179-188 1936
[26] J Liu Developing a soft sensor based on sparse partial least squares with
variable selection Journal of Process Control vol 24 pp 1046-1056 2014
[27] K Sun J Liu JL Kang SS Jang D SH Wong and DS Chen
Development of a variable selection method for soft sensor using artificial
neural network and nonnegative garrote Journal of Process Control vol 24 pp
1068-1075 2014
29
439
440
441
442
443
444
445
446
447
448
449
450
451
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
Obviously the research on the numerical model for the heating performance of
reheating furnace can be done based on basic combustion theory and heat transmission
characteristics Many scholars devote themselves to the simulations of the heat flow
phenomenon in the reheating furnace Zhang et al attempted to apply a computational
fluid dynamic (CFD) simulation to predict the combustion performance for a reheating
furnace by simplifying the furnace to a cuboid and assuming that the slab possesses
infinite length and enters the reheating furnace at a fixed speed [16] The CFD method
has been applied to the study of reaction turbulence radiation heat transmission and the
calculation for the steady state heat transmission rate of the slab under the given
temperature [17] In the works of references [16-20] the authors used the given
temperature data of slabs to compute the steady flow and temperature field However
owing to the changing operating conditions the actual implementation of these
numerical model methods still bristle with difficulties although the methods mentioned
above are feasible for the prediction Thus for an online application it is necessary to
adopt a real time data-driven model to resolve the time varying characteristics
Proper variable selection is an important step in model building for a large-scale
combustion system A well-trimmed variable dimension ensures the acquired model is
transparent comprehensible and robust Some studies reported that the combustion
model built by a selected subset of input variables provide more accurate predictions of
combustion efficiency than the entire set of variables [21-23] Recently shrinkage
methods which conduct variable selection by shrinking or setting some coefficients of a
ldquogreedyrdquo model to zero have received significant attention A popular form of these
methods is the non-negative garrote (NNG) [23 24]
5
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
Against this background this paper aims to propose a combustion efficiency index
for the reheating furnace and investigate for room in improvement regarding energy
conservation The primary contribution is a practical combustion efficiency index the
incorporation of the derived variables and soft-sensing method for the optimization of
combustion efficiency of reheating furnaces The derived variables are found more
physically meaningful than the plain variables when constructing the model of
combustion efficiency By employing a NNG variable selection procedure an adaptive
scheme for combustion efficiency modeling and adjustment is proposed and virtually
implemented for a rolling reheating furnace The results show that there is significant
room for energy conservation
The remainder of the paper is organized as follows In the next section the reheating
furnace and the data preprocessing is described In Section 3 the statistics analysis for
different variables and the formation of derived variables are presented In Section 4
the framework of an adaptive model based on NNG variable selection is presented and
two models developed for the temperature and temperature-gas (TG) ratio are
compared according to the model prediction precision A model-based optimization
scheme is provided and applied to the combustion efficiency improvement for an actual
case of a reheating furnace presented in Section 5 Several remarks and a summary
conclude the last section
2 Plant description and data preprocessing
The schematic of the heating process in the rolling mill reheating furnace is shown in
Fig 1 There are four zones in the reheating furnace including the preheating zone (P)
the first heating zone (1) the second heating zone (2) and the soaking zone (S) The
6
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
steel slab moves through the four zones in turn and is heated to the demanded state
using a specific temperature increase curve As is shown in Table 1 the soaking zone
has two areas that are defined as up and down and both of the areas possess the same
five variables including two manipulated variables the air flow (A) and the gas flow
(G) and three temperatures in left center and right sections of the area (T-l T-c and T-
r) The other zones have the same variables as the soaking zone hence there are 40
variables in total for the reheating furnace
Fig 1 The schematic of the heating process in the reheating furnace
Table 1 Variables and descriptions in the soaking zone
Variable Description unit
AS-u Air flow in the lsquouprsquo area Nm3h
AS-d Air flow in the lsquodownrsquo area Nm3h
GS-u Gas flow in the lsquouprsquo area Nm3h
GS-d Gas flow in the lsquodownrsquo area Nm3h
TS-ul Temperature in the left part of the lsquouprsquo area
TS-uc Temperature in the center part of the lsquouprsquo area
TS-ur Temperature in the right part of the lsquouprsquo area
7
143
144
145
146
147
148
149
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
TS-dl Temperature in the left part of the lsquodownrsquo area
TS-dc Temperature in the center part of the lsquodownrsquo area
TS-dr Temperature in the right part of the lsquodownrsquo area
A data set of 20000 samples was used in this study The samples were collected from
an actual reheating furnace in a large iron and steel plant located in Shanghai from
September 14 to September 27 2014 The operational data is taken on a per minute
basis
1
092
093
042
041
046
092
1
091
04
039
045
093
091
1
038
038
045
042
04
038
1
096
096
041
039
038
096
1
096
046
045
045
096
096
1
TS-ul
TS-uc
TS-ur
TS-dl
TS-dc
TS-dr
TS-ul TS-uc TS-ur TS-dl TS-dc TS-dr
04
05
06
07
08
09
1
Fig 2 Correlation between each temperature in the soaking zone
In order to investigate the relation among different temperatures in each zone
correlation analysis is conducted for the soaking zone as illustrated in Fig 2 It can be
seen that the temperature in three parts of the lsquouprsquo area is highly correlated with a
correlation coefficient greater than 09 The similar results exist for the lsquodownrsquo area as
well On the contrary the correlation coefficient between temperatures in one part of
8
150
151
152
153
154
155
156
157
158
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
lsquouprsquo area and any part of lsquodownrsquo area does not exceed 05 Therefore for the reduction
of the data dimension the three temperatures in the lsquouprsquo area or the lsquodownrsquo area can be
treated as only one variable which can be taken as the mean value or the first principal
component acquired from the PCA analysis Considering the reservation of the variable
physical meaning the former is preferred and used
3 Statistics analysis and incorporation of derived variables
In this section in order to uncover the physical knowledge for the actual operation
guidance and confirmation statistical analysis is performed for the 16 input manipulated
variables and the eight output variables (ie the temperatures in the four zones) of a
reheating furnace system For the combustion efficiency evaluation and modeling two
types of derived ratio variables are introduced which is helpful to reveal the
information included in the data
31 Correlation analysis
Correlation analysis between the temperatures (T) in each area and all of the air flows
(A) and the gas flows (G) is performed and shown in Fig 3 It can be seen that only the
temperatures in both areas of the soaking zone are the most highly related to the air flow
and the gas flow in its own zone However this phenomenon does not occur in the other
three zones The temperatures in the second heating zone mainly depend on the air flow
and the gas flow in its own zone as well as the nearby first heating zone As for the
preheating zone and first heating zone no apparent correlation can be observed
Obviously these analysis results could not tell us the explicit information about how to
evaluate the efficiency levels and key manipulated variables
9
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
-101
T P-u
AP GP A1 G1 A2 G2 AS GS
AP GP A1 G1 A2 G2 AS GS
AP GP A1 G1 A2 G2 AS GS
-101
T P-d
-101
T 1-u
-101
T 1-d
AP GP A1 G1 A2 G2 AS GS
-101
T 2-u
AP GP A1 G1 A2 G2 AS GS
-101
T 2-d
-101
T S-u
-101
T S-d
AP GP A1 G1 A2 G2 AS GS
AP GP A1 G1 A2 G2 AS GS
AP GP A1 G1 A2 G2 AS GS
UPDOWN
Fig 3 Correlation between air flows gas flows and temperatures
32 Incorporation of derived variables
During the stable heating stage the quantity of heat absorbed and removed from the
slabs from each furnace zone is relatively constant Hence the derived variable TG
ratio can be treated as an index for the combustion efficiency level This is because a
higher TG ratio signifies more combustion heat generated from unit gas ie higher
combustion efficiency
Moreover it is known that the appropriate air and fuel ratio is vital for the
combustion efficiency so the air-gas ratio (AG) is utilized as another derived variable
for the research Again the correlation analysis is performed for two types of derived
variables The correlations between different variables including the AG ratio G and
the TG ratio are shown in Fig 4 It can be clearly seen from the four red rectangle
blocks that TG of each zone is only remarkably related to AG and G in the same zone
10
183
184
185
186
187
188
189
190
191
192
193
194
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
Furthermore as related to the lsquouprsquo or lsquodownrsquo areas in one zone TG in each area has the
highest correlation with AG and G in the same area while AG and G in the opposite
area of the same zone is secondary This can be easily seen from the red and blue color
markings in each red rectangle block Thus it is of great significance to introduce these
derived ratio variables
-101
TG
P-u
AGP GP AG1G1 AG2 G2 AGS GS
-101
TG
P-d
AGP GP AG1 G1 AG2 G2 AGS GS
-101
TG
1-u
AGP GP AG1 G1 AG2 G2 AGS GS
-101
TG
1-d
AGP GP AG1 G1 AG2 G2 AGS GS
-101
TG
2-u
AGP GP AG1 G1 AG2 G2 AGS GS
-101
TG
2-d
AGP GP AG1 G1 AG2 G2 AGS GS
-101
TG
S-u
AGP GP AG1 G1 AG2 G2 AGS GS
-101
TG
S-d
AGP GP AG1 G1 AG2 G2 AGS GS
UPDOWN
Fig 4 Correlation between the air-gas ratio gas flow and TG ratio
33 LDA analysis
Linear discriminant analysis (LDA) aims to finding a projection direction that
maximizes the separation of class means and minimizes the within-class variance [25]
In this section LDA is utilized to identify the discriminating variables that play an
important role in determining combustion efficiency levels All the data are partitioned
into five groups according to their efficiency levels
11
195
196
197
198
199
200
201
202
203
204
205
206
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
Based on a descending order of TG ratios five efficiency levels for the lsquouprsquo area of
the soaking zone are denoted as HH H M L and LL LDA is conducted on the three
groups of the data with levels of HH M and LL
Fig 5 shows the scattering of the LDA projections of the process observations
collected at the three efficiency levels where y1 and y2 correspond to the first two LDA
components that contain most discriminant information The weighting factors ( and
) of the 16 input variables composing the projections y1 and y2 are shown in Fig 6
where and From left to right the 16 input variables are defined as
the eight AG ratio variables and the eight gas flow variables with the order of variables
of each kind P-u P-d 1-u 1-d 2-u 2-d S-u and S-d
-8 -6 -4 -2 0 2 4 6-4
-3
-2
-1
0
1
2
3
4
5
6
Projection y1
Pro
ject
ion
y2
HHMLL
Fig 5 Scattering of LDA projections y1 and y2 for three efficiency levels
12
207
208
209
210
211
212
213
214
215
216
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
AG_S-u-4
-2
0
2
Wei
ghtin
g fa
ctor
for y
1
AG_S-u
-101234
Wei
ghtin
g fa
ctor
for y
2
Fig 6 Weight factors of various input variables for projections y1 and y2
The two figures reveal that the three groups are clearly discriminated by the LDA
projection and the most significant variables for the different efficiency levels are the
derived variables ie the AG in the lsquouprsquo area of the soaking zone A similar result can
be obtained in the lsquodownrsquo area or any area of the other zones Therefore AG in each
heating area is the key manipulated variable that determines the different combustion
efficiency
4 Modeling and prediction of temperature and temperature-gas ratio
For the model-based operation optimization the models for the temperature and TG
ratio based on the NNG algorithm will be developed and compared
41 NNG variable selection algorithm
The NNG method can be generalized into a two-stage shrinkage method In the first
stage the sign for each variable is determined using the ordinary least square procedure
13
217
218
219
220
221
222
223
224
225
226
227
228
229
230
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
and in the second stage the corresponding magnitudes are computed by solving a series
of constrained quadratic programming
A set of observation data is provided where is the input matrix whose
columns represent the measured candidate variables and is the corresponding
vector of the response data The following expression is given with the number of the
response variable being equal to 1 but a similar procedure can be generalized to any
number of variables Let and be normalized to the zero-mean and unit standard
deviation Additionally let be a set of the ordinary least square estimates of the
coefficients of the following linear model then
(1)
The second stage shrinkage can be formulated as the following optimization problem
subject to
(2)
As decreased and the NNG is tightened more of the become zero and the
remaining nonzero coefficients are shrunk A solution path exists with on which
the appropriate shrinkage can be selected Conventionally the v-fold cross-validation is
used to estimate the prediction error and to select the best solution in the solution path
so as to minimize the prediction or model error
42 Modeling and prediction of temperature and TG ratio
The input-output relations change with time as the reheating process develops An
adaptive modeling strategy is often used to resolve time-varying characteristics of
14
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
industrial processes In this paper the NNG-based regression modeling prediction and
optimization is implemented in a moving window manner where the size of the
window and the length of the moving step are selected as 1000 and 100 respectively In
addition the prediction horizon is also set as 100 In each step the NNG regression
model is built based on the data in the current window which is then used for the
prediction and optimization in the subsequent prediction horizon Next the window
moves forward by replacing the oldest 100 samples for model training by the
observations collected in the previous prediction horizon This moving window strategy
is workable because in the reheating process the input-output relation is slowly time-
varying and the model is still valid for the prediction and optimization in the subsequent
short time
The modeling of the temperature T is based on the air flow rates and gas flow rates
while the modeling for the TG ratio is based on the AG ratios and gas flow rates
Taking the lsquouprsquo area of the soaking zone as an example the prediction result for T and
the TG ratio in the lsquouprsquo area of the soaking zone is shown in Figs 7 and 8 respectively
0 02 04 06 08 1 12 14 16 18 2
x 104
1140
1160
1180
1200
1220
1240
1260
1280
1300
1320
Sampling intervals
T S-u (
)
OriginalPredicted
Fig 7 Prediction of temperature in the lsquouprsquo area of the soaking zone
15
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
0 02 04 06 08 1 12 14 16 18 2
x 104
02
04
06
08
1
12
14
16
18
2
Sampling intervals
TG
S-u
OriginalPredicted
Fig 8 Prediction of TG ratio in the lsquouprsquo area of the soaking zone
A comparison between Figs 7 and 8 show that the prediction of the TG ratio is more
accurate than the prediction of temperature The average R2 in 190 NNG regressions is
0935 for the TG ratio while the average R2 is only 0814 for temperature This is
consistent with the statistical analysis which reveals that the correlation between AG
and TG is higher than that between A and T These results indicate that derived
variables are more meaningful for the purpose of prediction and the modeling of the
TG ratio is more appropriate for implementing optimization
The selected frequency for each variable in 190 NNG regressions and the coefficients
of each variable in 20 NNG regressions for the TG ratio modeling of the lsquouprsquo area of
the soaking zone are shown in Figs 9 and 10 respectively Fig 9 shows that the selected
frequency of the variables in the lsquouprsquo area of the soaking zone is much higher than
variables in other zones Similarly as is shown in Fig 10 the NNG regression
coefficients of the variables in the lsquouprsquo area of the soaking zone are much larger than the
variables in the other areas These results indicated that variables in the lsquouprsquo area of
16
269
270
271
272
273
274
275
276
277
278
279
280
281
282
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
soaking zone are mostly contributed to the modeling of the TG ratio in same area
which is quite reasonable Similar results can be obtained for the other areas
0
20
40
60
80
100
120
140
160
180
200
Sel
ecte
d fre
quen
cy
AGP-u AGP-d AG1-u AG1-d AG2-u AG2-d AGS-u AGS-d GP-u GP-d G1-u G1-d G2-u G2-d GS-u GS-d
Fig 9 Variable selected frequency of over 190 runs for the TG ratio model of the lsquouprsquo
area of the soaking zone
0 2 4 6 8 10 12 14 16 18 20-05
05
AG
P-u
0 2 4 6 8 10 12 14 16 18 20-05
05
GP
-u
0 2 4 6 8 10 12 14 16 18 20-1
0
AG
P-d
0 2 4 6 8 10 12 14 16 18 20-1
0
GP
-d
0 2 4 6 8 10 12 14 16 18 20-05
05
AG
1-u
0 2 4 6 8 10 12 14 16 18 20-05
05
G1-
u
0 2 4 6 8 10 12 14 16 18 20-05
05
AG
1-d
0 2 4 6 8 10 12 14 16 18 20-05
05
G1-
d
0 2 4 6 8 10 12 14 16 18 20-1
0
AG
2-u
0 2 4 6 8 10 12 14 16 18 200
1
G2-
u
0 2 4 6 8 10 12 14 16 18 20-1
0
AG
2-d
0 2 4 6 8 10 12 14 16 18 20-1
0
G2-
d
0 2 4 6 8 10 12 14 16 18 200
1
AG
S-u
0 2 4 6 8 10 12 14 16 18 20-1
0
GS
-u
0 2 4 6 8 10 12 14 16 18 20-05
05
AG
S-d
0 2 4 6 8 10 12 14 16 18 20-05
05
GS
-d
Fig 10 Part of the NNG regression coefficients of each variable over 190 runs for the
TG ratio model of the lsquouprsquo area of the soaking zone
It is remarkable that the NNG regression coefficients corresponding to the AG ratio
in the lsquouprsquo area of the soaking zone are consistently positive On the contrary the
coefficients of the gas flow rate in the same area are consistently negative This
indicates that under this condition the rise of the AG leads to the increase of the TG
17
283
284
285
286
287
288
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
while the rise of the G leads to the drop of the TG This property is helpful for
performing optimization and improve combustion efficiency
For comparison the modeling results for two other algorithms artificial neural
network (ANN) and partial least squares (PLS) [26] are compared with the NNG
algorithm in the following three aspects [27]
(1) Model Size the number of variables selected for modeling
(2) Model Magnitude the mean of the L1 norm of the regression coefficients
(3) Prediction Precision the mean squared prediction error (MSPE)
Summary of the algorithm comparison is shown in Table 2 It can be seen that the
superiorities of the NNG regression in model size model magnitude and model
precision are remarkable
Table 2 Summary of algorithm comparison
Index NNG ANN PLS
Model Size 849 16 16
Model Magnitude 10075 -- 12150
MSPE 00093 00108 00140
5 Model-based optimization
51 Implemention of model-based optimization operation
The goal of optimization is to seek an optimal combination of AG and G in the lsquouprsquo
area of the soaking zone so as to minimize the gas consumption while keeping the
temperature at the target value According to the heating schedule the most expected
temperature in the soaking zone is 1200 Therefore the target temperature
is set at this value to achieve the desired heating effect As shown in Fig 11 in the real
18
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
operation the temperature fluctuates around the target value because of the imperfect
control performance However in order to simplify the analysis and compute the
maximum possible energy saving perfect temperature control is assumed when
adopting the model-based optimization strategy In other words it is assumed that the
actual temperature in the lsquouprsquo area of the soaking zone can be adjusted to the expected
temperature ie 1200
0 02 04 06 08 1 12 14 16 18 2
x 104
1150
1200
1250
1300
Sampling intervals
T S-u (
)
OriginalTarget
Fig 11 Original and target temperature in the lsquouprsquo area of the soaking zone
The adjustment scheme takes the maximum value and minimum value of the original
operation data as the upper and lower bounds for the adjustment Moreover in order to
assure the validity of the linear model the increment or decrement of AG and G should
not beyond of the original value (considered as 10 for the purposes of this paper)
The adjustment strategy can be formulated as
19
307
308
309
310
311
312
313
314
315
316
317
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
(3)
where NNG() denotes the NNG regression model of the TG ratio and are the
NNG regression coefficients of the AG and G in the lsquouprsquo area of the soaking zone for
modeling TG in same area
With the model-based strategy and adaptive model of the TG ratio given in Section
42 the optimization adjustment results can be obtained as follows
The adjustment of gas flow in the lsquouprsquo area of the soaking zone shown in Fig 12
illustrates that the implementation of the model-based optimization operation can reduce
the consumption of the gas flow Compared with the original case 938 of the gas can
be saved on average by utilizing the model-based adjustment
0 02 04 06 08 1 12 14 16 18 2
x 104
-800
-700
-600
-500
-400
-300
-200
-100
0
Sampling intervals
Adj
ustm
ent o
f gas
flow
m(
3 h)
20
318
319
320
321
322
323
324
325
326
327
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
(a) Adjustment amount of gas flow
0 02 04 06 08 1 12 14 16 18 2
x 104
500
1000
1500
2000
2500
3000
3500
4000
4500
5000G
as fl
owm
(3 h
)
Sampling intervals
OriginalAdjusted
(b) Comparison of original and adjusted gas flow
095 096 097 098 099 1 101 102 103 104 105
x 104
500
1000
1500
2000
2500
3000
3500
4000
4500
5000
Gas
flow
m(
3 h)
Sampling intervals
OriginalAdjusted
(c) Comparison of original and adjusted gas flow in interval [9501-10500]
Fig 12 Adjustment of gas flow in the lsquouprsquo area of the soaking zone
The adjustment of the air flow in the lsquouprsquo area of the soaking zone given in Fig 13
shows that the air flow is reduced in most cases and only increased for a few cases In
21
328
329
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
general 681 of the air flow is reduced after the implementation of the model-based
optimization
0 02 04 06 08 1 12 14 16 18 2
x 104
-2000
-1500
-1000
-500
0
500
1000
Sampling intervals
Adj
ustm
ent o
f air
flow
(m3 h
)
(a) Adjustment amount of air flow
0 02 04 06 08 1 12 14 16 18 2
x 104
4000
5000
6000
7000
8000
9000
10000
Air
flow
m(
3 h)
Sampling intervals
OriginalAdjusted
(b) Comparison of original and adjusted air flow
22
330
331
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
095 096 097 098 099 1 101 102 103 104 105
x 104
4000
5000
6000
7000
8000
9000
10000
Air
flow
m(
3 h)
Sampling intervals
OriginalAdjusted
(c) Comparison of original and adjusted air flow in interval [9501-10500]
Fig 13 Adjustment of air flow in the lsquouprsquo area of the soaking zone
52 Discussions
In this section detailed analysis for the optimization operation results is provided
The statistics of optimization at the boundary conditions shown in Table 3 illustrates
that the adjusted value for G reaches its lower limit in most cases (8045 for
and 687 for ) which is the most energy-efficient point In a number of cases
(1079 for and 001 for ) the adjusted AG reaches its upper
limit These results indicate that the optimization operations maximize the combustion
efficiency by decreasing G and increasing the AG ratio up to the boundary conditions
However the lower limit is also unexpectedly attained in a small number of cases
(110 for and 078 for ) for which a more detailed analysis is
given below
23
332
333
334
335
336
337
338
339
340
341
342
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
Table 3 Statistics of optimization at boundary conditions
Total
Amount 0 1374 0 16090 17464
Percentag
e 0 687 0 8045
8732
Total
Amount 2 220 2158 156 2536
Percentag
e 001 110 1079 078
1268
Table 4 Statistics for different optimization operations
Adjustment
Gdarr
Guarr TotalAGuarr AGdarr
Tdarr Tuarr Tdarr Tuarr
Amount 9548 3744 6328 380 0 20000
Percentage 4774 1872 3164 190 0 100
The statistics for different optimization operations is performed and shown in Table
4 It can be seen that all the adjusted operations result in the reduction of gas
consumption In over half the cases (4774 with decreased temperature and 1872
with increased temperature) AG is adjusted to higher levels Meanwhile a number of
the adjustments (3164) lower the temperature by reducing the G and AG at the same
time This result indicates that the original temperature cannot be adjusted to the target
by only reducing G in the constraint conditions The remaining cases (19) are
24
343
344
345
346
347
348
349
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
relatively special in which an excess of air is supplied in the original operation and the
adjusted operation thereby increasing the temperature by a smaller AG with less gas
consumption
6 Conclusion
Based on the actual operation data this paper aims to explore the improvement of the
combustion efficiency and the room for energy conservation Correlation analysis and
LDA show that it is of great significance to introduce two derived ratio variables which
are the AG ratio and the TG ratio A type of combustion efficiency utilizing an on-line
soft sensor is put forward by employing a NNG variable selection algorithm which
provides a good solution to the combustion efficiency real-time measurement problem
of a reheating furnace The implementation of the model-based optimization is studied
based on the actual operational data Detailed analysis for the optimization results is
given for the different cases The results show that significant energy conservation can
be achieved when the furnace operation is optimized by using the developed soft sensor
model
Acknowledgement
The authors would like to thank the financial support provided by the National Nature
Science Foundation of China under Grant 61171145 Y Yao was supported by Ministry
of Science amp Technology ROC under Grant number MOST 104-2221-E-007-129
References
25
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[1] Z J Wang Q D Wu and T Y Chai Optimal-setting control for complicated
industrial processes and its application studyControl Engineering Practice vol
12 pp 65-74 2004
[2] A Steinboeck K Graichen and A Kugi Dynamic Optimization of a Slab
Reheating Furnace With Consistent Approximation of Control VariablesIEEE
Transactions on Control Systems Technology vol 19 pp 1444-1456 2011
[3] B T Zhang C Y Wang Q Qin and L Li Influence of Boiler Combustion
Adjustment on NOxEmission and Boiler EfficiencyAdvanced Materials
Research vol 732-733 pp 234-237 2013
[4] C K Yoo and IB Lee Soft Sensor and Adaptive Model-Based Dissolved
Oxygen Control for Biological Wastewater Treatment ProcessesEnvironmental
Engineering Science vol 21 pp 331-340 2004
[5] S A Bhat D N Saraf S Gupta and S K Gupta Use of Agitator Power as a
Soft Sensor for Bulk Free-Radical Polymerization of Methyl Methacrylate in
Batch ReactorsIndustrial amp Engineering Chemistry Research vol 45 pp 4243-
4255 2006
[6] K Desai Y Badhe S S Tambe and B D Kulkarni Soft-sensor development
for fed-batch bioreactors using support vector regressionBiochemical
Engineering Journal vol 27 pp 225-239 2006
[7] Y P Badhe Lonari J Tambe S S amp Kulkarni B D Improve polyethylene
process control and product qualityHydrocarbon Processing vol 86 pp 53-60
2007
26
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[8] N K Nath K Mandal A K Singh B Basu C Bhanu S Kumar et al Ladle
furnace on-line reckoner for prediction and control of steel temperature and
compositionIronmaking amp Steelmaking vol 33 pp 140-150 2006
[9] A J Yan T Y Chai F H Wu and P Wang Hybrid intelligent control of
combustion process for ore-roasting furnaceJournal of Control Theory and
Applications vol 6 pp 80-85 2008
[10] J Li W M Zhong H Cheng X D Kong and F Qian A data-driven soft
sensor modeling for furnace temperature of Opposed Multi-Burner gasifier pp
705-710 2011
[11] Y H Yang Y H Liu X Z Liu and S K Qin Billet temperature soft sensor
model of reheating furnace based on RVM method pp 4003-4006 2011
[12] J H Wang C Wang X F Zhu and X K Fang Application of soft sensor in
welding seam tracking prediction based on LSSVM and PSO with compression
factor pp 2441-2446 2013
[13] L Balbis J Balderud and M J Grimble Nonlinear predictive control of steel
slab reheating furnace pp 1679-1684 2008
[14] A Steinboeck D Wild T Kiefer and A Kugi A mathematical model of a slab
reheating furnace with radiative heat transfer and non-participating gaseous
media International Journal Of Heat And Mass Transfer vol 53 pp 5933-
5946 Dec 2010
[15] A Steinboeck D Wild and A Kugi Nonlinear model predictive control of a
continuous slab reheating furnace Control Engineering Practice vol 21 pp
495-508 2013
27
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[16] C Zhang T Ishii and S Sugiyama Numerical Modeling Of the Thermal
Performance Of Regenerative Slab Reheat Furnaces Numerical Heat Transfer
Part A Applications vol 32 pp 613-631 1997
[17] J G Kim and K Y Huh Prediction of Transient Slab Temperature Distribution
in the Re-heating Furnace of a Walking-beam Type for Rolling of Steel Slabs
ISIJ International vol 40 pp 1115-1123 2000
[18] J G Kim K Y and H I T K Three-Dimensional Analysis Of the Walking-
Beam-Type Slab Reheating Furnace In Hot Strip Mills Numerical Heat
Transfer Part A Applications vol 38 pp 589-609 2000
[19] CT Hsieh MJ Huang ST Lee and CH Wang Numerical Modeling of a
Walking-Beam-Type Slab Reheating Furnace Numerical Heat Transfer Part A
Applications vol 53 pp 966-981 2008
[20] MJ Huang CT Hsieh ST Lee and CH Wang A Coupled Numerical Study
of Slab Temperature and Gas Temperature in the Walking-Beam-Type Slab
Reheating Furnace Numerical Heat Transfer Part A Applications vol 54 pp
625-646 2008
[21] Z Song and A Kusiak Constraint-Based Control of Boiler Efficiency A Data-
Mining Approach IEEE Transactions on Industrial Informatics vol 3 pp 73-
83 2007
[22] J Q Li J J Gu and C L Niu The Operation Optimization based on
Correlation Analysis of Operation Parameters in Power Plant pp 138-141
2008
28
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[23] J G Wang S S Shieh S S Jang D S H Wong and C W Wu A two-tier
approach to the data-driven modeling on thermal efficiency of a BFGcoal co-
firing boiler Fuel vol 111 pp 528-534 Sep 2013
[24] L Breiman Better Subset Regression Using the Nonnegative Garrote
Technometrics vol 37 pp 373-384 1995
[25] R A Fisher The use of multiple measurements in taxonomic problems
AnnHum Genet vol 7 pp 179-188 1936
[26] J Liu Developing a soft sensor based on sparse partial least squares with
variable selection Journal of Process Control vol 24 pp 1046-1056 2014
[27] K Sun J Liu JL Kang SS Jang D SH Wong and DS Chen
Development of a variable selection method for soft sensor using artificial
neural network and nonnegative garrote Journal of Process Control vol 24 pp
1068-1075 2014
29
439
440
441
442
443
444
445
446
447
448
449
450
451
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
Against this background this paper aims to propose a combustion efficiency index
for the reheating furnace and investigate for room in improvement regarding energy
conservation The primary contribution is a practical combustion efficiency index the
incorporation of the derived variables and soft-sensing method for the optimization of
combustion efficiency of reheating furnaces The derived variables are found more
physically meaningful than the plain variables when constructing the model of
combustion efficiency By employing a NNG variable selection procedure an adaptive
scheme for combustion efficiency modeling and adjustment is proposed and virtually
implemented for a rolling reheating furnace The results show that there is significant
room for energy conservation
The remainder of the paper is organized as follows In the next section the reheating
furnace and the data preprocessing is described In Section 3 the statistics analysis for
different variables and the formation of derived variables are presented In Section 4
the framework of an adaptive model based on NNG variable selection is presented and
two models developed for the temperature and temperature-gas (TG) ratio are
compared according to the model prediction precision A model-based optimization
scheme is provided and applied to the combustion efficiency improvement for an actual
case of a reheating furnace presented in Section 5 Several remarks and a summary
conclude the last section
2 Plant description and data preprocessing
The schematic of the heating process in the rolling mill reheating furnace is shown in
Fig 1 There are four zones in the reheating furnace including the preheating zone (P)
the first heating zone (1) the second heating zone (2) and the soaking zone (S) The
6
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
steel slab moves through the four zones in turn and is heated to the demanded state
using a specific temperature increase curve As is shown in Table 1 the soaking zone
has two areas that are defined as up and down and both of the areas possess the same
five variables including two manipulated variables the air flow (A) and the gas flow
(G) and three temperatures in left center and right sections of the area (T-l T-c and T-
r) The other zones have the same variables as the soaking zone hence there are 40
variables in total for the reheating furnace
Fig 1 The schematic of the heating process in the reheating furnace
Table 1 Variables and descriptions in the soaking zone
Variable Description unit
AS-u Air flow in the lsquouprsquo area Nm3h
AS-d Air flow in the lsquodownrsquo area Nm3h
GS-u Gas flow in the lsquouprsquo area Nm3h
GS-d Gas flow in the lsquodownrsquo area Nm3h
TS-ul Temperature in the left part of the lsquouprsquo area
TS-uc Temperature in the center part of the lsquouprsquo area
TS-ur Temperature in the right part of the lsquouprsquo area
7
143
144
145
146
147
148
149
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
TS-dl Temperature in the left part of the lsquodownrsquo area
TS-dc Temperature in the center part of the lsquodownrsquo area
TS-dr Temperature in the right part of the lsquodownrsquo area
A data set of 20000 samples was used in this study The samples were collected from
an actual reheating furnace in a large iron and steel plant located in Shanghai from
September 14 to September 27 2014 The operational data is taken on a per minute
basis
1
092
093
042
041
046
092
1
091
04
039
045
093
091
1
038
038
045
042
04
038
1
096
096
041
039
038
096
1
096
046
045
045
096
096
1
TS-ul
TS-uc
TS-ur
TS-dl
TS-dc
TS-dr
TS-ul TS-uc TS-ur TS-dl TS-dc TS-dr
04
05
06
07
08
09
1
Fig 2 Correlation between each temperature in the soaking zone
In order to investigate the relation among different temperatures in each zone
correlation analysis is conducted for the soaking zone as illustrated in Fig 2 It can be
seen that the temperature in three parts of the lsquouprsquo area is highly correlated with a
correlation coefficient greater than 09 The similar results exist for the lsquodownrsquo area as
well On the contrary the correlation coefficient between temperatures in one part of
8
150
151
152
153
154
155
156
157
158
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
lsquouprsquo area and any part of lsquodownrsquo area does not exceed 05 Therefore for the reduction
of the data dimension the three temperatures in the lsquouprsquo area or the lsquodownrsquo area can be
treated as only one variable which can be taken as the mean value or the first principal
component acquired from the PCA analysis Considering the reservation of the variable
physical meaning the former is preferred and used
3 Statistics analysis and incorporation of derived variables
In this section in order to uncover the physical knowledge for the actual operation
guidance and confirmation statistical analysis is performed for the 16 input manipulated
variables and the eight output variables (ie the temperatures in the four zones) of a
reheating furnace system For the combustion efficiency evaluation and modeling two
types of derived ratio variables are introduced which is helpful to reveal the
information included in the data
31 Correlation analysis
Correlation analysis between the temperatures (T) in each area and all of the air flows
(A) and the gas flows (G) is performed and shown in Fig 3 It can be seen that only the
temperatures in both areas of the soaking zone are the most highly related to the air flow
and the gas flow in its own zone However this phenomenon does not occur in the other
three zones The temperatures in the second heating zone mainly depend on the air flow
and the gas flow in its own zone as well as the nearby first heating zone As for the
preheating zone and first heating zone no apparent correlation can be observed
Obviously these analysis results could not tell us the explicit information about how to
evaluate the efficiency levels and key manipulated variables
9
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
-101
T P-u
AP GP A1 G1 A2 G2 AS GS
AP GP A1 G1 A2 G2 AS GS
AP GP A1 G1 A2 G2 AS GS
-101
T P-d
-101
T 1-u
-101
T 1-d
AP GP A1 G1 A2 G2 AS GS
-101
T 2-u
AP GP A1 G1 A2 G2 AS GS
-101
T 2-d
-101
T S-u
-101
T S-d
AP GP A1 G1 A2 G2 AS GS
AP GP A1 G1 A2 G2 AS GS
AP GP A1 G1 A2 G2 AS GS
UPDOWN
Fig 3 Correlation between air flows gas flows and temperatures
32 Incorporation of derived variables
During the stable heating stage the quantity of heat absorbed and removed from the
slabs from each furnace zone is relatively constant Hence the derived variable TG
ratio can be treated as an index for the combustion efficiency level This is because a
higher TG ratio signifies more combustion heat generated from unit gas ie higher
combustion efficiency
Moreover it is known that the appropriate air and fuel ratio is vital for the
combustion efficiency so the air-gas ratio (AG) is utilized as another derived variable
for the research Again the correlation analysis is performed for two types of derived
variables The correlations between different variables including the AG ratio G and
the TG ratio are shown in Fig 4 It can be clearly seen from the four red rectangle
blocks that TG of each zone is only remarkably related to AG and G in the same zone
10
183
184
185
186
187
188
189
190
191
192
193
194
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
Furthermore as related to the lsquouprsquo or lsquodownrsquo areas in one zone TG in each area has the
highest correlation with AG and G in the same area while AG and G in the opposite
area of the same zone is secondary This can be easily seen from the red and blue color
markings in each red rectangle block Thus it is of great significance to introduce these
derived ratio variables
-101
TG
P-u
AGP GP AG1G1 AG2 G2 AGS GS
-101
TG
P-d
AGP GP AG1 G1 AG2 G2 AGS GS
-101
TG
1-u
AGP GP AG1 G1 AG2 G2 AGS GS
-101
TG
1-d
AGP GP AG1 G1 AG2 G2 AGS GS
-101
TG
2-u
AGP GP AG1 G1 AG2 G2 AGS GS
-101
TG
2-d
AGP GP AG1 G1 AG2 G2 AGS GS
-101
TG
S-u
AGP GP AG1 G1 AG2 G2 AGS GS
-101
TG
S-d
AGP GP AG1 G1 AG2 G2 AGS GS
UPDOWN
Fig 4 Correlation between the air-gas ratio gas flow and TG ratio
33 LDA analysis
Linear discriminant analysis (LDA) aims to finding a projection direction that
maximizes the separation of class means and minimizes the within-class variance [25]
In this section LDA is utilized to identify the discriminating variables that play an
important role in determining combustion efficiency levels All the data are partitioned
into five groups according to their efficiency levels
11
195
196
197
198
199
200
201
202
203
204
205
206
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
Based on a descending order of TG ratios five efficiency levels for the lsquouprsquo area of
the soaking zone are denoted as HH H M L and LL LDA is conducted on the three
groups of the data with levels of HH M and LL
Fig 5 shows the scattering of the LDA projections of the process observations
collected at the three efficiency levels where y1 and y2 correspond to the first two LDA
components that contain most discriminant information The weighting factors ( and
) of the 16 input variables composing the projections y1 and y2 are shown in Fig 6
where and From left to right the 16 input variables are defined as
the eight AG ratio variables and the eight gas flow variables with the order of variables
of each kind P-u P-d 1-u 1-d 2-u 2-d S-u and S-d
-8 -6 -4 -2 0 2 4 6-4
-3
-2
-1
0
1
2
3
4
5
6
Projection y1
Pro
ject
ion
y2
HHMLL
Fig 5 Scattering of LDA projections y1 and y2 for three efficiency levels
12
207
208
209
210
211
212
213
214
215
216
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
AG_S-u-4
-2
0
2
Wei
ghtin
g fa
ctor
for y
1
AG_S-u
-101234
Wei
ghtin
g fa
ctor
for y
2
Fig 6 Weight factors of various input variables for projections y1 and y2
The two figures reveal that the three groups are clearly discriminated by the LDA
projection and the most significant variables for the different efficiency levels are the
derived variables ie the AG in the lsquouprsquo area of the soaking zone A similar result can
be obtained in the lsquodownrsquo area or any area of the other zones Therefore AG in each
heating area is the key manipulated variable that determines the different combustion
efficiency
4 Modeling and prediction of temperature and temperature-gas ratio
For the model-based operation optimization the models for the temperature and TG
ratio based on the NNG algorithm will be developed and compared
41 NNG variable selection algorithm
The NNG method can be generalized into a two-stage shrinkage method In the first
stage the sign for each variable is determined using the ordinary least square procedure
13
217
218
219
220
221
222
223
224
225
226
227
228
229
230
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
and in the second stage the corresponding magnitudes are computed by solving a series
of constrained quadratic programming
A set of observation data is provided where is the input matrix whose
columns represent the measured candidate variables and is the corresponding
vector of the response data The following expression is given with the number of the
response variable being equal to 1 but a similar procedure can be generalized to any
number of variables Let and be normalized to the zero-mean and unit standard
deviation Additionally let be a set of the ordinary least square estimates of the
coefficients of the following linear model then
(1)
The second stage shrinkage can be formulated as the following optimization problem
subject to
(2)
As decreased and the NNG is tightened more of the become zero and the
remaining nonzero coefficients are shrunk A solution path exists with on which
the appropriate shrinkage can be selected Conventionally the v-fold cross-validation is
used to estimate the prediction error and to select the best solution in the solution path
so as to minimize the prediction or model error
42 Modeling and prediction of temperature and TG ratio
The input-output relations change with time as the reheating process develops An
adaptive modeling strategy is often used to resolve time-varying characteristics of
14
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
industrial processes In this paper the NNG-based regression modeling prediction and
optimization is implemented in a moving window manner where the size of the
window and the length of the moving step are selected as 1000 and 100 respectively In
addition the prediction horizon is also set as 100 In each step the NNG regression
model is built based on the data in the current window which is then used for the
prediction and optimization in the subsequent prediction horizon Next the window
moves forward by replacing the oldest 100 samples for model training by the
observations collected in the previous prediction horizon This moving window strategy
is workable because in the reheating process the input-output relation is slowly time-
varying and the model is still valid for the prediction and optimization in the subsequent
short time
The modeling of the temperature T is based on the air flow rates and gas flow rates
while the modeling for the TG ratio is based on the AG ratios and gas flow rates
Taking the lsquouprsquo area of the soaking zone as an example the prediction result for T and
the TG ratio in the lsquouprsquo area of the soaking zone is shown in Figs 7 and 8 respectively
0 02 04 06 08 1 12 14 16 18 2
x 104
1140
1160
1180
1200
1220
1240
1260
1280
1300
1320
Sampling intervals
T S-u (
)
OriginalPredicted
Fig 7 Prediction of temperature in the lsquouprsquo area of the soaking zone
15
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
0 02 04 06 08 1 12 14 16 18 2
x 104
02
04
06
08
1
12
14
16
18
2
Sampling intervals
TG
S-u
OriginalPredicted
Fig 8 Prediction of TG ratio in the lsquouprsquo area of the soaking zone
A comparison between Figs 7 and 8 show that the prediction of the TG ratio is more
accurate than the prediction of temperature The average R2 in 190 NNG regressions is
0935 for the TG ratio while the average R2 is only 0814 for temperature This is
consistent with the statistical analysis which reveals that the correlation between AG
and TG is higher than that between A and T These results indicate that derived
variables are more meaningful for the purpose of prediction and the modeling of the
TG ratio is more appropriate for implementing optimization
The selected frequency for each variable in 190 NNG regressions and the coefficients
of each variable in 20 NNG regressions for the TG ratio modeling of the lsquouprsquo area of
the soaking zone are shown in Figs 9 and 10 respectively Fig 9 shows that the selected
frequency of the variables in the lsquouprsquo area of the soaking zone is much higher than
variables in other zones Similarly as is shown in Fig 10 the NNG regression
coefficients of the variables in the lsquouprsquo area of the soaking zone are much larger than the
variables in the other areas These results indicated that variables in the lsquouprsquo area of
16
269
270
271
272
273
274
275
276
277
278
279
280
281
282
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
soaking zone are mostly contributed to the modeling of the TG ratio in same area
which is quite reasonable Similar results can be obtained for the other areas
0
20
40
60
80
100
120
140
160
180
200
Sel
ecte
d fre
quen
cy
AGP-u AGP-d AG1-u AG1-d AG2-u AG2-d AGS-u AGS-d GP-u GP-d G1-u G1-d G2-u G2-d GS-u GS-d
Fig 9 Variable selected frequency of over 190 runs for the TG ratio model of the lsquouprsquo
area of the soaking zone
0 2 4 6 8 10 12 14 16 18 20-05
05
AG
P-u
0 2 4 6 8 10 12 14 16 18 20-05
05
GP
-u
0 2 4 6 8 10 12 14 16 18 20-1
0
AG
P-d
0 2 4 6 8 10 12 14 16 18 20-1
0
GP
-d
0 2 4 6 8 10 12 14 16 18 20-05
05
AG
1-u
0 2 4 6 8 10 12 14 16 18 20-05
05
G1-
u
0 2 4 6 8 10 12 14 16 18 20-05
05
AG
1-d
0 2 4 6 8 10 12 14 16 18 20-05
05
G1-
d
0 2 4 6 8 10 12 14 16 18 20-1
0
AG
2-u
0 2 4 6 8 10 12 14 16 18 200
1
G2-
u
0 2 4 6 8 10 12 14 16 18 20-1
0
AG
2-d
0 2 4 6 8 10 12 14 16 18 20-1
0
G2-
d
0 2 4 6 8 10 12 14 16 18 200
1
AG
S-u
0 2 4 6 8 10 12 14 16 18 20-1
0
GS
-u
0 2 4 6 8 10 12 14 16 18 20-05
05
AG
S-d
0 2 4 6 8 10 12 14 16 18 20-05
05
GS
-d
Fig 10 Part of the NNG regression coefficients of each variable over 190 runs for the
TG ratio model of the lsquouprsquo area of the soaking zone
It is remarkable that the NNG regression coefficients corresponding to the AG ratio
in the lsquouprsquo area of the soaking zone are consistently positive On the contrary the
coefficients of the gas flow rate in the same area are consistently negative This
indicates that under this condition the rise of the AG leads to the increase of the TG
17
283
284
285
286
287
288
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
while the rise of the G leads to the drop of the TG This property is helpful for
performing optimization and improve combustion efficiency
For comparison the modeling results for two other algorithms artificial neural
network (ANN) and partial least squares (PLS) [26] are compared with the NNG
algorithm in the following three aspects [27]
(1) Model Size the number of variables selected for modeling
(2) Model Magnitude the mean of the L1 norm of the regression coefficients
(3) Prediction Precision the mean squared prediction error (MSPE)
Summary of the algorithm comparison is shown in Table 2 It can be seen that the
superiorities of the NNG regression in model size model magnitude and model
precision are remarkable
Table 2 Summary of algorithm comparison
Index NNG ANN PLS
Model Size 849 16 16
Model Magnitude 10075 -- 12150
MSPE 00093 00108 00140
5 Model-based optimization
51 Implemention of model-based optimization operation
The goal of optimization is to seek an optimal combination of AG and G in the lsquouprsquo
area of the soaking zone so as to minimize the gas consumption while keeping the
temperature at the target value According to the heating schedule the most expected
temperature in the soaking zone is 1200 Therefore the target temperature
is set at this value to achieve the desired heating effect As shown in Fig 11 in the real
18
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
operation the temperature fluctuates around the target value because of the imperfect
control performance However in order to simplify the analysis and compute the
maximum possible energy saving perfect temperature control is assumed when
adopting the model-based optimization strategy In other words it is assumed that the
actual temperature in the lsquouprsquo area of the soaking zone can be adjusted to the expected
temperature ie 1200
0 02 04 06 08 1 12 14 16 18 2
x 104
1150
1200
1250
1300
Sampling intervals
T S-u (
)
OriginalTarget
Fig 11 Original and target temperature in the lsquouprsquo area of the soaking zone
The adjustment scheme takes the maximum value and minimum value of the original
operation data as the upper and lower bounds for the adjustment Moreover in order to
assure the validity of the linear model the increment or decrement of AG and G should
not beyond of the original value (considered as 10 for the purposes of this paper)
The adjustment strategy can be formulated as
19
307
308
309
310
311
312
313
314
315
316
317
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
(3)
where NNG() denotes the NNG regression model of the TG ratio and are the
NNG regression coefficients of the AG and G in the lsquouprsquo area of the soaking zone for
modeling TG in same area
With the model-based strategy and adaptive model of the TG ratio given in Section
42 the optimization adjustment results can be obtained as follows
The adjustment of gas flow in the lsquouprsquo area of the soaking zone shown in Fig 12
illustrates that the implementation of the model-based optimization operation can reduce
the consumption of the gas flow Compared with the original case 938 of the gas can
be saved on average by utilizing the model-based adjustment
0 02 04 06 08 1 12 14 16 18 2
x 104
-800
-700
-600
-500
-400
-300
-200
-100
0
Sampling intervals
Adj
ustm
ent o
f gas
flow
m(
3 h)
20
318
319
320
321
322
323
324
325
326
327
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
(a) Adjustment amount of gas flow
0 02 04 06 08 1 12 14 16 18 2
x 104
500
1000
1500
2000
2500
3000
3500
4000
4500
5000G
as fl
owm
(3 h
)
Sampling intervals
OriginalAdjusted
(b) Comparison of original and adjusted gas flow
095 096 097 098 099 1 101 102 103 104 105
x 104
500
1000
1500
2000
2500
3000
3500
4000
4500
5000
Gas
flow
m(
3 h)
Sampling intervals
OriginalAdjusted
(c) Comparison of original and adjusted gas flow in interval [9501-10500]
Fig 12 Adjustment of gas flow in the lsquouprsquo area of the soaking zone
The adjustment of the air flow in the lsquouprsquo area of the soaking zone given in Fig 13
shows that the air flow is reduced in most cases and only increased for a few cases In
21
328
329
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
general 681 of the air flow is reduced after the implementation of the model-based
optimization
0 02 04 06 08 1 12 14 16 18 2
x 104
-2000
-1500
-1000
-500
0
500
1000
Sampling intervals
Adj
ustm
ent o
f air
flow
(m3 h
)
(a) Adjustment amount of air flow
0 02 04 06 08 1 12 14 16 18 2
x 104
4000
5000
6000
7000
8000
9000
10000
Air
flow
m(
3 h)
Sampling intervals
OriginalAdjusted
(b) Comparison of original and adjusted air flow
22
330
331
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
095 096 097 098 099 1 101 102 103 104 105
x 104
4000
5000
6000
7000
8000
9000
10000
Air
flow
m(
3 h)
Sampling intervals
OriginalAdjusted
(c) Comparison of original and adjusted air flow in interval [9501-10500]
Fig 13 Adjustment of air flow in the lsquouprsquo area of the soaking zone
52 Discussions
In this section detailed analysis for the optimization operation results is provided
The statistics of optimization at the boundary conditions shown in Table 3 illustrates
that the adjusted value for G reaches its lower limit in most cases (8045 for
and 687 for ) which is the most energy-efficient point In a number of cases
(1079 for and 001 for ) the adjusted AG reaches its upper
limit These results indicate that the optimization operations maximize the combustion
efficiency by decreasing G and increasing the AG ratio up to the boundary conditions
However the lower limit is also unexpectedly attained in a small number of cases
(110 for and 078 for ) for which a more detailed analysis is
given below
23
332
333
334
335
336
337
338
339
340
341
342
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
Table 3 Statistics of optimization at boundary conditions
Total
Amount 0 1374 0 16090 17464
Percentag
e 0 687 0 8045
8732
Total
Amount 2 220 2158 156 2536
Percentag
e 001 110 1079 078
1268
Table 4 Statistics for different optimization operations
Adjustment
Gdarr
Guarr TotalAGuarr AGdarr
Tdarr Tuarr Tdarr Tuarr
Amount 9548 3744 6328 380 0 20000
Percentage 4774 1872 3164 190 0 100
The statistics for different optimization operations is performed and shown in Table
4 It can be seen that all the adjusted operations result in the reduction of gas
consumption In over half the cases (4774 with decreased temperature and 1872
with increased temperature) AG is adjusted to higher levels Meanwhile a number of
the adjustments (3164) lower the temperature by reducing the G and AG at the same
time This result indicates that the original temperature cannot be adjusted to the target
by only reducing G in the constraint conditions The remaining cases (19) are
24
343
344
345
346
347
348
349
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
relatively special in which an excess of air is supplied in the original operation and the
adjusted operation thereby increasing the temperature by a smaller AG with less gas
consumption
6 Conclusion
Based on the actual operation data this paper aims to explore the improvement of the
combustion efficiency and the room for energy conservation Correlation analysis and
LDA show that it is of great significance to introduce two derived ratio variables which
are the AG ratio and the TG ratio A type of combustion efficiency utilizing an on-line
soft sensor is put forward by employing a NNG variable selection algorithm which
provides a good solution to the combustion efficiency real-time measurement problem
of a reheating furnace The implementation of the model-based optimization is studied
based on the actual operational data Detailed analysis for the optimization results is
given for the different cases The results show that significant energy conservation can
be achieved when the furnace operation is optimized by using the developed soft sensor
model
Acknowledgement
The authors would like to thank the financial support provided by the National Nature
Science Foundation of China under Grant 61171145 Y Yao was supported by Ministry
of Science amp Technology ROC under Grant number MOST 104-2221-E-007-129
References
25
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[1] Z J Wang Q D Wu and T Y Chai Optimal-setting control for complicated
industrial processes and its application studyControl Engineering Practice vol
12 pp 65-74 2004
[2] A Steinboeck K Graichen and A Kugi Dynamic Optimization of a Slab
Reheating Furnace With Consistent Approximation of Control VariablesIEEE
Transactions on Control Systems Technology vol 19 pp 1444-1456 2011
[3] B T Zhang C Y Wang Q Qin and L Li Influence of Boiler Combustion
Adjustment on NOxEmission and Boiler EfficiencyAdvanced Materials
Research vol 732-733 pp 234-237 2013
[4] C K Yoo and IB Lee Soft Sensor and Adaptive Model-Based Dissolved
Oxygen Control for Biological Wastewater Treatment ProcessesEnvironmental
Engineering Science vol 21 pp 331-340 2004
[5] S A Bhat D N Saraf S Gupta and S K Gupta Use of Agitator Power as a
Soft Sensor for Bulk Free-Radical Polymerization of Methyl Methacrylate in
Batch ReactorsIndustrial amp Engineering Chemistry Research vol 45 pp 4243-
4255 2006
[6] K Desai Y Badhe S S Tambe and B D Kulkarni Soft-sensor development
for fed-batch bioreactors using support vector regressionBiochemical
Engineering Journal vol 27 pp 225-239 2006
[7] Y P Badhe Lonari J Tambe S S amp Kulkarni B D Improve polyethylene
process control and product qualityHydrocarbon Processing vol 86 pp 53-60
2007
26
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[8] N K Nath K Mandal A K Singh B Basu C Bhanu S Kumar et al Ladle
furnace on-line reckoner for prediction and control of steel temperature and
compositionIronmaking amp Steelmaking vol 33 pp 140-150 2006
[9] A J Yan T Y Chai F H Wu and P Wang Hybrid intelligent control of
combustion process for ore-roasting furnaceJournal of Control Theory and
Applications vol 6 pp 80-85 2008
[10] J Li W M Zhong H Cheng X D Kong and F Qian A data-driven soft
sensor modeling for furnace temperature of Opposed Multi-Burner gasifier pp
705-710 2011
[11] Y H Yang Y H Liu X Z Liu and S K Qin Billet temperature soft sensor
model of reheating furnace based on RVM method pp 4003-4006 2011
[12] J H Wang C Wang X F Zhu and X K Fang Application of soft sensor in
welding seam tracking prediction based on LSSVM and PSO with compression
factor pp 2441-2446 2013
[13] L Balbis J Balderud and M J Grimble Nonlinear predictive control of steel
slab reheating furnace pp 1679-1684 2008
[14] A Steinboeck D Wild T Kiefer and A Kugi A mathematical model of a slab
reheating furnace with radiative heat transfer and non-participating gaseous
media International Journal Of Heat And Mass Transfer vol 53 pp 5933-
5946 Dec 2010
[15] A Steinboeck D Wild and A Kugi Nonlinear model predictive control of a
continuous slab reheating furnace Control Engineering Practice vol 21 pp
495-508 2013
27
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[16] C Zhang T Ishii and S Sugiyama Numerical Modeling Of the Thermal
Performance Of Regenerative Slab Reheat Furnaces Numerical Heat Transfer
Part A Applications vol 32 pp 613-631 1997
[17] J G Kim and K Y Huh Prediction of Transient Slab Temperature Distribution
in the Re-heating Furnace of a Walking-beam Type for Rolling of Steel Slabs
ISIJ International vol 40 pp 1115-1123 2000
[18] J G Kim K Y and H I T K Three-Dimensional Analysis Of the Walking-
Beam-Type Slab Reheating Furnace In Hot Strip Mills Numerical Heat
Transfer Part A Applications vol 38 pp 589-609 2000
[19] CT Hsieh MJ Huang ST Lee and CH Wang Numerical Modeling of a
Walking-Beam-Type Slab Reheating Furnace Numerical Heat Transfer Part A
Applications vol 53 pp 966-981 2008
[20] MJ Huang CT Hsieh ST Lee and CH Wang A Coupled Numerical Study
of Slab Temperature and Gas Temperature in the Walking-Beam-Type Slab
Reheating Furnace Numerical Heat Transfer Part A Applications vol 54 pp
625-646 2008
[21] Z Song and A Kusiak Constraint-Based Control of Boiler Efficiency A Data-
Mining Approach IEEE Transactions on Industrial Informatics vol 3 pp 73-
83 2007
[22] J Q Li J J Gu and C L Niu The Operation Optimization based on
Correlation Analysis of Operation Parameters in Power Plant pp 138-141
2008
28
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[23] J G Wang S S Shieh S S Jang D S H Wong and C W Wu A two-tier
approach to the data-driven modeling on thermal efficiency of a BFGcoal co-
firing boiler Fuel vol 111 pp 528-534 Sep 2013
[24] L Breiman Better Subset Regression Using the Nonnegative Garrote
Technometrics vol 37 pp 373-384 1995
[25] R A Fisher The use of multiple measurements in taxonomic problems
AnnHum Genet vol 7 pp 179-188 1936
[26] J Liu Developing a soft sensor based on sparse partial least squares with
variable selection Journal of Process Control vol 24 pp 1046-1056 2014
[27] K Sun J Liu JL Kang SS Jang D SH Wong and DS Chen
Development of a variable selection method for soft sensor using artificial
neural network and nonnegative garrote Journal of Process Control vol 24 pp
1068-1075 2014
29
439
440
441
442
443
444
445
446
447
448
449
450
451
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
steel slab moves through the four zones in turn and is heated to the demanded state
using a specific temperature increase curve As is shown in Table 1 the soaking zone
has two areas that are defined as up and down and both of the areas possess the same
five variables including two manipulated variables the air flow (A) and the gas flow
(G) and three temperatures in left center and right sections of the area (T-l T-c and T-
r) The other zones have the same variables as the soaking zone hence there are 40
variables in total for the reheating furnace
Fig 1 The schematic of the heating process in the reheating furnace
Table 1 Variables and descriptions in the soaking zone
Variable Description unit
AS-u Air flow in the lsquouprsquo area Nm3h
AS-d Air flow in the lsquodownrsquo area Nm3h
GS-u Gas flow in the lsquouprsquo area Nm3h
GS-d Gas flow in the lsquodownrsquo area Nm3h
TS-ul Temperature in the left part of the lsquouprsquo area
TS-uc Temperature in the center part of the lsquouprsquo area
TS-ur Temperature in the right part of the lsquouprsquo area
7
143
144
145
146
147
148
149
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
TS-dl Temperature in the left part of the lsquodownrsquo area
TS-dc Temperature in the center part of the lsquodownrsquo area
TS-dr Temperature in the right part of the lsquodownrsquo area
A data set of 20000 samples was used in this study The samples were collected from
an actual reheating furnace in a large iron and steel plant located in Shanghai from
September 14 to September 27 2014 The operational data is taken on a per minute
basis
1
092
093
042
041
046
092
1
091
04
039
045
093
091
1
038
038
045
042
04
038
1
096
096
041
039
038
096
1
096
046
045
045
096
096
1
TS-ul
TS-uc
TS-ur
TS-dl
TS-dc
TS-dr
TS-ul TS-uc TS-ur TS-dl TS-dc TS-dr
04
05
06
07
08
09
1
Fig 2 Correlation between each temperature in the soaking zone
In order to investigate the relation among different temperatures in each zone
correlation analysis is conducted for the soaking zone as illustrated in Fig 2 It can be
seen that the temperature in three parts of the lsquouprsquo area is highly correlated with a
correlation coefficient greater than 09 The similar results exist for the lsquodownrsquo area as
well On the contrary the correlation coefficient between temperatures in one part of
8
150
151
152
153
154
155
156
157
158
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
lsquouprsquo area and any part of lsquodownrsquo area does not exceed 05 Therefore for the reduction
of the data dimension the three temperatures in the lsquouprsquo area or the lsquodownrsquo area can be
treated as only one variable which can be taken as the mean value or the first principal
component acquired from the PCA analysis Considering the reservation of the variable
physical meaning the former is preferred and used
3 Statistics analysis and incorporation of derived variables
In this section in order to uncover the physical knowledge for the actual operation
guidance and confirmation statistical analysis is performed for the 16 input manipulated
variables and the eight output variables (ie the temperatures in the four zones) of a
reheating furnace system For the combustion efficiency evaluation and modeling two
types of derived ratio variables are introduced which is helpful to reveal the
information included in the data
31 Correlation analysis
Correlation analysis between the temperatures (T) in each area and all of the air flows
(A) and the gas flows (G) is performed and shown in Fig 3 It can be seen that only the
temperatures in both areas of the soaking zone are the most highly related to the air flow
and the gas flow in its own zone However this phenomenon does not occur in the other
three zones The temperatures in the second heating zone mainly depend on the air flow
and the gas flow in its own zone as well as the nearby first heating zone As for the
preheating zone and first heating zone no apparent correlation can be observed
Obviously these analysis results could not tell us the explicit information about how to
evaluate the efficiency levels and key manipulated variables
9
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
-101
T P-u
AP GP A1 G1 A2 G2 AS GS
AP GP A1 G1 A2 G2 AS GS
AP GP A1 G1 A2 G2 AS GS
-101
T P-d
-101
T 1-u
-101
T 1-d
AP GP A1 G1 A2 G2 AS GS
-101
T 2-u
AP GP A1 G1 A2 G2 AS GS
-101
T 2-d
-101
T S-u
-101
T S-d
AP GP A1 G1 A2 G2 AS GS
AP GP A1 G1 A2 G2 AS GS
AP GP A1 G1 A2 G2 AS GS
UPDOWN
Fig 3 Correlation between air flows gas flows and temperatures
32 Incorporation of derived variables
During the stable heating stage the quantity of heat absorbed and removed from the
slabs from each furnace zone is relatively constant Hence the derived variable TG
ratio can be treated as an index for the combustion efficiency level This is because a
higher TG ratio signifies more combustion heat generated from unit gas ie higher
combustion efficiency
Moreover it is known that the appropriate air and fuel ratio is vital for the
combustion efficiency so the air-gas ratio (AG) is utilized as another derived variable
for the research Again the correlation analysis is performed for two types of derived
variables The correlations between different variables including the AG ratio G and
the TG ratio are shown in Fig 4 It can be clearly seen from the four red rectangle
blocks that TG of each zone is only remarkably related to AG and G in the same zone
10
183
184
185
186
187
188
189
190
191
192
193
194
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
Furthermore as related to the lsquouprsquo or lsquodownrsquo areas in one zone TG in each area has the
highest correlation with AG and G in the same area while AG and G in the opposite
area of the same zone is secondary This can be easily seen from the red and blue color
markings in each red rectangle block Thus it is of great significance to introduce these
derived ratio variables
-101
TG
P-u
AGP GP AG1G1 AG2 G2 AGS GS
-101
TG
P-d
AGP GP AG1 G1 AG2 G2 AGS GS
-101
TG
1-u
AGP GP AG1 G1 AG2 G2 AGS GS
-101
TG
1-d
AGP GP AG1 G1 AG2 G2 AGS GS
-101
TG
2-u
AGP GP AG1 G1 AG2 G2 AGS GS
-101
TG
2-d
AGP GP AG1 G1 AG2 G2 AGS GS
-101
TG
S-u
AGP GP AG1 G1 AG2 G2 AGS GS
-101
TG
S-d
AGP GP AG1 G1 AG2 G2 AGS GS
UPDOWN
Fig 4 Correlation between the air-gas ratio gas flow and TG ratio
33 LDA analysis
Linear discriminant analysis (LDA) aims to finding a projection direction that
maximizes the separation of class means and minimizes the within-class variance [25]
In this section LDA is utilized to identify the discriminating variables that play an
important role in determining combustion efficiency levels All the data are partitioned
into five groups according to their efficiency levels
11
195
196
197
198
199
200
201
202
203
204
205
206
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
Based on a descending order of TG ratios five efficiency levels for the lsquouprsquo area of
the soaking zone are denoted as HH H M L and LL LDA is conducted on the three
groups of the data with levels of HH M and LL
Fig 5 shows the scattering of the LDA projections of the process observations
collected at the three efficiency levels where y1 and y2 correspond to the first two LDA
components that contain most discriminant information The weighting factors ( and
) of the 16 input variables composing the projections y1 and y2 are shown in Fig 6
where and From left to right the 16 input variables are defined as
the eight AG ratio variables and the eight gas flow variables with the order of variables
of each kind P-u P-d 1-u 1-d 2-u 2-d S-u and S-d
-8 -6 -4 -2 0 2 4 6-4
-3
-2
-1
0
1
2
3
4
5
6
Projection y1
Pro
ject
ion
y2
HHMLL
Fig 5 Scattering of LDA projections y1 and y2 for three efficiency levels
12
207
208
209
210
211
212
213
214
215
216
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
AG_S-u-4
-2
0
2
Wei
ghtin
g fa
ctor
for y
1
AG_S-u
-101234
Wei
ghtin
g fa
ctor
for y
2
Fig 6 Weight factors of various input variables for projections y1 and y2
The two figures reveal that the three groups are clearly discriminated by the LDA
projection and the most significant variables for the different efficiency levels are the
derived variables ie the AG in the lsquouprsquo area of the soaking zone A similar result can
be obtained in the lsquodownrsquo area or any area of the other zones Therefore AG in each
heating area is the key manipulated variable that determines the different combustion
efficiency
4 Modeling and prediction of temperature and temperature-gas ratio
For the model-based operation optimization the models for the temperature and TG
ratio based on the NNG algorithm will be developed and compared
41 NNG variable selection algorithm
The NNG method can be generalized into a two-stage shrinkage method In the first
stage the sign for each variable is determined using the ordinary least square procedure
13
217
218
219
220
221
222
223
224
225
226
227
228
229
230
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
and in the second stage the corresponding magnitudes are computed by solving a series
of constrained quadratic programming
A set of observation data is provided where is the input matrix whose
columns represent the measured candidate variables and is the corresponding
vector of the response data The following expression is given with the number of the
response variable being equal to 1 but a similar procedure can be generalized to any
number of variables Let and be normalized to the zero-mean and unit standard
deviation Additionally let be a set of the ordinary least square estimates of the
coefficients of the following linear model then
(1)
The second stage shrinkage can be formulated as the following optimization problem
subject to
(2)
As decreased and the NNG is tightened more of the become zero and the
remaining nonzero coefficients are shrunk A solution path exists with on which
the appropriate shrinkage can be selected Conventionally the v-fold cross-validation is
used to estimate the prediction error and to select the best solution in the solution path
so as to minimize the prediction or model error
42 Modeling and prediction of temperature and TG ratio
The input-output relations change with time as the reheating process develops An
adaptive modeling strategy is often used to resolve time-varying characteristics of
14
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
industrial processes In this paper the NNG-based regression modeling prediction and
optimization is implemented in a moving window manner where the size of the
window and the length of the moving step are selected as 1000 and 100 respectively In
addition the prediction horizon is also set as 100 In each step the NNG regression
model is built based on the data in the current window which is then used for the
prediction and optimization in the subsequent prediction horizon Next the window
moves forward by replacing the oldest 100 samples for model training by the
observations collected in the previous prediction horizon This moving window strategy
is workable because in the reheating process the input-output relation is slowly time-
varying and the model is still valid for the prediction and optimization in the subsequent
short time
The modeling of the temperature T is based on the air flow rates and gas flow rates
while the modeling for the TG ratio is based on the AG ratios and gas flow rates
Taking the lsquouprsquo area of the soaking zone as an example the prediction result for T and
the TG ratio in the lsquouprsquo area of the soaking zone is shown in Figs 7 and 8 respectively
0 02 04 06 08 1 12 14 16 18 2
x 104
1140
1160
1180
1200
1220
1240
1260
1280
1300
1320
Sampling intervals
T S-u (
)
OriginalPredicted
Fig 7 Prediction of temperature in the lsquouprsquo area of the soaking zone
15
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
0 02 04 06 08 1 12 14 16 18 2
x 104
02
04
06
08
1
12
14
16
18
2
Sampling intervals
TG
S-u
OriginalPredicted
Fig 8 Prediction of TG ratio in the lsquouprsquo area of the soaking zone
A comparison between Figs 7 and 8 show that the prediction of the TG ratio is more
accurate than the prediction of temperature The average R2 in 190 NNG regressions is
0935 for the TG ratio while the average R2 is only 0814 for temperature This is
consistent with the statistical analysis which reveals that the correlation between AG
and TG is higher than that between A and T These results indicate that derived
variables are more meaningful for the purpose of prediction and the modeling of the
TG ratio is more appropriate for implementing optimization
The selected frequency for each variable in 190 NNG regressions and the coefficients
of each variable in 20 NNG regressions for the TG ratio modeling of the lsquouprsquo area of
the soaking zone are shown in Figs 9 and 10 respectively Fig 9 shows that the selected
frequency of the variables in the lsquouprsquo area of the soaking zone is much higher than
variables in other zones Similarly as is shown in Fig 10 the NNG regression
coefficients of the variables in the lsquouprsquo area of the soaking zone are much larger than the
variables in the other areas These results indicated that variables in the lsquouprsquo area of
16
269
270
271
272
273
274
275
276
277
278
279
280
281
282
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
soaking zone are mostly contributed to the modeling of the TG ratio in same area
which is quite reasonable Similar results can be obtained for the other areas
0
20
40
60
80
100
120
140
160
180
200
Sel
ecte
d fre
quen
cy
AGP-u AGP-d AG1-u AG1-d AG2-u AG2-d AGS-u AGS-d GP-u GP-d G1-u G1-d G2-u G2-d GS-u GS-d
Fig 9 Variable selected frequency of over 190 runs for the TG ratio model of the lsquouprsquo
area of the soaking zone
0 2 4 6 8 10 12 14 16 18 20-05
05
AG
P-u
0 2 4 6 8 10 12 14 16 18 20-05
05
GP
-u
0 2 4 6 8 10 12 14 16 18 20-1
0
AG
P-d
0 2 4 6 8 10 12 14 16 18 20-1
0
GP
-d
0 2 4 6 8 10 12 14 16 18 20-05
05
AG
1-u
0 2 4 6 8 10 12 14 16 18 20-05
05
G1-
u
0 2 4 6 8 10 12 14 16 18 20-05
05
AG
1-d
0 2 4 6 8 10 12 14 16 18 20-05
05
G1-
d
0 2 4 6 8 10 12 14 16 18 20-1
0
AG
2-u
0 2 4 6 8 10 12 14 16 18 200
1
G2-
u
0 2 4 6 8 10 12 14 16 18 20-1
0
AG
2-d
0 2 4 6 8 10 12 14 16 18 20-1
0
G2-
d
0 2 4 6 8 10 12 14 16 18 200
1
AG
S-u
0 2 4 6 8 10 12 14 16 18 20-1
0
GS
-u
0 2 4 6 8 10 12 14 16 18 20-05
05
AG
S-d
0 2 4 6 8 10 12 14 16 18 20-05
05
GS
-d
Fig 10 Part of the NNG regression coefficients of each variable over 190 runs for the
TG ratio model of the lsquouprsquo area of the soaking zone
It is remarkable that the NNG regression coefficients corresponding to the AG ratio
in the lsquouprsquo area of the soaking zone are consistently positive On the contrary the
coefficients of the gas flow rate in the same area are consistently negative This
indicates that under this condition the rise of the AG leads to the increase of the TG
17
283
284
285
286
287
288
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
while the rise of the G leads to the drop of the TG This property is helpful for
performing optimization and improve combustion efficiency
For comparison the modeling results for two other algorithms artificial neural
network (ANN) and partial least squares (PLS) [26] are compared with the NNG
algorithm in the following three aspects [27]
(1) Model Size the number of variables selected for modeling
(2) Model Magnitude the mean of the L1 norm of the regression coefficients
(3) Prediction Precision the mean squared prediction error (MSPE)
Summary of the algorithm comparison is shown in Table 2 It can be seen that the
superiorities of the NNG regression in model size model magnitude and model
precision are remarkable
Table 2 Summary of algorithm comparison
Index NNG ANN PLS
Model Size 849 16 16
Model Magnitude 10075 -- 12150
MSPE 00093 00108 00140
5 Model-based optimization
51 Implemention of model-based optimization operation
The goal of optimization is to seek an optimal combination of AG and G in the lsquouprsquo
area of the soaking zone so as to minimize the gas consumption while keeping the
temperature at the target value According to the heating schedule the most expected
temperature in the soaking zone is 1200 Therefore the target temperature
is set at this value to achieve the desired heating effect As shown in Fig 11 in the real
18
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
operation the temperature fluctuates around the target value because of the imperfect
control performance However in order to simplify the analysis and compute the
maximum possible energy saving perfect temperature control is assumed when
adopting the model-based optimization strategy In other words it is assumed that the
actual temperature in the lsquouprsquo area of the soaking zone can be adjusted to the expected
temperature ie 1200
0 02 04 06 08 1 12 14 16 18 2
x 104
1150
1200
1250
1300
Sampling intervals
T S-u (
)
OriginalTarget
Fig 11 Original and target temperature in the lsquouprsquo area of the soaking zone
The adjustment scheme takes the maximum value and minimum value of the original
operation data as the upper and lower bounds for the adjustment Moreover in order to
assure the validity of the linear model the increment or decrement of AG and G should
not beyond of the original value (considered as 10 for the purposes of this paper)
The adjustment strategy can be formulated as
19
307
308
309
310
311
312
313
314
315
316
317
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
(3)
where NNG() denotes the NNG regression model of the TG ratio and are the
NNG regression coefficients of the AG and G in the lsquouprsquo area of the soaking zone for
modeling TG in same area
With the model-based strategy and adaptive model of the TG ratio given in Section
42 the optimization adjustment results can be obtained as follows
The adjustment of gas flow in the lsquouprsquo area of the soaking zone shown in Fig 12
illustrates that the implementation of the model-based optimization operation can reduce
the consumption of the gas flow Compared with the original case 938 of the gas can
be saved on average by utilizing the model-based adjustment
0 02 04 06 08 1 12 14 16 18 2
x 104
-800
-700
-600
-500
-400
-300
-200
-100
0
Sampling intervals
Adj
ustm
ent o
f gas
flow
m(
3 h)
20
318
319
320
321
322
323
324
325
326
327
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
(a) Adjustment amount of gas flow
0 02 04 06 08 1 12 14 16 18 2
x 104
500
1000
1500
2000
2500
3000
3500
4000
4500
5000G
as fl
owm
(3 h
)
Sampling intervals
OriginalAdjusted
(b) Comparison of original and adjusted gas flow
095 096 097 098 099 1 101 102 103 104 105
x 104
500
1000
1500
2000
2500
3000
3500
4000
4500
5000
Gas
flow
m(
3 h)
Sampling intervals
OriginalAdjusted
(c) Comparison of original and adjusted gas flow in interval [9501-10500]
Fig 12 Adjustment of gas flow in the lsquouprsquo area of the soaking zone
The adjustment of the air flow in the lsquouprsquo area of the soaking zone given in Fig 13
shows that the air flow is reduced in most cases and only increased for a few cases In
21
328
329
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
general 681 of the air flow is reduced after the implementation of the model-based
optimization
0 02 04 06 08 1 12 14 16 18 2
x 104
-2000
-1500
-1000
-500
0
500
1000
Sampling intervals
Adj
ustm
ent o
f air
flow
(m3 h
)
(a) Adjustment amount of air flow
0 02 04 06 08 1 12 14 16 18 2
x 104
4000
5000
6000
7000
8000
9000
10000
Air
flow
m(
3 h)
Sampling intervals
OriginalAdjusted
(b) Comparison of original and adjusted air flow
22
330
331
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
095 096 097 098 099 1 101 102 103 104 105
x 104
4000
5000
6000
7000
8000
9000
10000
Air
flow
m(
3 h)
Sampling intervals
OriginalAdjusted
(c) Comparison of original and adjusted air flow in interval [9501-10500]
Fig 13 Adjustment of air flow in the lsquouprsquo area of the soaking zone
52 Discussions
In this section detailed analysis for the optimization operation results is provided
The statistics of optimization at the boundary conditions shown in Table 3 illustrates
that the adjusted value for G reaches its lower limit in most cases (8045 for
and 687 for ) which is the most energy-efficient point In a number of cases
(1079 for and 001 for ) the adjusted AG reaches its upper
limit These results indicate that the optimization operations maximize the combustion
efficiency by decreasing G and increasing the AG ratio up to the boundary conditions
However the lower limit is also unexpectedly attained in a small number of cases
(110 for and 078 for ) for which a more detailed analysis is
given below
23
332
333
334
335
336
337
338
339
340
341
342
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
Table 3 Statistics of optimization at boundary conditions
Total
Amount 0 1374 0 16090 17464
Percentag
e 0 687 0 8045
8732
Total
Amount 2 220 2158 156 2536
Percentag
e 001 110 1079 078
1268
Table 4 Statistics for different optimization operations
Adjustment
Gdarr
Guarr TotalAGuarr AGdarr
Tdarr Tuarr Tdarr Tuarr
Amount 9548 3744 6328 380 0 20000
Percentage 4774 1872 3164 190 0 100
The statistics for different optimization operations is performed and shown in Table
4 It can be seen that all the adjusted operations result in the reduction of gas
consumption In over half the cases (4774 with decreased temperature and 1872
with increased temperature) AG is adjusted to higher levels Meanwhile a number of
the adjustments (3164) lower the temperature by reducing the G and AG at the same
time This result indicates that the original temperature cannot be adjusted to the target
by only reducing G in the constraint conditions The remaining cases (19) are
24
343
344
345
346
347
348
349
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
relatively special in which an excess of air is supplied in the original operation and the
adjusted operation thereby increasing the temperature by a smaller AG with less gas
consumption
6 Conclusion
Based on the actual operation data this paper aims to explore the improvement of the
combustion efficiency and the room for energy conservation Correlation analysis and
LDA show that it is of great significance to introduce two derived ratio variables which
are the AG ratio and the TG ratio A type of combustion efficiency utilizing an on-line
soft sensor is put forward by employing a NNG variable selection algorithm which
provides a good solution to the combustion efficiency real-time measurement problem
of a reheating furnace The implementation of the model-based optimization is studied
based on the actual operational data Detailed analysis for the optimization results is
given for the different cases The results show that significant energy conservation can
be achieved when the furnace operation is optimized by using the developed soft sensor
model
Acknowledgement
The authors would like to thank the financial support provided by the National Nature
Science Foundation of China under Grant 61171145 Y Yao was supported by Ministry
of Science amp Technology ROC under Grant number MOST 104-2221-E-007-129
References
25
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[1] Z J Wang Q D Wu and T Y Chai Optimal-setting control for complicated
industrial processes and its application studyControl Engineering Practice vol
12 pp 65-74 2004
[2] A Steinboeck K Graichen and A Kugi Dynamic Optimization of a Slab
Reheating Furnace With Consistent Approximation of Control VariablesIEEE
Transactions on Control Systems Technology vol 19 pp 1444-1456 2011
[3] B T Zhang C Y Wang Q Qin and L Li Influence of Boiler Combustion
Adjustment on NOxEmission and Boiler EfficiencyAdvanced Materials
Research vol 732-733 pp 234-237 2013
[4] C K Yoo and IB Lee Soft Sensor and Adaptive Model-Based Dissolved
Oxygen Control for Biological Wastewater Treatment ProcessesEnvironmental
Engineering Science vol 21 pp 331-340 2004
[5] S A Bhat D N Saraf S Gupta and S K Gupta Use of Agitator Power as a
Soft Sensor for Bulk Free-Radical Polymerization of Methyl Methacrylate in
Batch ReactorsIndustrial amp Engineering Chemistry Research vol 45 pp 4243-
4255 2006
[6] K Desai Y Badhe S S Tambe and B D Kulkarni Soft-sensor development
for fed-batch bioreactors using support vector regressionBiochemical
Engineering Journal vol 27 pp 225-239 2006
[7] Y P Badhe Lonari J Tambe S S amp Kulkarni B D Improve polyethylene
process control and product qualityHydrocarbon Processing vol 86 pp 53-60
2007
26
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[8] N K Nath K Mandal A K Singh B Basu C Bhanu S Kumar et al Ladle
furnace on-line reckoner for prediction and control of steel temperature and
compositionIronmaking amp Steelmaking vol 33 pp 140-150 2006
[9] A J Yan T Y Chai F H Wu and P Wang Hybrid intelligent control of
combustion process for ore-roasting furnaceJournal of Control Theory and
Applications vol 6 pp 80-85 2008
[10] J Li W M Zhong H Cheng X D Kong and F Qian A data-driven soft
sensor modeling for furnace temperature of Opposed Multi-Burner gasifier pp
705-710 2011
[11] Y H Yang Y H Liu X Z Liu and S K Qin Billet temperature soft sensor
model of reheating furnace based on RVM method pp 4003-4006 2011
[12] J H Wang C Wang X F Zhu and X K Fang Application of soft sensor in
welding seam tracking prediction based on LSSVM and PSO with compression
factor pp 2441-2446 2013
[13] L Balbis J Balderud and M J Grimble Nonlinear predictive control of steel
slab reheating furnace pp 1679-1684 2008
[14] A Steinboeck D Wild T Kiefer and A Kugi A mathematical model of a slab
reheating furnace with radiative heat transfer and non-participating gaseous
media International Journal Of Heat And Mass Transfer vol 53 pp 5933-
5946 Dec 2010
[15] A Steinboeck D Wild and A Kugi Nonlinear model predictive control of a
continuous slab reheating furnace Control Engineering Practice vol 21 pp
495-508 2013
27
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[16] C Zhang T Ishii and S Sugiyama Numerical Modeling Of the Thermal
Performance Of Regenerative Slab Reheat Furnaces Numerical Heat Transfer
Part A Applications vol 32 pp 613-631 1997
[17] J G Kim and K Y Huh Prediction of Transient Slab Temperature Distribution
in the Re-heating Furnace of a Walking-beam Type for Rolling of Steel Slabs
ISIJ International vol 40 pp 1115-1123 2000
[18] J G Kim K Y and H I T K Three-Dimensional Analysis Of the Walking-
Beam-Type Slab Reheating Furnace In Hot Strip Mills Numerical Heat
Transfer Part A Applications vol 38 pp 589-609 2000
[19] CT Hsieh MJ Huang ST Lee and CH Wang Numerical Modeling of a
Walking-Beam-Type Slab Reheating Furnace Numerical Heat Transfer Part A
Applications vol 53 pp 966-981 2008
[20] MJ Huang CT Hsieh ST Lee and CH Wang A Coupled Numerical Study
of Slab Temperature and Gas Temperature in the Walking-Beam-Type Slab
Reheating Furnace Numerical Heat Transfer Part A Applications vol 54 pp
625-646 2008
[21] Z Song and A Kusiak Constraint-Based Control of Boiler Efficiency A Data-
Mining Approach IEEE Transactions on Industrial Informatics vol 3 pp 73-
83 2007
[22] J Q Li J J Gu and C L Niu The Operation Optimization based on
Correlation Analysis of Operation Parameters in Power Plant pp 138-141
2008
28
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[23] J G Wang S S Shieh S S Jang D S H Wong and C W Wu A two-tier
approach to the data-driven modeling on thermal efficiency of a BFGcoal co-
firing boiler Fuel vol 111 pp 528-534 Sep 2013
[24] L Breiman Better Subset Regression Using the Nonnegative Garrote
Technometrics vol 37 pp 373-384 1995
[25] R A Fisher The use of multiple measurements in taxonomic problems
AnnHum Genet vol 7 pp 179-188 1936
[26] J Liu Developing a soft sensor based on sparse partial least squares with
variable selection Journal of Process Control vol 24 pp 1046-1056 2014
[27] K Sun J Liu JL Kang SS Jang D SH Wong and DS Chen
Development of a variable selection method for soft sensor using artificial
neural network and nonnegative garrote Journal of Process Control vol 24 pp
1068-1075 2014
29
439
440
441
442
443
444
445
446
447
448
449
450
451
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
TS-dl Temperature in the left part of the lsquodownrsquo area
TS-dc Temperature in the center part of the lsquodownrsquo area
TS-dr Temperature in the right part of the lsquodownrsquo area
A data set of 20000 samples was used in this study The samples were collected from
an actual reheating furnace in a large iron and steel plant located in Shanghai from
September 14 to September 27 2014 The operational data is taken on a per minute
basis
1
092
093
042
041
046
092
1
091
04
039
045
093
091
1
038
038
045
042
04
038
1
096
096
041
039
038
096
1
096
046
045
045
096
096
1
TS-ul
TS-uc
TS-ur
TS-dl
TS-dc
TS-dr
TS-ul TS-uc TS-ur TS-dl TS-dc TS-dr
04
05
06
07
08
09
1
Fig 2 Correlation between each temperature in the soaking zone
In order to investigate the relation among different temperatures in each zone
correlation analysis is conducted for the soaking zone as illustrated in Fig 2 It can be
seen that the temperature in three parts of the lsquouprsquo area is highly correlated with a
correlation coefficient greater than 09 The similar results exist for the lsquodownrsquo area as
well On the contrary the correlation coefficient between temperatures in one part of
8
150
151
152
153
154
155
156
157
158
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
lsquouprsquo area and any part of lsquodownrsquo area does not exceed 05 Therefore for the reduction
of the data dimension the three temperatures in the lsquouprsquo area or the lsquodownrsquo area can be
treated as only one variable which can be taken as the mean value or the first principal
component acquired from the PCA analysis Considering the reservation of the variable
physical meaning the former is preferred and used
3 Statistics analysis and incorporation of derived variables
In this section in order to uncover the physical knowledge for the actual operation
guidance and confirmation statistical analysis is performed for the 16 input manipulated
variables and the eight output variables (ie the temperatures in the four zones) of a
reheating furnace system For the combustion efficiency evaluation and modeling two
types of derived ratio variables are introduced which is helpful to reveal the
information included in the data
31 Correlation analysis
Correlation analysis between the temperatures (T) in each area and all of the air flows
(A) and the gas flows (G) is performed and shown in Fig 3 It can be seen that only the
temperatures in both areas of the soaking zone are the most highly related to the air flow
and the gas flow in its own zone However this phenomenon does not occur in the other
three zones The temperatures in the second heating zone mainly depend on the air flow
and the gas flow in its own zone as well as the nearby first heating zone As for the
preheating zone and first heating zone no apparent correlation can be observed
Obviously these analysis results could not tell us the explicit information about how to
evaluate the efficiency levels and key manipulated variables
9
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
-101
T P-u
AP GP A1 G1 A2 G2 AS GS
AP GP A1 G1 A2 G2 AS GS
AP GP A1 G1 A2 G2 AS GS
-101
T P-d
-101
T 1-u
-101
T 1-d
AP GP A1 G1 A2 G2 AS GS
-101
T 2-u
AP GP A1 G1 A2 G2 AS GS
-101
T 2-d
-101
T S-u
-101
T S-d
AP GP A1 G1 A2 G2 AS GS
AP GP A1 G1 A2 G2 AS GS
AP GP A1 G1 A2 G2 AS GS
UPDOWN
Fig 3 Correlation between air flows gas flows and temperatures
32 Incorporation of derived variables
During the stable heating stage the quantity of heat absorbed and removed from the
slabs from each furnace zone is relatively constant Hence the derived variable TG
ratio can be treated as an index for the combustion efficiency level This is because a
higher TG ratio signifies more combustion heat generated from unit gas ie higher
combustion efficiency
Moreover it is known that the appropriate air and fuel ratio is vital for the
combustion efficiency so the air-gas ratio (AG) is utilized as another derived variable
for the research Again the correlation analysis is performed for two types of derived
variables The correlations between different variables including the AG ratio G and
the TG ratio are shown in Fig 4 It can be clearly seen from the four red rectangle
blocks that TG of each zone is only remarkably related to AG and G in the same zone
10
183
184
185
186
187
188
189
190
191
192
193
194
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
Furthermore as related to the lsquouprsquo or lsquodownrsquo areas in one zone TG in each area has the
highest correlation with AG and G in the same area while AG and G in the opposite
area of the same zone is secondary This can be easily seen from the red and blue color
markings in each red rectangle block Thus it is of great significance to introduce these
derived ratio variables
-101
TG
P-u
AGP GP AG1G1 AG2 G2 AGS GS
-101
TG
P-d
AGP GP AG1 G1 AG2 G2 AGS GS
-101
TG
1-u
AGP GP AG1 G1 AG2 G2 AGS GS
-101
TG
1-d
AGP GP AG1 G1 AG2 G2 AGS GS
-101
TG
2-u
AGP GP AG1 G1 AG2 G2 AGS GS
-101
TG
2-d
AGP GP AG1 G1 AG2 G2 AGS GS
-101
TG
S-u
AGP GP AG1 G1 AG2 G2 AGS GS
-101
TG
S-d
AGP GP AG1 G1 AG2 G2 AGS GS
UPDOWN
Fig 4 Correlation between the air-gas ratio gas flow and TG ratio
33 LDA analysis
Linear discriminant analysis (LDA) aims to finding a projection direction that
maximizes the separation of class means and minimizes the within-class variance [25]
In this section LDA is utilized to identify the discriminating variables that play an
important role in determining combustion efficiency levels All the data are partitioned
into five groups according to their efficiency levels
11
195
196
197
198
199
200
201
202
203
204
205
206
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
Based on a descending order of TG ratios five efficiency levels for the lsquouprsquo area of
the soaking zone are denoted as HH H M L and LL LDA is conducted on the three
groups of the data with levels of HH M and LL
Fig 5 shows the scattering of the LDA projections of the process observations
collected at the three efficiency levels where y1 and y2 correspond to the first two LDA
components that contain most discriminant information The weighting factors ( and
) of the 16 input variables composing the projections y1 and y2 are shown in Fig 6
where and From left to right the 16 input variables are defined as
the eight AG ratio variables and the eight gas flow variables with the order of variables
of each kind P-u P-d 1-u 1-d 2-u 2-d S-u and S-d
-8 -6 -4 -2 0 2 4 6-4
-3
-2
-1
0
1
2
3
4
5
6
Projection y1
Pro
ject
ion
y2
HHMLL
Fig 5 Scattering of LDA projections y1 and y2 for three efficiency levels
12
207
208
209
210
211
212
213
214
215
216
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
AG_S-u-4
-2
0
2
Wei
ghtin
g fa
ctor
for y
1
AG_S-u
-101234
Wei
ghtin
g fa
ctor
for y
2
Fig 6 Weight factors of various input variables for projections y1 and y2
The two figures reveal that the three groups are clearly discriminated by the LDA
projection and the most significant variables for the different efficiency levels are the
derived variables ie the AG in the lsquouprsquo area of the soaking zone A similar result can
be obtained in the lsquodownrsquo area or any area of the other zones Therefore AG in each
heating area is the key manipulated variable that determines the different combustion
efficiency
4 Modeling and prediction of temperature and temperature-gas ratio
For the model-based operation optimization the models for the temperature and TG
ratio based on the NNG algorithm will be developed and compared
41 NNG variable selection algorithm
The NNG method can be generalized into a two-stage shrinkage method In the first
stage the sign for each variable is determined using the ordinary least square procedure
13
217
218
219
220
221
222
223
224
225
226
227
228
229
230
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
and in the second stage the corresponding magnitudes are computed by solving a series
of constrained quadratic programming
A set of observation data is provided where is the input matrix whose
columns represent the measured candidate variables and is the corresponding
vector of the response data The following expression is given with the number of the
response variable being equal to 1 but a similar procedure can be generalized to any
number of variables Let and be normalized to the zero-mean and unit standard
deviation Additionally let be a set of the ordinary least square estimates of the
coefficients of the following linear model then
(1)
The second stage shrinkage can be formulated as the following optimization problem
subject to
(2)
As decreased and the NNG is tightened more of the become zero and the
remaining nonzero coefficients are shrunk A solution path exists with on which
the appropriate shrinkage can be selected Conventionally the v-fold cross-validation is
used to estimate the prediction error and to select the best solution in the solution path
so as to minimize the prediction or model error
42 Modeling and prediction of temperature and TG ratio
The input-output relations change with time as the reheating process develops An
adaptive modeling strategy is often used to resolve time-varying characteristics of
14
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
industrial processes In this paper the NNG-based regression modeling prediction and
optimization is implemented in a moving window manner where the size of the
window and the length of the moving step are selected as 1000 and 100 respectively In
addition the prediction horizon is also set as 100 In each step the NNG regression
model is built based on the data in the current window which is then used for the
prediction and optimization in the subsequent prediction horizon Next the window
moves forward by replacing the oldest 100 samples for model training by the
observations collected in the previous prediction horizon This moving window strategy
is workable because in the reheating process the input-output relation is slowly time-
varying and the model is still valid for the prediction and optimization in the subsequent
short time
The modeling of the temperature T is based on the air flow rates and gas flow rates
while the modeling for the TG ratio is based on the AG ratios and gas flow rates
Taking the lsquouprsquo area of the soaking zone as an example the prediction result for T and
the TG ratio in the lsquouprsquo area of the soaking zone is shown in Figs 7 and 8 respectively
0 02 04 06 08 1 12 14 16 18 2
x 104
1140
1160
1180
1200
1220
1240
1260
1280
1300
1320
Sampling intervals
T S-u (
)
OriginalPredicted
Fig 7 Prediction of temperature in the lsquouprsquo area of the soaking zone
15
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
0 02 04 06 08 1 12 14 16 18 2
x 104
02
04
06
08
1
12
14
16
18
2
Sampling intervals
TG
S-u
OriginalPredicted
Fig 8 Prediction of TG ratio in the lsquouprsquo area of the soaking zone
A comparison between Figs 7 and 8 show that the prediction of the TG ratio is more
accurate than the prediction of temperature The average R2 in 190 NNG regressions is
0935 for the TG ratio while the average R2 is only 0814 for temperature This is
consistent with the statistical analysis which reveals that the correlation between AG
and TG is higher than that between A and T These results indicate that derived
variables are more meaningful for the purpose of prediction and the modeling of the
TG ratio is more appropriate for implementing optimization
The selected frequency for each variable in 190 NNG regressions and the coefficients
of each variable in 20 NNG regressions for the TG ratio modeling of the lsquouprsquo area of
the soaking zone are shown in Figs 9 and 10 respectively Fig 9 shows that the selected
frequency of the variables in the lsquouprsquo area of the soaking zone is much higher than
variables in other zones Similarly as is shown in Fig 10 the NNG regression
coefficients of the variables in the lsquouprsquo area of the soaking zone are much larger than the
variables in the other areas These results indicated that variables in the lsquouprsquo area of
16
269
270
271
272
273
274
275
276
277
278
279
280
281
282
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
soaking zone are mostly contributed to the modeling of the TG ratio in same area
which is quite reasonable Similar results can be obtained for the other areas
0
20
40
60
80
100
120
140
160
180
200
Sel
ecte
d fre
quen
cy
AGP-u AGP-d AG1-u AG1-d AG2-u AG2-d AGS-u AGS-d GP-u GP-d G1-u G1-d G2-u G2-d GS-u GS-d
Fig 9 Variable selected frequency of over 190 runs for the TG ratio model of the lsquouprsquo
area of the soaking zone
0 2 4 6 8 10 12 14 16 18 20-05
05
AG
P-u
0 2 4 6 8 10 12 14 16 18 20-05
05
GP
-u
0 2 4 6 8 10 12 14 16 18 20-1
0
AG
P-d
0 2 4 6 8 10 12 14 16 18 20-1
0
GP
-d
0 2 4 6 8 10 12 14 16 18 20-05
05
AG
1-u
0 2 4 6 8 10 12 14 16 18 20-05
05
G1-
u
0 2 4 6 8 10 12 14 16 18 20-05
05
AG
1-d
0 2 4 6 8 10 12 14 16 18 20-05
05
G1-
d
0 2 4 6 8 10 12 14 16 18 20-1
0
AG
2-u
0 2 4 6 8 10 12 14 16 18 200
1
G2-
u
0 2 4 6 8 10 12 14 16 18 20-1
0
AG
2-d
0 2 4 6 8 10 12 14 16 18 20-1
0
G2-
d
0 2 4 6 8 10 12 14 16 18 200
1
AG
S-u
0 2 4 6 8 10 12 14 16 18 20-1
0
GS
-u
0 2 4 6 8 10 12 14 16 18 20-05
05
AG
S-d
0 2 4 6 8 10 12 14 16 18 20-05
05
GS
-d
Fig 10 Part of the NNG regression coefficients of each variable over 190 runs for the
TG ratio model of the lsquouprsquo area of the soaking zone
It is remarkable that the NNG regression coefficients corresponding to the AG ratio
in the lsquouprsquo area of the soaking zone are consistently positive On the contrary the
coefficients of the gas flow rate in the same area are consistently negative This
indicates that under this condition the rise of the AG leads to the increase of the TG
17
283
284
285
286
287
288
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
while the rise of the G leads to the drop of the TG This property is helpful for
performing optimization and improve combustion efficiency
For comparison the modeling results for two other algorithms artificial neural
network (ANN) and partial least squares (PLS) [26] are compared with the NNG
algorithm in the following three aspects [27]
(1) Model Size the number of variables selected for modeling
(2) Model Magnitude the mean of the L1 norm of the regression coefficients
(3) Prediction Precision the mean squared prediction error (MSPE)
Summary of the algorithm comparison is shown in Table 2 It can be seen that the
superiorities of the NNG regression in model size model magnitude and model
precision are remarkable
Table 2 Summary of algorithm comparison
Index NNG ANN PLS
Model Size 849 16 16
Model Magnitude 10075 -- 12150
MSPE 00093 00108 00140
5 Model-based optimization
51 Implemention of model-based optimization operation
The goal of optimization is to seek an optimal combination of AG and G in the lsquouprsquo
area of the soaking zone so as to minimize the gas consumption while keeping the
temperature at the target value According to the heating schedule the most expected
temperature in the soaking zone is 1200 Therefore the target temperature
is set at this value to achieve the desired heating effect As shown in Fig 11 in the real
18
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
operation the temperature fluctuates around the target value because of the imperfect
control performance However in order to simplify the analysis and compute the
maximum possible energy saving perfect temperature control is assumed when
adopting the model-based optimization strategy In other words it is assumed that the
actual temperature in the lsquouprsquo area of the soaking zone can be adjusted to the expected
temperature ie 1200
0 02 04 06 08 1 12 14 16 18 2
x 104
1150
1200
1250
1300
Sampling intervals
T S-u (
)
OriginalTarget
Fig 11 Original and target temperature in the lsquouprsquo area of the soaking zone
The adjustment scheme takes the maximum value and minimum value of the original
operation data as the upper and lower bounds for the adjustment Moreover in order to
assure the validity of the linear model the increment or decrement of AG and G should
not beyond of the original value (considered as 10 for the purposes of this paper)
The adjustment strategy can be formulated as
19
307
308
309
310
311
312
313
314
315
316
317
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
(3)
where NNG() denotes the NNG regression model of the TG ratio and are the
NNG regression coefficients of the AG and G in the lsquouprsquo area of the soaking zone for
modeling TG in same area
With the model-based strategy and adaptive model of the TG ratio given in Section
42 the optimization adjustment results can be obtained as follows
The adjustment of gas flow in the lsquouprsquo area of the soaking zone shown in Fig 12
illustrates that the implementation of the model-based optimization operation can reduce
the consumption of the gas flow Compared with the original case 938 of the gas can
be saved on average by utilizing the model-based adjustment
0 02 04 06 08 1 12 14 16 18 2
x 104
-800
-700
-600
-500
-400
-300
-200
-100
0
Sampling intervals
Adj
ustm
ent o
f gas
flow
m(
3 h)
20
318
319
320
321
322
323
324
325
326
327
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
(a) Adjustment amount of gas flow
0 02 04 06 08 1 12 14 16 18 2
x 104
500
1000
1500
2000
2500
3000
3500
4000
4500
5000G
as fl
owm
(3 h
)
Sampling intervals
OriginalAdjusted
(b) Comparison of original and adjusted gas flow
095 096 097 098 099 1 101 102 103 104 105
x 104
500
1000
1500
2000
2500
3000
3500
4000
4500
5000
Gas
flow
m(
3 h)
Sampling intervals
OriginalAdjusted
(c) Comparison of original and adjusted gas flow in interval [9501-10500]
Fig 12 Adjustment of gas flow in the lsquouprsquo area of the soaking zone
The adjustment of the air flow in the lsquouprsquo area of the soaking zone given in Fig 13
shows that the air flow is reduced in most cases and only increased for a few cases In
21
328
329
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
general 681 of the air flow is reduced after the implementation of the model-based
optimization
0 02 04 06 08 1 12 14 16 18 2
x 104
-2000
-1500
-1000
-500
0
500
1000
Sampling intervals
Adj
ustm
ent o
f air
flow
(m3 h
)
(a) Adjustment amount of air flow
0 02 04 06 08 1 12 14 16 18 2
x 104
4000
5000
6000
7000
8000
9000
10000
Air
flow
m(
3 h)
Sampling intervals
OriginalAdjusted
(b) Comparison of original and adjusted air flow
22
330
331
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
095 096 097 098 099 1 101 102 103 104 105
x 104
4000
5000
6000
7000
8000
9000
10000
Air
flow
m(
3 h)
Sampling intervals
OriginalAdjusted
(c) Comparison of original and adjusted air flow in interval [9501-10500]
Fig 13 Adjustment of air flow in the lsquouprsquo area of the soaking zone
52 Discussions
In this section detailed analysis for the optimization operation results is provided
The statistics of optimization at the boundary conditions shown in Table 3 illustrates
that the adjusted value for G reaches its lower limit in most cases (8045 for
and 687 for ) which is the most energy-efficient point In a number of cases
(1079 for and 001 for ) the adjusted AG reaches its upper
limit These results indicate that the optimization operations maximize the combustion
efficiency by decreasing G and increasing the AG ratio up to the boundary conditions
However the lower limit is also unexpectedly attained in a small number of cases
(110 for and 078 for ) for which a more detailed analysis is
given below
23
332
333
334
335
336
337
338
339
340
341
342
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
Table 3 Statistics of optimization at boundary conditions
Total
Amount 0 1374 0 16090 17464
Percentag
e 0 687 0 8045
8732
Total
Amount 2 220 2158 156 2536
Percentag
e 001 110 1079 078
1268
Table 4 Statistics for different optimization operations
Adjustment
Gdarr
Guarr TotalAGuarr AGdarr
Tdarr Tuarr Tdarr Tuarr
Amount 9548 3744 6328 380 0 20000
Percentage 4774 1872 3164 190 0 100
The statistics for different optimization operations is performed and shown in Table
4 It can be seen that all the adjusted operations result in the reduction of gas
consumption In over half the cases (4774 with decreased temperature and 1872
with increased temperature) AG is adjusted to higher levels Meanwhile a number of
the adjustments (3164) lower the temperature by reducing the G and AG at the same
time This result indicates that the original temperature cannot be adjusted to the target
by only reducing G in the constraint conditions The remaining cases (19) are
24
343
344
345
346
347
348
349
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
relatively special in which an excess of air is supplied in the original operation and the
adjusted operation thereby increasing the temperature by a smaller AG with less gas
consumption
6 Conclusion
Based on the actual operation data this paper aims to explore the improvement of the
combustion efficiency and the room for energy conservation Correlation analysis and
LDA show that it is of great significance to introduce two derived ratio variables which
are the AG ratio and the TG ratio A type of combustion efficiency utilizing an on-line
soft sensor is put forward by employing a NNG variable selection algorithm which
provides a good solution to the combustion efficiency real-time measurement problem
of a reheating furnace The implementation of the model-based optimization is studied
based on the actual operational data Detailed analysis for the optimization results is
given for the different cases The results show that significant energy conservation can
be achieved when the furnace operation is optimized by using the developed soft sensor
model
Acknowledgement
The authors would like to thank the financial support provided by the National Nature
Science Foundation of China under Grant 61171145 Y Yao was supported by Ministry
of Science amp Technology ROC under Grant number MOST 104-2221-E-007-129
References
25
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[1] Z J Wang Q D Wu and T Y Chai Optimal-setting control for complicated
industrial processes and its application studyControl Engineering Practice vol
12 pp 65-74 2004
[2] A Steinboeck K Graichen and A Kugi Dynamic Optimization of a Slab
Reheating Furnace With Consistent Approximation of Control VariablesIEEE
Transactions on Control Systems Technology vol 19 pp 1444-1456 2011
[3] B T Zhang C Y Wang Q Qin and L Li Influence of Boiler Combustion
Adjustment on NOxEmission and Boiler EfficiencyAdvanced Materials
Research vol 732-733 pp 234-237 2013
[4] C K Yoo and IB Lee Soft Sensor and Adaptive Model-Based Dissolved
Oxygen Control for Biological Wastewater Treatment ProcessesEnvironmental
Engineering Science vol 21 pp 331-340 2004
[5] S A Bhat D N Saraf S Gupta and S K Gupta Use of Agitator Power as a
Soft Sensor for Bulk Free-Radical Polymerization of Methyl Methacrylate in
Batch ReactorsIndustrial amp Engineering Chemistry Research vol 45 pp 4243-
4255 2006
[6] K Desai Y Badhe S S Tambe and B D Kulkarni Soft-sensor development
for fed-batch bioreactors using support vector regressionBiochemical
Engineering Journal vol 27 pp 225-239 2006
[7] Y P Badhe Lonari J Tambe S S amp Kulkarni B D Improve polyethylene
process control and product qualityHydrocarbon Processing vol 86 pp 53-60
2007
26
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[8] N K Nath K Mandal A K Singh B Basu C Bhanu S Kumar et al Ladle
furnace on-line reckoner for prediction and control of steel temperature and
compositionIronmaking amp Steelmaking vol 33 pp 140-150 2006
[9] A J Yan T Y Chai F H Wu and P Wang Hybrid intelligent control of
combustion process for ore-roasting furnaceJournal of Control Theory and
Applications vol 6 pp 80-85 2008
[10] J Li W M Zhong H Cheng X D Kong and F Qian A data-driven soft
sensor modeling for furnace temperature of Opposed Multi-Burner gasifier pp
705-710 2011
[11] Y H Yang Y H Liu X Z Liu and S K Qin Billet temperature soft sensor
model of reheating furnace based on RVM method pp 4003-4006 2011
[12] J H Wang C Wang X F Zhu and X K Fang Application of soft sensor in
welding seam tracking prediction based on LSSVM and PSO with compression
factor pp 2441-2446 2013
[13] L Balbis J Balderud and M J Grimble Nonlinear predictive control of steel
slab reheating furnace pp 1679-1684 2008
[14] A Steinboeck D Wild T Kiefer and A Kugi A mathematical model of a slab
reheating furnace with radiative heat transfer and non-participating gaseous
media International Journal Of Heat And Mass Transfer vol 53 pp 5933-
5946 Dec 2010
[15] A Steinboeck D Wild and A Kugi Nonlinear model predictive control of a
continuous slab reheating furnace Control Engineering Practice vol 21 pp
495-508 2013
27
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[16] C Zhang T Ishii and S Sugiyama Numerical Modeling Of the Thermal
Performance Of Regenerative Slab Reheat Furnaces Numerical Heat Transfer
Part A Applications vol 32 pp 613-631 1997
[17] J G Kim and K Y Huh Prediction of Transient Slab Temperature Distribution
in the Re-heating Furnace of a Walking-beam Type for Rolling of Steel Slabs
ISIJ International vol 40 pp 1115-1123 2000
[18] J G Kim K Y and H I T K Three-Dimensional Analysis Of the Walking-
Beam-Type Slab Reheating Furnace In Hot Strip Mills Numerical Heat
Transfer Part A Applications vol 38 pp 589-609 2000
[19] CT Hsieh MJ Huang ST Lee and CH Wang Numerical Modeling of a
Walking-Beam-Type Slab Reheating Furnace Numerical Heat Transfer Part A
Applications vol 53 pp 966-981 2008
[20] MJ Huang CT Hsieh ST Lee and CH Wang A Coupled Numerical Study
of Slab Temperature and Gas Temperature in the Walking-Beam-Type Slab
Reheating Furnace Numerical Heat Transfer Part A Applications vol 54 pp
625-646 2008
[21] Z Song and A Kusiak Constraint-Based Control of Boiler Efficiency A Data-
Mining Approach IEEE Transactions on Industrial Informatics vol 3 pp 73-
83 2007
[22] J Q Li J J Gu and C L Niu The Operation Optimization based on
Correlation Analysis of Operation Parameters in Power Plant pp 138-141
2008
28
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[23] J G Wang S S Shieh S S Jang D S H Wong and C W Wu A two-tier
approach to the data-driven modeling on thermal efficiency of a BFGcoal co-
firing boiler Fuel vol 111 pp 528-534 Sep 2013
[24] L Breiman Better Subset Regression Using the Nonnegative Garrote
Technometrics vol 37 pp 373-384 1995
[25] R A Fisher The use of multiple measurements in taxonomic problems
AnnHum Genet vol 7 pp 179-188 1936
[26] J Liu Developing a soft sensor based on sparse partial least squares with
variable selection Journal of Process Control vol 24 pp 1046-1056 2014
[27] K Sun J Liu JL Kang SS Jang D SH Wong and DS Chen
Development of a variable selection method for soft sensor using artificial
neural network and nonnegative garrote Journal of Process Control vol 24 pp
1068-1075 2014
29
439
440
441
442
443
444
445
446
447
448
449
450
451
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
lsquouprsquo area and any part of lsquodownrsquo area does not exceed 05 Therefore for the reduction
of the data dimension the three temperatures in the lsquouprsquo area or the lsquodownrsquo area can be
treated as only one variable which can be taken as the mean value or the first principal
component acquired from the PCA analysis Considering the reservation of the variable
physical meaning the former is preferred and used
3 Statistics analysis and incorporation of derived variables
In this section in order to uncover the physical knowledge for the actual operation
guidance and confirmation statistical analysis is performed for the 16 input manipulated
variables and the eight output variables (ie the temperatures in the four zones) of a
reheating furnace system For the combustion efficiency evaluation and modeling two
types of derived ratio variables are introduced which is helpful to reveal the
information included in the data
31 Correlation analysis
Correlation analysis between the temperatures (T) in each area and all of the air flows
(A) and the gas flows (G) is performed and shown in Fig 3 It can be seen that only the
temperatures in both areas of the soaking zone are the most highly related to the air flow
and the gas flow in its own zone However this phenomenon does not occur in the other
three zones The temperatures in the second heating zone mainly depend on the air flow
and the gas flow in its own zone as well as the nearby first heating zone As for the
preheating zone and first heating zone no apparent correlation can be observed
Obviously these analysis results could not tell us the explicit information about how to
evaluate the efficiency levels and key manipulated variables
9
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
-101
T P-u
AP GP A1 G1 A2 G2 AS GS
AP GP A1 G1 A2 G2 AS GS
AP GP A1 G1 A2 G2 AS GS
-101
T P-d
-101
T 1-u
-101
T 1-d
AP GP A1 G1 A2 G2 AS GS
-101
T 2-u
AP GP A1 G1 A2 G2 AS GS
-101
T 2-d
-101
T S-u
-101
T S-d
AP GP A1 G1 A2 G2 AS GS
AP GP A1 G1 A2 G2 AS GS
AP GP A1 G1 A2 G2 AS GS
UPDOWN
Fig 3 Correlation between air flows gas flows and temperatures
32 Incorporation of derived variables
During the stable heating stage the quantity of heat absorbed and removed from the
slabs from each furnace zone is relatively constant Hence the derived variable TG
ratio can be treated as an index for the combustion efficiency level This is because a
higher TG ratio signifies more combustion heat generated from unit gas ie higher
combustion efficiency
Moreover it is known that the appropriate air and fuel ratio is vital for the
combustion efficiency so the air-gas ratio (AG) is utilized as another derived variable
for the research Again the correlation analysis is performed for two types of derived
variables The correlations between different variables including the AG ratio G and
the TG ratio are shown in Fig 4 It can be clearly seen from the four red rectangle
blocks that TG of each zone is only remarkably related to AG and G in the same zone
10
183
184
185
186
187
188
189
190
191
192
193
194
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
Furthermore as related to the lsquouprsquo or lsquodownrsquo areas in one zone TG in each area has the
highest correlation with AG and G in the same area while AG and G in the opposite
area of the same zone is secondary This can be easily seen from the red and blue color
markings in each red rectangle block Thus it is of great significance to introduce these
derived ratio variables
-101
TG
P-u
AGP GP AG1G1 AG2 G2 AGS GS
-101
TG
P-d
AGP GP AG1 G1 AG2 G2 AGS GS
-101
TG
1-u
AGP GP AG1 G1 AG2 G2 AGS GS
-101
TG
1-d
AGP GP AG1 G1 AG2 G2 AGS GS
-101
TG
2-u
AGP GP AG1 G1 AG2 G2 AGS GS
-101
TG
2-d
AGP GP AG1 G1 AG2 G2 AGS GS
-101
TG
S-u
AGP GP AG1 G1 AG2 G2 AGS GS
-101
TG
S-d
AGP GP AG1 G1 AG2 G2 AGS GS
UPDOWN
Fig 4 Correlation between the air-gas ratio gas flow and TG ratio
33 LDA analysis
Linear discriminant analysis (LDA) aims to finding a projection direction that
maximizes the separation of class means and minimizes the within-class variance [25]
In this section LDA is utilized to identify the discriminating variables that play an
important role in determining combustion efficiency levels All the data are partitioned
into five groups according to their efficiency levels
11
195
196
197
198
199
200
201
202
203
204
205
206
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
Based on a descending order of TG ratios five efficiency levels for the lsquouprsquo area of
the soaking zone are denoted as HH H M L and LL LDA is conducted on the three
groups of the data with levels of HH M and LL
Fig 5 shows the scattering of the LDA projections of the process observations
collected at the three efficiency levels where y1 and y2 correspond to the first two LDA
components that contain most discriminant information The weighting factors ( and
) of the 16 input variables composing the projections y1 and y2 are shown in Fig 6
where and From left to right the 16 input variables are defined as
the eight AG ratio variables and the eight gas flow variables with the order of variables
of each kind P-u P-d 1-u 1-d 2-u 2-d S-u and S-d
-8 -6 -4 -2 0 2 4 6-4
-3
-2
-1
0
1
2
3
4
5
6
Projection y1
Pro
ject
ion
y2
HHMLL
Fig 5 Scattering of LDA projections y1 and y2 for three efficiency levels
12
207
208
209
210
211
212
213
214
215
216
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
AG_S-u-4
-2
0
2
Wei
ghtin
g fa
ctor
for y
1
AG_S-u
-101234
Wei
ghtin
g fa
ctor
for y
2
Fig 6 Weight factors of various input variables for projections y1 and y2
The two figures reveal that the three groups are clearly discriminated by the LDA
projection and the most significant variables for the different efficiency levels are the
derived variables ie the AG in the lsquouprsquo area of the soaking zone A similar result can
be obtained in the lsquodownrsquo area or any area of the other zones Therefore AG in each
heating area is the key manipulated variable that determines the different combustion
efficiency
4 Modeling and prediction of temperature and temperature-gas ratio
For the model-based operation optimization the models for the temperature and TG
ratio based on the NNG algorithm will be developed and compared
41 NNG variable selection algorithm
The NNG method can be generalized into a two-stage shrinkage method In the first
stage the sign for each variable is determined using the ordinary least square procedure
13
217
218
219
220
221
222
223
224
225
226
227
228
229
230
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
and in the second stage the corresponding magnitudes are computed by solving a series
of constrained quadratic programming
A set of observation data is provided where is the input matrix whose
columns represent the measured candidate variables and is the corresponding
vector of the response data The following expression is given with the number of the
response variable being equal to 1 but a similar procedure can be generalized to any
number of variables Let and be normalized to the zero-mean and unit standard
deviation Additionally let be a set of the ordinary least square estimates of the
coefficients of the following linear model then
(1)
The second stage shrinkage can be formulated as the following optimization problem
subject to
(2)
As decreased and the NNG is tightened more of the become zero and the
remaining nonzero coefficients are shrunk A solution path exists with on which
the appropriate shrinkage can be selected Conventionally the v-fold cross-validation is
used to estimate the prediction error and to select the best solution in the solution path
so as to minimize the prediction or model error
42 Modeling and prediction of temperature and TG ratio
The input-output relations change with time as the reheating process develops An
adaptive modeling strategy is often used to resolve time-varying characteristics of
14
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
industrial processes In this paper the NNG-based regression modeling prediction and
optimization is implemented in a moving window manner where the size of the
window and the length of the moving step are selected as 1000 and 100 respectively In
addition the prediction horizon is also set as 100 In each step the NNG regression
model is built based on the data in the current window which is then used for the
prediction and optimization in the subsequent prediction horizon Next the window
moves forward by replacing the oldest 100 samples for model training by the
observations collected in the previous prediction horizon This moving window strategy
is workable because in the reheating process the input-output relation is slowly time-
varying and the model is still valid for the prediction and optimization in the subsequent
short time
The modeling of the temperature T is based on the air flow rates and gas flow rates
while the modeling for the TG ratio is based on the AG ratios and gas flow rates
Taking the lsquouprsquo area of the soaking zone as an example the prediction result for T and
the TG ratio in the lsquouprsquo area of the soaking zone is shown in Figs 7 and 8 respectively
0 02 04 06 08 1 12 14 16 18 2
x 104
1140
1160
1180
1200
1220
1240
1260
1280
1300
1320
Sampling intervals
T S-u (
)
OriginalPredicted
Fig 7 Prediction of temperature in the lsquouprsquo area of the soaking zone
15
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
0 02 04 06 08 1 12 14 16 18 2
x 104
02
04
06
08
1
12
14
16
18
2
Sampling intervals
TG
S-u
OriginalPredicted
Fig 8 Prediction of TG ratio in the lsquouprsquo area of the soaking zone
A comparison between Figs 7 and 8 show that the prediction of the TG ratio is more
accurate than the prediction of temperature The average R2 in 190 NNG regressions is
0935 for the TG ratio while the average R2 is only 0814 for temperature This is
consistent with the statistical analysis which reveals that the correlation between AG
and TG is higher than that between A and T These results indicate that derived
variables are more meaningful for the purpose of prediction and the modeling of the
TG ratio is more appropriate for implementing optimization
The selected frequency for each variable in 190 NNG regressions and the coefficients
of each variable in 20 NNG regressions for the TG ratio modeling of the lsquouprsquo area of
the soaking zone are shown in Figs 9 and 10 respectively Fig 9 shows that the selected
frequency of the variables in the lsquouprsquo area of the soaking zone is much higher than
variables in other zones Similarly as is shown in Fig 10 the NNG regression
coefficients of the variables in the lsquouprsquo area of the soaking zone are much larger than the
variables in the other areas These results indicated that variables in the lsquouprsquo area of
16
269
270
271
272
273
274
275
276
277
278
279
280
281
282
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
soaking zone are mostly contributed to the modeling of the TG ratio in same area
which is quite reasonable Similar results can be obtained for the other areas
0
20
40
60
80
100
120
140
160
180
200
Sel
ecte
d fre
quen
cy
AGP-u AGP-d AG1-u AG1-d AG2-u AG2-d AGS-u AGS-d GP-u GP-d G1-u G1-d G2-u G2-d GS-u GS-d
Fig 9 Variable selected frequency of over 190 runs for the TG ratio model of the lsquouprsquo
area of the soaking zone
0 2 4 6 8 10 12 14 16 18 20-05
05
AG
P-u
0 2 4 6 8 10 12 14 16 18 20-05
05
GP
-u
0 2 4 6 8 10 12 14 16 18 20-1
0
AG
P-d
0 2 4 6 8 10 12 14 16 18 20-1
0
GP
-d
0 2 4 6 8 10 12 14 16 18 20-05
05
AG
1-u
0 2 4 6 8 10 12 14 16 18 20-05
05
G1-
u
0 2 4 6 8 10 12 14 16 18 20-05
05
AG
1-d
0 2 4 6 8 10 12 14 16 18 20-05
05
G1-
d
0 2 4 6 8 10 12 14 16 18 20-1
0
AG
2-u
0 2 4 6 8 10 12 14 16 18 200
1
G2-
u
0 2 4 6 8 10 12 14 16 18 20-1
0
AG
2-d
0 2 4 6 8 10 12 14 16 18 20-1
0
G2-
d
0 2 4 6 8 10 12 14 16 18 200
1
AG
S-u
0 2 4 6 8 10 12 14 16 18 20-1
0
GS
-u
0 2 4 6 8 10 12 14 16 18 20-05
05
AG
S-d
0 2 4 6 8 10 12 14 16 18 20-05
05
GS
-d
Fig 10 Part of the NNG regression coefficients of each variable over 190 runs for the
TG ratio model of the lsquouprsquo area of the soaking zone
It is remarkable that the NNG regression coefficients corresponding to the AG ratio
in the lsquouprsquo area of the soaking zone are consistently positive On the contrary the
coefficients of the gas flow rate in the same area are consistently negative This
indicates that under this condition the rise of the AG leads to the increase of the TG
17
283
284
285
286
287
288
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
while the rise of the G leads to the drop of the TG This property is helpful for
performing optimization and improve combustion efficiency
For comparison the modeling results for two other algorithms artificial neural
network (ANN) and partial least squares (PLS) [26] are compared with the NNG
algorithm in the following three aspects [27]
(1) Model Size the number of variables selected for modeling
(2) Model Magnitude the mean of the L1 norm of the regression coefficients
(3) Prediction Precision the mean squared prediction error (MSPE)
Summary of the algorithm comparison is shown in Table 2 It can be seen that the
superiorities of the NNG regression in model size model magnitude and model
precision are remarkable
Table 2 Summary of algorithm comparison
Index NNG ANN PLS
Model Size 849 16 16
Model Magnitude 10075 -- 12150
MSPE 00093 00108 00140
5 Model-based optimization
51 Implemention of model-based optimization operation
The goal of optimization is to seek an optimal combination of AG and G in the lsquouprsquo
area of the soaking zone so as to minimize the gas consumption while keeping the
temperature at the target value According to the heating schedule the most expected
temperature in the soaking zone is 1200 Therefore the target temperature
is set at this value to achieve the desired heating effect As shown in Fig 11 in the real
18
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
operation the temperature fluctuates around the target value because of the imperfect
control performance However in order to simplify the analysis and compute the
maximum possible energy saving perfect temperature control is assumed when
adopting the model-based optimization strategy In other words it is assumed that the
actual temperature in the lsquouprsquo area of the soaking zone can be adjusted to the expected
temperature ie 1200
0 02 04 06 08 1 12 14 16 18 2
x 104
1150
1200
1250
1300
Sampling intervals
T S-u (
)
OriginalTarget
Fig 11 Original and target temperature in the lsquouprsquo area of the soaking zone
The adjustment scheme takes the maximum value and minimum value of the original
operation data as the upper and lower bounds for the adjustment Moreover in order to
assure the validity of the linear model the increment or decrement of AG and G should
not beyond of the original value (considered as 10 for the purposes of this paper)
The adjustment strategy can be formulated as
19
307
308
309
310
311
312
313
314
315
316
317
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
(3)
where NNG() denotes the NNG regression model of the TG ratio and are the
NNG regression coefficients of the AG and G in the lsquouprsquo area of the soaking zone for
modeling TG in same area
With the model-based strategy and adaptive model of the TG ratio given in Section
42 the optimization adjustment results can be obtained as follows
The adjustment of gas flow in the lsquouprsquo area of the soaking zone shown in Fig 12
illustrates that the implementation of the model-based optimization operation can reduce
the consumption of the gas flow Compared with the original case 938 of the gas can
be saved on average by utilizing the model-based adjustment
0 02 04 06 08 1 12 14 16 18 2
x 104
-800
-700
-600
-500
-400
-300
-200
-100
0
Sampling intervals
Adj
ustm
ent o
f gas
flow
m(
3 h)
20
318
319
320
321
322
323
324
325
326
327
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
(a) Adjustment amount of gas flow
0 02 04 06 08 1 12 14 16 18 2
x 104
500
1000
1500
2000
2500
3000
3500
4000
4500
5000G
as fl
owm
(3 h
)
Sampling intervals
OriginalAdjusted
(b) Comparison of original and adjusted gas flow
095 096 097 098 099 1 101 102 103 104 105
x 104
500
1000
1500
2000
2500
3000
3500
4000
4500
5000
Gas
flow
m(
3 h)
Sampling intervals
OriginalAdjusted
(c) Comparison of original and adjusted gas flow in interval [9501-10500]
Fig 12 Adjustment of gas flow in the lsquouprsquo area of the soaking zone
The adjustment of the air flow in the lsquouprsquo area of the soaking zone given in Fig 13
shows that the air flow is reduced in most cases and only increased for a few cases In
21
328
329
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
general 681 of the air flow is reduced after the implementation of the model-based
optimization
0 02 04 06 08 1 12 14 16 18 2
x 104
-2000
-1500
-1000
-500
0
500
1000
Sampling intervals
Adj
ustm
ent o
f air
flow
(m3 h
)
(a) Adjustment amount of air flow
0 02 04 06 08 1 12 14 16 18 2
x 104
4000
5000
6000
7000
8000
9000
10000
Air
flow
m(
3 h)
Sampling intervals
OriginalAdjusted
(b) Comparison of original and adjusted air flow
22
330
331
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
095 096 097 098 099 1 101 102 103 104 105
x 104
4000
5000
6000
7000
8000
9000
10000
Air
flow
m(
3 h)
Sampling intervals
OriginalAdjusted
(c) Comparison of original and adjusted air flow in interval [9501-10500]
Fig 13 Adjustment of air flow in the lsquouprsquo area of the soaking zone
52 Discussions
In this section detailed analysis for the optimization operation results is provided
The statistics of optimization at the boundary conditions shown in Table 3 illustrates
that the adjusted value for G reaches its lower limit in most cases (8045 for
and 687 for ) which is the most energy-efficient point In a number of cases
(1079 for and 001 for ) the adjusted AG reaches its upper
limit These results indicate that the optimization operations maximize the combustion
efficiency by decreasing G and increasing the AG ratio up to the boundary conditions
However the lower limit is also unexpectedly attained in a small number of cases
(110 for and 078 for ) for which a more detailed analysis is
given below
23
332
333
334
335
336
337
338
339
340
341
342
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
Table 3 Statistics of optimization at boundary conditions
Total
Amount 0 1374 0 16090 17464
Percentag
e 0 687 0 8045
8732
Total
Amount 2 220 2158 156 2536
Percentag
e 001 110 1079 078
1268
Table 4 Statistics for different optimization operations
Adjustment
Gdarr
Guarr TotalAGuarr AGdarr
Tdarr Tuarr Tdarr Tuarr
Amount 9548 3744 6328 380 0 20000
Percentage 4774 1872 3164 190 0 100
The statistics for different optimization operations is performed and shown in Table
4 It can be seen that all the adjusted operations result in the reduction of gas
consumption In over half the cases (4774 with decreased temperature and 1872
with increased temperature) AG is adjusted to higher levels Meanwhile a number of
the adjustments (3164) lower the temperature by reducing the G and AG at the same
time This result indicates that the original temperature cannot be adjusted to the target
by only reducing G in the constraint conditions The remaining cases (19) are
24
343
344
345
346
347
348
349
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
relatively special in which an excess of air is supplied in the original operation and the
adjusted operation thereby increasing the temperature by a smaller AG with less gas
consumption
6 Conclusion
Based on the actual operation data this paper aims to explore the improvement of the
combustion efficiency and the room for energy conservation Correlation analysis and
LDA show that it is of great significance to introduce two derived ratio variables which
are the AG ratio and the TG ratio A type of combustion efficiency utilizing an on-line
soft sensor is put forward by employing a NNG variable selection algorithm which
provides a good solution to the combustion efficiency real-time measurement problem
of a reheating furnace The implementation of the model-based optimization is studied
based on the actual operational data Detailed analysis for the optimization results is
given for the different cases The results show that significant energy conservation can
be achieved when the furnace operation is optimized by using the developed soft sensor
model
Acknowledgement
The authors would like to thank the financial support provided by the National Nature
Science Foundation of China under Grant 61171145 Y Yao was supported by Ministry
of Science amp Technology ROC under Grant number MOST 104-2221-E-007-129
References
25
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[1] Z J Wang Q D Wu and T Y Chai Optimal-setting control for complicated
industrial processes and its application studyControl Engineering Practice vol
12 pp 65-74 2004
[2] A Steinboeck K Graichen and A Kugi Dynamic Optimization of a Slab
Reheating Furnace With Consistent Approximation of Control VariablesIEEE
Transactions on Control Systems Technology vol 19 pp 1444-1456 2011
[3] B T Zhang C Y Wang Q Qin and L Li Influence of Boiler Combustion
Adjustment on NOxEmission and Boiler EfficiencyAdvanced Materials
Research vol 732-733 pp 234-237 2013
[4] C K Yoo and IB Lee Soft Sensor and Adaptive Model-Based Dissolved
Oxygen Control for Biological Wastewater Treatment ProcessesEnvironmental
Engineering Science vol 21 pp 331-340 2004
[5] S A Bhat D N Saraf S Gupta and S K Gupta Use of Agitator Power as a
Soft Sensor for Bulk Free-Radical Polymerization of Methyl Methacrylate in
Batch ReactorsIndustrial amp Engineering Chemistry Research vol 45 pp 4243-
4255 2006
[6] K Desai Y Badhe S S Tambe and B D Kulkarni Soft-sensor development
for fed-batch bioreactors using support vector regressionBiochemical
Engineering Journal vol 27 pp 225-239 2006
[7] Y P Badhe Lonari J Tambe S S amp Kulkarni B D Improve polyethylene
process control and product qualityHydrocarbon Processing vol 86 pp 53-60
2007
26
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[8] N K Nath K Mandal A K Singh B Basu C Bhanu S Kumar et al Ladle
furnace on-line reckoner for prediction and control of steel temperature and
compositionIronmaking amp Steelmaking vol 33 pp 140-150 2006
[9] A J Yan T Y Chai F H Wu and P Wang Hybrid intelligent control of
combustion process for ore-roasting furnaceJournal of Control Theory and
Applications vol 6 pp 80-85 2008
[10] J Li W M Zhong H Cheng X D Kong and F Qian A data-driven soft
sensor modeling for furnace temperature of Opposed Multi-Burner gasifier pp
705-710 2011
[11] Y H Yang Y H Liu X Z Liu and S K Qin Billet temperature soft sensor
model of reheating furnace based on RVM method pp 4003-4006 2011
[12] J H Wang C Wang X F Zhu and X K Fang Application of soft sensor in
welding seam tracking prediction based on LSSVM and PSO with compression
factor pp 2441-2446 2013
[13] L Balbis J Balderud and M J Grimble Nonlinear predictive control of steel
slab reheating furnace pp 1679-1684 2008
[14] A Steinboeck D Wild T Kiefer and A Kugi A mathematical model of a slab
reheating furnace with radiative heat transfer and non-participating gaseous
media International Journal Of Heat And Mass Transfer vol 53 pp 5933-
5946 Dec 2010
[15] A Steinboeck D Wild and A Kugi Nonlinear model predictive control of a
continuous slab reheating furnace Control Engineering Practice vol 21 pp
495-508 2013
27
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[16] C Zhang T Ishii and S Sugiyama Numerical Modeling Of the Thermal
Performance Of Regenerative Slab Reheat Furnaces Numerical Heat Transfer
Part A Applications vol 32 pp 613-631 1997
[17] J G Kim and K Y Huh Prediction of Transient Slab Temperature Distribution
in the Re-heating Furnace of a Walking-beam Type for Rolling of Steel Slabs
ISIJ International vol 40 pp 1115-1123 2000
[18] J G Kim K Y and H I T K Three-Dimensional Analysis Of the Walking-
Beam-Type Slab Reheating Furnace In Hot Strip Mills Numerical Heat
Transfer Part A Applications vol 38 pp 589-609 2000
[19] CT Hsieh MJ Huang ST Lee and CH Wang Numerical Modeling of a
Walking-Beam-Type Slab Reheating Furnace Numerical Heat Transfer Part A
Applications vol 53 pp 966-981 2008
[20] MJ Huang CT Hsieh ST Lee and CH Wang A Coupled Numerical Study
of Slab Temperature and Gas Temperature in the Walking-Beam-Type Slab
Reheating Furnace Numerical Heat Transfer Part A Applications vol 54 pp
625-646 2008
[21] Z Song and A Kusiak Constraint-Based Control of Boiler Efficiency A Data-
Mining Approach IEEE Transactions on Industrial Informatics vol 3 pp 73-
83 2007
[22] J Q Li J J Gu and C L Niu The Operation Optimization based on
Correlation Analysis of Operation Parameters in Power Plant pp 138-141
2008
28
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[23] J G Wang S S Shieh S S Jang D S H Wong and C W Wu A two-tier
approach to the data-driven modeling on thermal efficiency of a BFGcoal co-
firing boiler Fuel vol 111 pp 528-534 Sep 2013
[24] L Breiman Better Subset Regression Using the Nonnegative Garrote
Technometrics vol 37 pp 373-384 1995
[25] R A Fisher The use of multiple measurements in taxonomic problems
AnnHum Genet vol 7 pp 179-188 1936
[26] J Liu Developing a soft sensor based on sparse partial least squares with
variable selection Journal of Process Control vol 24 pp 1046-1056 2014
[27] K Sun J Liu JL Kang SS Jang D SH Wong and DS Chen
Development of a variable selection method for soft sensor using artificial
neural network and nonnegative garrote Journal of Process Control vol 24 pp
1068-1075 2014
29
439
440
441
442
443
444
445
446
447
448
449
450
451
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
-101
T P-u
AP GP A1 G1 A2 G2 AS GS
AP GP A1 G1 A2 G2 AS GS
AP GP A1 G1 A2 G2 AS GS
-101
T P-d
-101
T 1-u
-101
T 1-d
AP GP A1 G1 A2 G2 AS GS
-101
T 2-u
AP GP A1 G1 A2 G2 AS GS
-101
T 2-d
-101
T S-u
-101
T S-d
AP GP A1 G1 A2 G2 AS GS
AP GP A1 G1 A2 G2 AS GS
AP GP A1 G1 A2 G2 AS GS
UPDOWN
Fig 3 Correlation between air flows gas flows and temperatures
32 Incorporation of derived variables
During the stable heating stage the quantity of heat absorbed and removed from the
slabs from each furnace zone is relatively constant Hence the derived variable TG
ratio can be treated as an index for the combustion efficiency level This is because a
higher TG ratio signifies more combustion heat generated from unit gas ie higher
combustion efficiency
Moreover it is known that the appropriate air and fuel ratio is vital for the
combustion efficiency so the air-gas ratio (AG) is utilized as another derived variable
for the research Again the correlation analysis is performed for two types of derived
variables The correlations between different variables including the AG ratio G and
the TG ratio are shown in Fig 4 It can be clearly seen from the four red rectangle
blocks that TG of each zone is only remarkably related to AG and G in the same zone
10
183
184
185
186
187
188
189
190
191
192
193
194
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
Furthermore as related to the lsquouprsquo or lsquodownrsquo areas in one zone TG in each area has the
highest correlation with AG and G in the same area while AG and G in the opposite
area of the same zone is secondary This can be easily seen from the red and blue color
markings in each red rectangle block Thus it is of great significance to introduce these
derived ratio variables
-101
TG
P-u
AGP GP AG1G1 AG2 G2 AGS GS
-101
TG
P-d
AGP GP AG1 G1 AG2 G2 AGS GS
-101
TG
1-u
AGP GP AG1 G1 AG2 G2 AGS GS
-101
TG
1-d
AGP GP AG1 G1 AG2 G2 AGS GS
-101
TG
2-u
AGP GP AG1 G1 AG2 G2 AGS GS
-101
TG
2-d
AGP GP AG1 G1 AG2 G2 AGS GS
-101
TG
S-u
AGP GP AG1 G1 AG2 G2 AGS GS
-101
TG
S-d
AGP GP AG1 G1 AG2 G2 AGS GS
UPDOWN
Fig 4 Correlation between the air-gas ratio gas flow and TG ratio
33 LDA analysis
Linear discriminant analysis (LDA) aims to finding a projection direction that
maximizes the separation of class means and minimizes the within-class variance [25]
In this section LDA is utilized to identify the discriminating variables that play an
important role in determining combustion efficiency levels All the data are partitioned
into five groups according to their efficiency levels
11
195
196
197
198
199
200
201
202
203
204
205
206
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
Based on a descending order of TG ratios five efficiency levels for the lsquouprsquo area of
the soaking zone are denoted as HH H M L and LL LDA is conducted on the three
groups of the data with levels of HH M and LL
Fig 5 shows the scattering of the LDA projections of the process observations
collected at the three efficiency levels where y1 and y2 correspond to the first two LDA
components that contain most discriminant information The weighting factors ( and
) of the 16 input variables composing the projections y1 and y2 are shown in Fig 6
where and From left to right the 16 input variables are defined as
the eight AG ratio variables and the eight gas flow variables with the order of variables
of each kind P-u P-d 1-u 1-d 2-u 2-d S-u and S-d
-8 -6 -4 -2 0 2 4 6-4
-3
-2
-1
0
1
2
3
4
5
6
Projection y1
Pro
ject
ion
y2
HHMLL
Fig 5 Scattering of LDA projections y1 and y2 for three efficiency levels
12
207
208
209
210
211
212
213
214
215
216
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
AG_S-u-4
-2
0
2
Wei
ghtin
g fa
ctor
for y
1
AG_S-u
-101234
Wei
ghtin
g fa
ctor
for y
2
Fig 6 Weight factors of various input variables for projections y1 and y2
The two figures reveal that the three groups are clearly discriminated by the LDA
projection and the most significant variables for the different efficiency levels are the
derived variables ie the AG in the lsquouprsquo area of the soaking zone A similar result can
be obtained in the lsquodownrsquo area or any area of the other zones Therefore AG in each
heating area is the key manipulated variable that determines the different combustion
efficiency
4 Modeling and prediction of temperature and temperature-gas ratio
For the model-based operation optimization the models for the temperature and TG
ratio based on the NNG algorithm will be developed and compared
41 NNG variable selection algorithm
The NNG method can be generalized into a two-stage shrinkage method In the first
stage the sign for each variable is determined using the ordinary least square procedure
13
217
218
219
220
221
222
223
224
225
226
227
228
229
230
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
and in the second stage the corresponding magnitudes are computed by solving a series
of constrained quadratic programming
A set of observation data is provided where is the input matrix whose
columns represent the measured candidate variables and is the corresponding
vector of the response data The following expression is given with the number of the
response variable being equal to 1 but a similar procedure can be generalized to any
number of variables Let and be normalized to the zero-mean and unit standard
deviation Additionally let be a set of the ordinary least square estimates of the
coefficients of the following linear model then
(1)
The second stage shrinkage can be formulated as the following optimization problem
subject to
(2)
As decreased and the NNG is tightened more of the become zero and the
remaining nonzero coefficients are shrunk A solution path exists with on which
the appropriate shrinkage can be selected Conventionally the v-fold cross-validation is
used to estimate the prediction error and to select the best solution in the solution path
so as to minimize the prediction or model error
42 Modeling and prediction of temperature and TG ratio
The input-output relations change with time as the reheating process develops An
adaptive modeling strategy is often used to resolve time-varying characteristics of
14
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
industrial processes In this paper the NNG-based regression modeling prediction and
optimization is implemented in a moving window manner where the size of the
window and the length of the moving step are selected as 1000 and 100 respectively In
addition the prediction horizon is also set as 100 In each step the NNG regression
model is built based on the data in the current window which is then used for the
prediction and optimization in the subsequent prediction horizon Next the window
moves forward by replacing the oldest 100 samples for model training by the
observations collected in the previous prediction horizon This moving window strategy
is workable because in the reheating process the input-output relation is slowly time-
varying and the model is still valid for the prediction and optimization in the subsequent
short time
The modeling of the temperature T is based on the air flow rates and gas flow rates
while the modeling for the TG ratio is based on the AG ratios and gas flow rates
Taking the lsquouprsquo area of the soaking zone as an example the prediction result for T and
the TG ratio in the lsquouprsquo area of the soaking zone is shown in Figs 7 and 8 respectively
0 02 04 06 08 1 12 14 16 18 2
x 104
1140
1160
1180
1200
1220
1240
1260
1280
1300
1320
Sampling intervals
T S-u (
)
OriginalPredicted
Fig 7 Prediction of temperature in the lsquouprsquo area of the soaking zone
15
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
0 02 04 06 08 1 12 14 16 18 2
x 104
02
04
06
08
1
12
14
16
18
2
Sampling intervals
TG
S-u
OriginalPredicted
Fig 8 Prediction of TG ratio in the lsquouprsquo area of the soaking zone
A comparison between Figs 7 and 8 show that the prediction of the TG ratio is more
accurate than the prediction of temperature The average R2 in 190 NNG regressions is
0935 for the TG ratio while the average R2 is only 0814 for temperature This is
consistent with the statistical analysis which reveals that the correlation between AG
and TG is higher than that between A and T These results indicate that derived
variables are more meaningful for the purpose of prediction and the modeling of the
TG ratio is more appropriate for implementing optimization
The selected frequency for each variable in 190 NNG regressions and the coefficients
of each variable in 20 NNG regressions for the TG ratio modeling of the lsquouprsquo area of
the soaking zone are shown in Figs 9 and 10 respectively Fig 9 shows that the selected
frequency of the variables in the lsquouprsquo area of the soaking zone is much higher than
variables in other zones Similarly as is shown in Fig 10 the NNG regression
coefficients of the variables in the lsquouprsquo area of the soaking zone are much larger than the
variables in the other areas These results indicated that variables in the lsquouprsquo area of
16
269
270
271
272
273
274
275
276
277
278
279
280
281
282
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
soaking zone are mostly contributed to the modeling of the TG ratio in same area
which is quite reasonable Similar results can be obtained for the other areas
0
20
40
60
80
100
120
140
160
180
200
Sel
ecte
d fre
quen
cy
AGP-u AGP-d AG1-u AG1-d AG2-u AG2-d AGS-u AGS-d GP-u GP-d G1-u G1-d G2-u G2-d GS-u GS-d
Fig 9 Variable selected frequency of over 190 runs for the TG ratio model of the lsquouprsquo
area of the soaking zone
0 2 4 6 8 10 12 14 16 18 20-05
05
AG
P-u
0 2 4 6 8 10 12 14 16 18 20-05
05
GP
-u
0 2 4 6 8 10 12 14 16 18 20-1
0
AG
P-d
0 2 4 6 8 10 12 14 16 18 20-1
0
GP
-d
0 2 4 6 8 10 12 14 16 18 20-05
05
AG
1-u
0 2 4 6 8 10 12 14 16 18 20-05
05
G1-
u
0 2 4 6 8 10 12 14 16 18 20-05
05
AG
1-d
0 2 4 6 8 10 12 14 16 18 20-05
05
G1-
d
0 2 4 6 8 10 12 14 16 18 20-1
0
AG
2-u
0 2 4 6 8 10 12 14 16 18 200
1
G2-
u
0 2 4 6 8 10 12 14 16 18 20-1
0
AG
2-d
0 2 4 6 8 10 12 14 16 18 20-1
0
G2-
d
0 2 4 6 8 10 12 14 16 18 200
1
AG
S-u
0 2 4 6 8 10 12 14 16 18 20-1
0
GS
-u
0 2 4 6 8 10 12 14 16 18 20-05
05
AG
S-d
0 2 4 6 8 10 12 14 16 18 20-05
05
GS
-d
Fig 10 Part of the NNG regression coefficients of each variable over 190 runs for the
TG ratio model of the lsquouprsquo area of the soaking zone
It is remarkable that the NNG regression coefficients corresponding to the AG ratio
in the lsquouprsquo area of the soaking zone are consistently positive On the contrary the
coefficients of the gas flow rate in the same area are consistently negative This
indicates that under this condition the rise of the AG leads to the increase of the TG
17
283
284
285
286
287
288
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
while the rise of the G leads to the drop of the TG This property is helpful for
performing optimization and improve combustion efficiency
For comparison the modeling results for two other algorithms artificial neural
network (ANN) and partial least squares (PLS) [26] are compared with the NNG
algorithm in the following three aspects [27]
(1) Model Size the number of variables selected for modeling
(2) Model Magnitude the mean of the L1 norm of the regression coefficients
(3) Prediction Precision the mean squared prediction error (MSPE)
Summary of the algorithm comparison is shown in Table 2 It can be seen that the
superiorities of the NNG regression in model size model magnitude and model
precision are remarkable
Table 2 Summary of algorithm comparison
Index NNG ANN PLS
Model Size 849 16 16
Model Magnitude 10075 -- 12150
MSPE 00093 00108 00140
5 Model-based optimization
51 Implemention of model-based optimization operation
The goal of optimization is to seek an optimal combination of AG and G in the lsquouprsquo
area of the soaking zone so as to minimize the gas consumption while keeping the
temperature at the target value According to the heating schedule the most expected
temperature in the soaking zone is 1200 Therefore the target temperature
is set at this value to achieve the desired heating effect As shown in Fig 11 in the real
18
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
operation the temperature fluctuates around the target value because of the imperfect
control performance However in order to simplify the analysis and compute the
maximum possible energy saving perfect temperature control is assumed when
adopting the model-based optimization strategy In other words it is assumed that the
actual temperature in the lsquouprsquo area of the soaking zone can be adjusted to the expected
temperature ie 1200
0 02 04 06 08 1 12 14 16 18 2
x 104
1150
1200
1250
1300
Sampling intervals
T S-u (
)
OriginalTarget
Fig 11 Original and target temperature in the lsquouprsquo area of the soaking zone
The adjustment scheme takes the maximum value and minimum value of the original
operation data as the upper and lower bounds for the adjustment Moreover in order to
assure the validity of the linear model the increment or decrement of AG and G should
not beyond of the original value (considered as 10 for the purposes of this paper)
The adjustment strategy can be formulated as
19
307
308
309
310
311
312
313
314
315
316
317
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
(3)
where NNG() denotes the NNG regression model of the TG ratio and are the
NNG regression coefficients of the AG and G in the lsquouprsquo area of the soaking zone for
modeling TG in same area
With the model-based strategy and adaptive model of the TG ratio given in Section
42 the optimization adjustment results can be obtained as follows
The adjustment of gas flow in the lsquouprsquo area of the soaking zone shown in Fig 12
illustrates that the implementation of the model-based optimization operation can reduce
the consumption of the gas flow Compared with the original case 938 of the gas can
be saved on average by utilizing the model-based adjustment
0 02 04 06 08 1 12 14 16 18 2
x 104
-800
-700
-600
-500
-400
-300
-200
-100
0
Sampling intervals
Adj
ustm
ent o
f gas
flow
m(
3 h)
20
318
319
320
321
322
323
324
325
326
327
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
(a) Adjustment amount of gas flow
0 02 04 06 08 1 12 14 16 18 2
x 104
500
1000
1500
2000
2500
3000
3500
4000
4500
5000G
as fl
owm
(3 h
)
Sampling intervals
OriginalAdjusted
(b) Comparison of original and adjusted gas flow
095 096 097 098 099 1 101 102 103 104 105
x 104
500
1000
1500
2000
2500
3000
3500
4000
4500
5000
Gas
flow
m(
3 h)
Sampling intervals
OriginalAdjusted
(c) Comparison of original and adjusted gas flow in interval [9501-10500]
Fig 12 Adjustment of gas flow in the lsquouprsquo area of the soaking zone
The adjustment of the air flow in the lsquouprsquo area of the soaking zone given in Fig 13
shows that the air flow is reduced in most cases and only increased for a few cases In
21
328
329
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
general 681 of the air flow is reduced after the implementation of the model-based
optimization
0 02 04 06 08 1 12 14 16 18 2
x 104
-2000
-1500
-1000
-500
0
500
1000
Sampling intervals
Adj
ustm
ent o
f air
flow
(m3 h
)
(a) Adjustment amount of air flow
0 02 04 06 08 1 12 14 16 18 2
x 104
4000
5000
6000
7000
8000
9000
10000
Air
flow
m(
3 h)
Sampling intervals
OriginalAdjusted
(b) Comparison of original and adjusted air flow
22
330
331
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
095 096 097 098 099 1 101 102 103 104 105
x 104
4000
5000
6000
7000
8000
9000
10000
Air
flow
m(
3 h)
Sampling intervals
OriginalAdjusted
(c) Comparison of original and adjusted air flow in interval [9501-10500]
Fig 13 Adjustment of air flow in the lsquouprsquo area of the soaking zone
52 Discussions
In this section detailed analysis for the optimization operation results is provided
The statistics of optimization at the boundary conditions shown in Table 3 illustrates
that the adjusted value for G reaches its lower limit in most cases (8045 for
and 687 for ) which is the most energy-efficient point In a number of cases
(1079 for and 001 for ) the adjusted AG reaches its upper
limit These results indicate that the optimization operations maximize the combustion
efficiency by decreasing G and increasing the AG ratio up to the boundary conditions
However the lower limit is also unexpectedly attained in a small number of cases
(110 for and 078 for ) for which a more detailed analysis is
given below
23
332
333
334
335
336
337
338
339
340
341
342
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
Table 3 Statistics of optimization at boundary conditions
Total
Amount 0 1374 0 16090 17464
Percentag
e 0 687 0 8045
8732
Total
Amount 2 220 2158 156 2536
Percentag
e 001 110 1079 078
1268
Table 4 Statistics for different optimization operations
Adjustment
Gdarr
Guarr TotalAGuarr AGdarr
Tdarr Tuarr Tdarr Tuarr
Amount 9548 3744 6328 380 0 20000
Percentage 4774 1872 3164 190 0 100
The statistics for different optimization operations is performed and shown in Table
4 It can be seen that all the adjusted operations result in the reduction of gas
consumption In over half the cases (4774 with decreased temperature and 1872
with increased temperature) AG is adjusted to higher levels Meanwhile a number of
the adjustments (3164) lower the temperature by reducing the G and AG at the same
time This result indicates that the original temperature cannot be adjusted to the target
by only reducing G in the constraint conditions The remaining cases (19) are
24
343
344
345
346
347
348
349
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
relatively special in which an excess of air is supplied in the original operation and the
adjusted operation thereby increasing the temperature by a smaller AG with less gas
consumption
6 Conclusion
Based on the actual operation data this paper aims to explore the improvement of the
combustion efficiency and the room for energy conservation Correlation analysis and
LDA show that it is of great significance to introduce two derived ratio variables which
are the AG ratio and the TG ratio A type of combustion efficiency utilizing an on-line
soft sensor is put forward by employing a NNG variable selection algorithm which
provides a good solution to the combustion efficiency real-time measurement problem
of a reheating furnace The implementation of the model-based optimization is studied
based on the actual operational data Detailed analysis for the optimization results is
given for the different cases The results show that significant energy conservation can
be achieved when the furnace operation is optimized by using the developed soft sensor
model
Acknowledgement
The authors would like to thank the financial support provided by the National Nature
Science Foundation of China under Grant 61171145 Y Yao was supported by Ministry
of Science amp Technology ROC under Grant number MOST 104-2221-E-007-129
References
25
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[1] Z J Wang Q D Wu and T Y Chai Optimal-setting control for complicated
industrial processes and its application studyControl Engineering Practice vol
12 pp 65-74 2004
[2] A Steinboeck K Graichen and A Kugi Dynamic Optimization of a Slab
Reheating Furnace With Consistent Approximation of Control VariablesIEEE
Transactions on Control Systems Technology vol 19 pp 1444-1456 2011
[3] B T Zhang C Y Wang Q Qin and L Li Influence of Boiler Combustion
Adjustment on NOxEmission and Boiler EfficiencyAdvanced Materials
Research vol 732-733 pp 234-237 2013
[4] C K Yoo and IB Lee Soft Sensor and Adaptive Model-Based Dissolved
Oxygen Control for Biological Wastewater Treatment ProcessesEnvironmental
Engineering Science vol 21 pp 331-340 2004
[5] S A Bhat D N Saraf S Gupta and S K Gupta Use of Agitator Power as a
Soft Sensor for Bulk Free-Radical Polymerization of Methyl Methacrylate in
Batch ReactorsIndustrial amp Engineering Chemistry Research vol 45 pp 4243-
4255 2006
[6] K Desai Y Badhe S S Tambe and B D Kulkarni Soft-sensor development
for fed-batch bioreactors using support vector regressionBiochemical
Engineering Journal vol 27 pp 225-239 2006
[7] Y P Badhe Lonari J Tambe S S amp Kulkarni B D Improve polyethylene
process control and product qualityHydrocarbon Processing vol 86 pp 53-60
2007
26
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[8] N K Nath K Mandal A K Singh B Basu C Bhanu S Kumar et al Ladle
furnace on-line reckoner for prediction and control of steel temperature and
compositionIronmaking amp Steelmaking vol 33 pp 140-150 2006
[9] A J Yan T Y Chai F H Wu and P Wang Hybrid intelligent control of
combustion process for ore-roasting furnaceJournal of Control Theory and
Applications vol 6 pp 80-85 2008
[10] J Li W M Zhong H Cheng X D Kong and F Qian A data-driven soft
sensor modeling for furnace temperature of Opposed Multi-Burner gasifier pp
705-710 2011
[11] Y H Yang Y H Liu X Z Liu and S K Qin Billet temperature soft sensor
model of reheating furnace based on RVM method pp 4003-4006 2011
[12] J H Wang C Wang X F Zhu and X K Fang Application of soft sensor in
welding seam tracking prediction based on LSSVM and PSO with compression
factor pp 2441-2446 2013
[13] L Balbis J Balderud and M J Grimble Nonlinear predictive control of steel
slab reheating furnace pp 1679-1684 2008
[14] A Steinboeck D Wild T Kiefer and A Kugi A mathematical model of a slab
reheating furnace with radiative heat transfer and non-participating gaseous
media International Journal Of Heat And Mass Transfer vol 53 pp 5933-
5946 Dec 2010
[15] A Steinboeck D Wild and A Kugi Nonlinear model predictive control of a
continuous slab reheating furnace Control Engineering Practice vol 21 pp
495-508 2013
27
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[16] C Zhang T Ishii and S Sugiyama Numerical Modeling Of the Thermal
Performance Of Regenerative Slab Reheat Furnaces Numerical Heat Transfer
Part A Applications vol 32 pp 613-631 1997
[17] J G Kim and K Y Huh Prediction of Transient Slab Temperature Distribution
in the Re-heating Furnace of a Walking-beam Type for Rolling of Steel Slabs
ISIJ International vol 40 pp 1115-1123 2000
[18] J G Kim K Y and H I T K Three-Dimensional Analysis Of the Walking-
Beam-Type Slab Reheating Furnace In Hot Strip Mills Numerical Heat
Transfer Part A Applications vol 38 pp 589-609 2000
[19] CT Hsieh MJ Huang ST Lee and CH Wang Numerical Modeling of a
Walking-Beam-Type Slab Reheating Furnace Numerical Heat Transfer Part A
Applications vol 53 pp 966-981 2008
[20] MJ Huang CT Hsieh ST Lee and CH Wang A Coupled Numerical Study
of Slab Temperature and Gas Temperature in the Walking-Beam-Type Slab
Reheating Furnace Numerical Heat Transfer Part A Applications vol 54 pp
625-646 2008
[21] Z Song and A Kusiak Constraint-Based Control of Boiler Efficiency A Data-
Mining Approach IEEE Transactions on Industrial Informatics vol 3 pp 73-
83 2007
[22] J Q Li J J Gu and C L Niu The Operation Optimization based on
Correlation Analysis of Operation Parameters in Power Plant pp 138-141
2008
28
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[23] J G Wang S S Shieh S S Jang D S H Wong and C W Wu A two-tier
approach to the data-driven modeling on thermal efficiency of a BFGcoal co-
firing boiler Fuel vol 111 pp 528-534 Sep 2013
[24] L Breiman Better Subset Regression Using the Nonnegative Garrote
Technometrics vol 37 pp 373-384 1995
[25] R A Fisher The use of multiple measurements in taxonomic problems
AnnHum Genet vol 7 pp 179-188 1936
[26] J Liu Developing a soft sensor based on sparse partial least squares with
variable selection Journal of Process Control vol 24 pp 1046-1056 2014
[27] K Sun J Liu JL Kang SS Jang D SH Wong and DS Chen
Development of a variable selection method for soft sensor using artificial
neural network and nonnegative garrote Journal of Process Control vol 24 pp
1068-1075 2014
29
439
440
441
442
443
444
445
446
447
448
449
450
451
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
Furthermore as related to the lsquouprsquo or lsquodownrsquo areas in one zone TG in each area has the
highest correlation with AG and G in the same area while AG and G in the opposite
area of the same zone is secondary This can be easily seen from the red and blue color
markings in each red rectangle block Thus it is of great significance to introduce these
derived ratio variables
-101
TG
P-u
AGP GP AG1G1 AG2 G2 AGS GS
-101
TG
P-d
AGP GP AG1 G1 AG2 G2 AGS GS
-101
TG
1-u
AGP GP AG1 G1 AG2 G2 AGS GS
-101
TG
1-d
AGP GP AG1 G1 AG2 G2 AGS GS
-101
TG
2-u
AGP GP AG1 G1 AG2 G2 AGS GS
-101
TG
2-d
AGP GP AG1 G1 AG2 G2 AGS GS
-101
TG
S-u
AGP GP AG1 G1 AG2 G2 AGS GS
-101
TG
S-d
AGP GP AG1 G1 AG2 G2 AGS GS
UPDOWN
Fig 4 Correlation between the air-gas ratio gas flow and TG ratio
33 LDA analysis
Linear discriminant analysis (LDA) aims to finding a projection direction that
maximizes the separation of class means and minimizes the within-class variance [25]
In this section LDA is utilized to identify the discriminating variables that play an
important role in determining combustion efficiency levels All the data are partitioned
into five groups according to their efficiency levels
11
195
196
197
198
199
200
201
202
203
204
205
206
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
Based on a descending order of TG ratios five efficiency levels for the lsquouprsquo area of
the soaking zone are denoted as HH H M L and LL LDA is conducted on the three
groups of the data with levels of HH M and LL
Fig 5 shows the scattering of the LDA projections of the process observations
collected at the three efficiency levels where y1 and y2 correspond to the first two LDA
components that contain most discriminant information The weighting factors ( and
) of the 16 input variables composing the projections y1 and y2 are shown in Fig 6
where and From left to right the 16 input variables are defined as
the eight AG ratio variables and the eight gas flow variables with the order of variables
of each kind P-u P-d 1-u 1-d 2-u 2-d S-u and S-d
-8 -6 -4 -2 0 2 4 6-4
-3
-2
-1
0
1
2
3
4
5
6
Projection y1
Pro
ject
ion
y2
HHMLL
Fig 5 Scattering of LDA projections y1 and y2 for three efficiency levels
12
207
208
209
210
211
212
213
214
215
216
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
AG_S-u-4
-2
0
2
Wei
ghtin
g fa
ctor
for y
1
AG_S-u
-101234
Wei
ghtin
g fa
ctor
for y
2
Fig 6 Weight factors of various input variables for projections y1 and y2
The two figures reveal that the three groups are clearly discriminated by the LDA
projection and the most significant variables for the different efficiency levels are the
derived variables ie the AG in the lsquouprsquo area of the soaking zone A similar result can
be obtained in the lsquodownrsquo area or any area of the other zones Therefore AG in each
heating area is the key manipulated variable that determines the different combustion
efficiency
4 Modeling and prediction of temperature and temperature-gas ratio
For the model-based operation optimization the models for the temperature and TG
ratio based on the NNG algorithm will be developed and compared
41 NNG variable selection algorithm
The NNG method can be generalized into a two-stage shrinkage method In the first
stage the sign for each variable is determined using the ordinary least square procedure
13
217
218
219
220
221
222
223
224
225
226
227
228
229
230
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
and in the second stage the corresponding magnitudes are computed by solving a series
of constrained quadratic programming
A set of observation data is provided where is the input matrix whose
columns represent the measured candidate variables and is the corresponding
vector of the response data The following expression is given with the number of the
response variable being equal to 1 but a similar procedure can be generalized to any
number of variables Let and be normalized to the zero-mean and unit standard
deviation Additionally let be a set of the ordinary least square estimates of the
coefficients of the following linear model then
(1)
The second stage shrinkage can be formulated as the following optimization problem
subject to
(2)
As decreased and the NNG is tightened more of the become zero and the
remaining nonzero coefficients are shrunk A solution path exists with on which
the appropriate shrinkage can be selected Conventionally the v-fold cross-validation is
used to estimate the prediction error and to select the best solution in the solution path
so as to minimize the prediction or model error
42 Modeling and prediction of temperature and TG ratio
The input-output relations change with time as the reheating process develops An
adaptive modeling strategy is often used to resolve time-varying characteristics of
14
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
industrial processes In this paper the NNG-based regression modeling prediction and
optimization is implemented in a moving window manner where the size of the
window and the length of the moving step are selected as 1000 and 100 respectively In
addition the prediction horizon is also set as 100 In each step the NNG regression
model is built based on the data in the current window which is then used for the
prediction and optimization in the subsequent prediction horizon Next the window
moves forward by replacing the oldest 100 samples for model training by the
observations collected in the previous prediction horizon This moving window strategy
is workable because in the reheating process the input-output relation is slowly time-
varying and the model is still valid for the prediction and optimization in the subsequent
short time
The modeling of the temperature T is based on the air flow rates and gas flow rates
while the modeling for the TG ratio is based on the AG ratios and gas flow rates
Taking the lsquouprsquo area of the soaking zone as an example the prediction result for T and
the TG ratio in the lsquouprsquo area of the soaking zone is shown in Figs 7 and 8 respectively
0 02 04 06 08 1 12 14 16 18 2
x 104
1140
1160
1180
1200
1220
1240
1260
1280
1300
1320
Sampling intervals
T S-u (
)
OriginalPredicted
Fig 7 Prediction of temperature in the lsquouprsquo area of the soaking zone
15
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
0 02 04 06 08 1 12 14 16 18 2
x 104
02
04
06
08
1
12
14
16
18
2
Sampling intervals
TG
S-u
OriginalPredicted
Fig 8 Prediction of TG ratio in the lsquouprsquo area of the soaking zone
A comparison between Figs 7 and 8 show that the prediction of the TG ratio is more
accurate than the prediction of temperature The average R2 in 190 NNG regressions is
0935 for the TG ratio while the average R2 is only 0814 for temperature This is
consistent with the statistical analysis which reveals that the correlation between AG
and TG is higher than that between A and T These results indicate that derived
variables are more meaningful for the purpose of prediction and the modeling of the
TG ratio is more appropriate for implementing optimization
The selected frequency for each variable in 190 NNG regressions and the coefficients
of each variable in 20 NNG regressions for the TG ratio modeling of the lsquouprsquo area of
the soaking zone are shown in Figs 9 and 10 respectively Fig 9 shows that the selected
frequency of the variables in the lsquouprsquo area of the soaking zone is much higher than
variables in other zones Similarly as is shown in Fig 10 the NNG regression
coefficients of the variables in the lsquouprsquo area of the soaking zone are much larger than the
variables in the other areas These results indicated that variables in the lsquouprsquo area of
16
269
270
271
272
273
274
275
276
277
278
279
280
281
282
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
soaking zone are mostly contributed to the modeling of the TG ratio in same area
which is quite reasonable Similar results can be obtained for the other areas
0
20
40
60
80
100
120
140
160
180
200
Sel
ecte
d fre
quen
cy
AGP-u AGP-d AG1-u AG1-d AG2-u AG2-d AGS-u AGS-d GP-u GP-d G1-u G1-d G2-u G2-d GS-u GS-d
Fig 9 Variable selected frequency of over 190 runs for the TG ratio model of the lsquouprsquo
area of the soaking zone
0 2 4 6 8 10 12 14 16 18 20-05
05
AG
P-u
0 2 4 6 8 10 12 14 16 18 20-05
05
GP
-u
0 2 4 6 8 10 12 14 16 18 20-1
0
AG
P-d
0 2 4 6 8 10 12 14 16 18 20-1
0
GP
-d
0 2 4 6 8 10 12 14 16 18 20-05
05
AG
1-u
0 2 4 6 8 10 12 14 16 18 20-05
05
G1-
u
0 2 4 6 8 10 12 14 16 18 20-05
05
AG
1-d
0 2 4 6 8 10 12 14 16 18 20-05
05
G1-
d
0 2 4 6 8 10 12 14 16 18 20-1
0
AG
2-u
0 2 4 6 8 10 12 14 16 18 200
1
G2-
u
0 2 4 6 8 10 12 14 16 18 20-1
0
AG
2-d
0 2 4 6 8 10 12 14 16 18 20-1
0
G2-
d
0 2 4 6 8 10 12 14 16 18 200
1
AG
S-u
0 2 4 6 8 10 12 14 16 18 20-1
0
GS
-u
0 2 4 6 8 10 12 14 16 18 20-05
05
AG
S-d
0 2 4 6 8 10 12 14 16 18 20-05
05
GS
-d
Fig 10 Part of the NNG regression coefficients of each variable over 190 runs for the
TG ratio model of the lsquouprsquo area of the soaking zone
It is remarkable that the NNG regression coefficients corresponding to the AG ratio
in the lsquouprsquo area of the soaking zone are consistently positive On the contrary the
coefficients of the gas flow rate in the same area are consistently negative This
indicates that under this condition the rise of the AG leads to the increase of the TG
17
283
284
285
286
287
288
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
while the rise of the G leads to the drop of the TG This property is helpful for
performing optimization and improve combustion efficiency
For comparison the modeling results for two other algorithms artificial neural
network (ANN) and partial least squares (PLS) [26] are compared with the NNG
algorithm in the following three aspects [27]
(1) Model Size the number of variables selected for modeling
(2) Model Magnitude the mean of the L1 norm of the regression coefficients
(3) Prediction Precision the mean squared prediction error (MSPE)
Summary of the algorithm comparison is shown in Table 2 It can be seen that the
superiorities of the NNG regression in model size model magnitude and model
precision are remarkable
Table 2 Summary of algorithm comparison
Index NNG ANN PLS
Model Size 849 16 16
Model Magnitude 10075 -- 12150
MSPE 00093 00108 00140
5 Model-based optimization
51 Implemention of model-based optimization operation
The goal of optimization is to seek an optimal combination of AG and G in the lsquouprsquo
area of the soaking zone so as to minimize the gas consumption while keeping the
temperature at the target value According to the heating schedule the most expected
temperature in the soaking zone is 1200 Therefore the target temperature
is set at this value to achieve the desired heating effect As shown in Fig 11 in the real
18
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
operation the temperature fluctuates around the target value because of the imperfect
control performance However in order to simplify the analysis and compute the
maximum possible energy saving perfect temperature control is assumed when
adopting the model-based optimization strategy In other words it is assumed that the
actual temperature in the lsquouprsquo area of the soaking zone can be adjusted to the expected
temperature ie 1200
0 02 04 06 08 1 12 14 16 18 2
x 104
1150
1200
1250
1300
Sampling intervals
T S-u (
)
OriginalTarget
Fig 11 Original and target temperature in the lsquouprsquo area of the soaking zone
The adjustment scheme takes the maximum value and minimum value of the original
operation data as the upper and lower bounds for the adjustment Moreover in order to
assure the validity of the linear model the increment or decrement of AG and G should
not beyond of the original value (considered as 10 for the purposes of this paper)
The adjustment strategy can be formulated as
19
307
308
309
310
311
312
313
314
315
316
317
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
(3)
where NNG() denotes the NNG regression model of the TG ratio and are the
NNG regression coefficients of the AG and G in the lsquouprsquo area of the soaking zone for
modeling TG in same area
With the model-based strategy and adaptive model of the TG ratio given in Section
42 the optimization adjustment results can be obtained as follows
The adjustment of gas flow in the lsquouprsquo area of the soaking zone shown in Fig 12
illustrates that the implementation of the model-based optimization operation can reduce
the consumption of the gas flow Compared with the original case 938 of the gas can
be saved on average by utilizing the model-based adjustment
0 02 04 06 08 1 12 14 16 18 2
x 104
-800
-700
-600
-500
-400
-300
-200
-100
0
Sampling intervals
Adj
ustm
ent o
f gas
flow
m(
3 h)
20
318
319
320
321
322
323
324
325
326
327
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
(a) Adjustment amount of gas flow
0 02 04 06 08 1 12 14 16 18 2
x 104
500
1000
1500
2000
2500
3000
3500
4000
4500
5000G
as fl
owm
(3 h
)
Sampling intervals
OriginalAdjusted
(b) Comparison of original and adjusted gas flow
095 096 097 098 099 1 101 102 103 104 105
x 104
500
1000
1500
2000
2500
3000
3500
4000
4500
5000
Gas
flow
m(
3 h)
Sampling intervals
OriginalAdjusted
(c) Comparison of original and adjusted gas flow in interval [9501-10500]
Fig 12 Adjustment of gas flow in the lsquouprsquo area of the soaking zone
The adjustment of the air flow in the lsquouprsquo area of the soaking zone given in Fig 13
shows that the air flow is reduced in most cases and only increased for a few cases In
21
328
329
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
general 681 of the air flow is reduced after the implementation of the model-based
optimization
0 02 04 06 08 1 12 14 16 18 2
x 104
-2000
-1500
-1000
-500
0
500
1000
Sampling intervals
Adj
ustm
ent o
f air
flow
(m3 h
)
(a) Adjustment amount of air flow
0 02 04 06 08 1 12 14 16 18 2
x 104
4000
5000
6000
7000
8000
9000
10000
Air
flow
m(
3 h)
Sampling intervals
OriginalAdjusted
(b) Comparison of original and adjusted air flow
22
330
331
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
095 096 097 098 099 1 101 102 103 104 105
x 104
4000
5000
6000
7000
8000
9000
10000
Air
flow
m(
3 h)
Sampling intervals
OriginalAdjusted
(c) Comparison of original and adjusted air flow in interval [9501-10500]
Fig 13 Adjustment of air flow in the lsquouprsquo area of the soaking zone
52 Discussions
In this section detailed analysis for the optimization operation results is provided
The statistics of optimization at the boundary conditions shown in Table 3 illustrates
that the adjusted value for G reaches its lower limit in most cases (8045 for
and 687 for ) which is the most energy-efficient point In a number of cases
(1079 for and 001 for ) the adjusted AG reaches its upper
limit These results indicate that the optimization operations maximize the combustion
efficiency by decreasing G and increasing the AG ratio up to the boundary conditions
However the lower limit is also unexpectedly attained in a small number of cases
(110 for and 078 for ) for which a more detailed analysis is
given below
23
332
333
334
335
336
337
338
339
340
341
342
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
Table 3 Statistics of optimization at boundary conditions
Total
Amount 0 1374 0 16090 17464
Percentag
e 0 687 0 8045
8732
Total
Amount 2 220 2158 156 2536
Percentag
e 001 110 1079 078
1268
Table 4 Statistics for different optimization operations
Adjustment
Gdarr
Guarr TotalAGuarr AGdarr
Tdarr Tuarr Tdarr Tuarr
Amount 9548 3744 6328 380 0 20000
Percentage 4774 1872 3164 190 0 100
The statistics for different optimization operations is performed and shown in Table
4 It can be seen that all the adjusted operations result in the reduction of gas
consumption In over half the cases (4774 with decreased temperature and 1872
with increased temperature) AG is adjusted to higher levels Meanwhile a number of
the adjustments (3164) lower the temperature by reducing the G and AG at the same
time This result indicates that the original temperature cannot be adjusted to the target
by only reducing G in the constraint conditions The remaining cases (19) are
24
343
344
345
346
347
348
349
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
relatively special in which an excess of air is supplied in the original operation and the
adjusted operation thereby increasing the temperature by a smaller AG with less gas
consumption
6 Conclusion
Based on the actual operation data this paper aims to explore the improvement of the
combustion efficiency and the room for energy conservation Correlation analysis and
LDA show that it is of great significance to introduce two derived ratio variables which
are the AG ratio and the TG ratio A type of combustion efficiency utilizing an on-line
soft sensor is put forward by employing a NNG variable selection algorithm which
provides a good solution to the combustion efficiency real-time measurement problem
of a reheating furnace The implementation of the model-based optimization is studied
based on the actual operational data Detailed analysis for the optimization results is
given for the different cases The results show that significant energy conservation can
be achieved when the furnace operation is optimized by using the developed soft sensor
model
Acknowledgement
The authors would like to thank the financial support provided by the National Nature
Science Foundation of China under Grant 61171145 Y Yao was supported by Ministry
of Science amp Technology ROC under Grant number MOST 104-2221-E-007-129
References
25
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[1] Z J Wang Q D Wu and T Y Chai Optimal-setting control for complicated
industrial processes and its application studyControl Engineering Practice vol
12 pp 65-74 2004
[2] A Steinboeck K Graichen and A Kugi Dynamic Optimization of a Slab
Reheating Furnace With Consistent Approximation of Control VariablesIEEE
Transactions on Control Systems Technology vol 19 pp 1444-1456 2011
[3] B T Zhang C Y Wang Q Qin and L Li Influence of Boiler Combustion
Adjustment on NOxEmission and Boiler EfficiencyAdvanced Materials
Research vol 732-733 pp 234-237 2013
[4] C K Yoo and IB Lee Soft Sensor and Adaptive Model-Based Dissolved
Oxygen Control for Biological Wastewater Treatment ProcessesEnvironmental
Engineering Science vol 21 pp 331-340 2004
[5] S A Bhat D N Saraf S Gupta and S K Gupta Use of Agitator Power as a
Soft Sensor for Bulk Free-Radical Polymerization of Methyl Methacrylate in
Batch ReactorsIndustrial amp Engineering Chemistry Research vol 45 pp 4243-
4255 2006
[6] K Desai Y Badhe S S Tambe and B D Kulkarni Soft-sensor development
for fed-batch bioreactors using support vector regressionBiochemical
Engineering Journal vol 27 pp 225-239 2006
[7] Y P Badhe Lonari J Tambe S S amp Kulkarni B D Improve polyethylene
process control and product qualityHydrocarbon Processing vol 86 pp 53-60
2007
26
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[8] N K Nath K Mandal A K Singh B Basu C Bhanu S Kumar et al Ladle
furnace on-line reckoner for prediction and control of steel temperature and
compositionIronmaking amp Steelmaking vol 33 pp 140-150 2006
[9] A J Yan T Y Chai F H Wu and P Wang Hybrid intelligent control of
combustion process for ore-roasting furnaceJournal of Control Theory and
Applications vol 6 pp 80-85 2008
[10] J Li W M Zhong H Cheng X D Kong and F Qian A data-driven soft
sensor modeling for furnace temperature of Opposed Multi-Burner gasifier pp
705-710 2011
[11] Y H Yang Y H Liu X Z Liu and S K Qin Billet temperature soft sensor
model of reheating furnace based on RVM method pp 4003-4006 2011
[12] J H Wang C Wang X F Zhu and X K Fang Application of soft sensor in
welding seam tracking prediction based on LSSVM and PSO with compression
factor pp 2441-2446 2013
[13] L Balbis J Balderud and M J Grimble Nonlinear predictive control of steel
slab reheating furnace pp 1679-1684 2008
[14] A Steinboeck D Wild T Kiefer and A Kugi A mathematical model of a slab
reheating furnace with radiative heat transfer and non-participating gaseous
media International Journal Of Heat And Mass Transfer vol 53 pp 5933-
5946 Dec 2010
[15] A Steinboeck D Wild and A Kugi Nonlinear model predictive control of a
continuous slab reheating furnace Control Engineering Practice vol 21 pp
495-508 2013
27
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[16] C Zhang T Ishii and S Sugiyama Numerical Modeling Of the Thermal
Performance Of Regenerative Slab Reheat Furnaces Numerical Heat Transfer
Part A Applications vol 32 pp 613-631 1997
[17] J G Kim and K Y Huh Prediction of Transient Slab Temperature Distribution
in the Re-heating Furnace of a Walking-beam Type for Rolling of Steel Slabs
ISIJ International vol 40 pp 1115-1123 2000
[18] J G Kim K Y and H I T K Three-Dimensional Analysis Of the Walking-
Beam-Type Slab Reheating Furnace In Hot Strip Mills Numerical Heat
Transfer Part A Applications vol 38 pp 589-609 2000
[19] CT Hsieh MJ Huang ST Lee and CH Wang Numerical Modeling of a
Walking-Beam-Type Slab Reheating Furnace Numerical Heat Transfer Part A
Applications vol 53 pp 966-981 2008
[20] MJ Huang CT Hsieh ST Lee and CH Wang A Coupled Numerical Study
of Slab Temperature and Gas Temperature in the Walking-Beam-Type Slab
Reheating Furnace Numerical Heat Transfer Part A Applications vol 54 pp
625-646 2008
[21] Z Song and A Kusiak Constraint-Based Control of Boiler Efficiency A Data-
Mining Approach IEEE Transactions on Industrial Informatics vol 3 pp 73-
83 2007
[22] J Q Li J J Gu and C L Niu The Operation Optimization based on
Correlation Analysis of Operation Parameters in Power Plant pp 138-141
2008
28
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[23] J G Wang S S Shieh S S Jang D S H Wong and C W Wu A two-tier
approach to the data-driven modeling on thermal efficiency of a BFGcoal co-
firing boiler Fuel vol 111 pp 528-534 Sep 2013
[24] L Breiman Better Subset Regression Using the Nonnegative Garrote
Technometrics vol 37 pp 373-384 1995
[25] R A Fisher The use of multiple measurements in taxonomic problems
AnnHum Genet vol 7 pp 179-188 1936
[26] J Liu Developing a soft sensor based on sparse partial least squares with
variable selection Journal of Process Control vol 24 pp 1046-1056 2014
[27] K Sun J Liu JL Kang SS Jang D SH Wong and DS Chen
Development of a variable selection method for soft sensor using artificial
neural network and nonnegative garrote Journal of Process Control vol 24 pp
1068-1075 2014
29
439
440
441
442
443
444
445
446
447
448
449
450
451
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
Based on a descending order of TG ratios five efficiency levels for the lsquouprsquo area of
the soaking zone are denoted as HH H M L and LL LDA is conducted on the three
groups of the data with levels of HH M and LL
Fig 5 shows the scattering of the LDA projections of the process observations
collected at the three efficiency levels where y1 and y2 correspond to the first two LDA
components that contain most discriminant information The weighting factors ( and
) of the 16 input variables composing the projections y1 and y2 are shown in Fig 6
where and From left to right the 16 input variables are defined as
the eight AG ratio variables and the eight gas flow variables with the order of variables
of each kind P-u P-d 1-u 1-d 2-u 2-d S-u and S-d
-8 -6 -4 -2 0 2 4 6-4
-3
-2
-1
0
1
2
3
4
5
6
Projection y1
Pro
ject
ion
y2
HHMLL
Fig 5 Scattering of LDA projections y1 and y2 for three efficiency levels
12
207
208
209
210
211
212
213
214
215
216
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
AG_S-u-4
-2
0
2
Wei
ghtin
g fa
ctor
for y
1
AG_S-u
-101234
Wei
ghtin
g fa
ctor
for y
2
Fig 6 Weight factors of various input variables for projections y1 and y2
The two figures reveal that the three groups are clearly discriminated by the LDA
projection and the most significant variables for the different efficiency levels are the
derived variables ie the AG in the lsquouprsquo area of the soaking zone A similar result can
be obtained in the lsquodownrsquo area or any area of the other zones Therefore AG in each
heating area is the key manipulated variable that determines the different combustion
efficiency
4 Modeling and prediction of temperature and temperature-gas ratio
For the model-based operation optimization the models for the temperature and TG
ratio based on the NNG algorithm will be developed and compared
41 NNG variable selection algorithm
The NNG method can be generalized into a two-stage shrinkage method In the first
stage the sign for each variable is determined using the ordinary least square procedure
13
217
218
219
220
221
222
223
224
225
226
227
228
229
230
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
and in the second stage the corresponding magnitudes are computed by solving a series
of constrained quadratic programming
A set of observation data is provided where is the input matrix whose
columns represent the measured candidate variables and is the corresponding
vector of the response data The following expression is given with the number of the
response variable being equal to 1 but a similar procedure can be generalized to any
number of variables Let and be normalized to the zero-mean and unit standard
deviation Additionally let be a set of the ordinary least square estimates of the
coefficients of the following linear model then
(1)
The second stage shrinkage can be formulated as the following optimization problem
subject to
(2)
As decreased and the NNG is tightened more of the become zero and the
remaining nonzero coefficients are shrunk A solution path exists with on which
the appropriate shrinkage can be selected Conventionally the v-fold cross-validation is
used to estimate the prediction error and to select the best solution in the solution path
so as to minimize the prediction or model error
42 Modeling and prediction of temperature and TG ratio
The input-output relations change with time as the reheating process develops An
adaptive modeling strategy is often used to resolve time-varying characteristics of
14
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
industrial processes In this paper the NNG-based regression modeling prediction and
optimization is implemented in a moving window manner where the size of the
window and the length of the moving step are selected as 1000 and 100 respectively In
addition the prediction horizon is also set as 100 In each step the NNG regression
model is built based on the data in the current window which is then used for the
prediction and optimization in the subsequent prediction horizon Next the window
moves forward by replacing the oldest 100 samples for model training by the
observations collected in the previous prediction horizon This moving window strategy
is workable because in the reheating process the input-output relation is slowly time-
varying and the model is still valid for the prediction and optimization in the subsequent
short time
The modeling of the temperature T is based on the air flow rates and gas flow rates
while the modeling for the TG ratio is based on the AG ratios and gas flow rates
Taking the lsquouprsquo area of the soaking zone as an example the prediction result for T and
the TG ratio in the lsquouprsquo area of the soaking zone is shown in Figs 7 and 8 respectively
0 02 04 06 08 1 12 14 16 18 2
x 104
1140
1160
1180
1200
1220
1240
1260
1280
1300
1320
Sampling intervals
T S-u (
)
OriginalPredicted
Fig 7 Prediction of temperature in the lsquouprsquo area of the soaking zone
15
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
0 02 04 06 08 1 12 14 16 18 2
x 104
02
04
06
08
1
12
14
16
18
2
Sampling intervals
TG
S-u
OriginalPredicted
Fig 8 Prediction of TG ratio in the lsquouprsquo area of the soaking zone
A comparison between Figs 7 and 8 show that the prediction of the TG ratio is more
accurate than the prediction of temperature The average R2 in 190 NNG regressions is
0935 for the TG ratio while the average R2 is only 0814 for temperature This is
consistent with the statistical analysis which reveals that the correlation between AG
and TG is higher than that between A and T These results indicate that derived
variables are more meaningful for the purpose of prediction and the modeling of the
TG ratio is more appropriate for implementing optimization
The selected frequency for each variable in 190 NNG regressions and the coefficients
of each variable in 20 NNG regressions for the TG ratio modeling of the lsquouprsquo area of
the soaking zone are shown in Figs 9 and 10 respectively Fig 9 shows that the selected
frequency of the variables in the lsquouprsquo area of the soaking zone is much higher than
variables in other zones Similarly as is shown in Fig 10 the NNG regression
coefficients of the variables in the lsquouprsquo area of the soaking zone are much larger than the
variables in the other areas These results indicated that variables in the lsquouprsquo area of
16
269
270
271
272
273
274
275
276
277
278
279
280
281
282
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
soaking zone are mostly contributed to the modeling of the TG ratio in same area
which is quite reasonable Similar results can be obtained for the other areas
0
20
40
60
80
100
120
140
160
180
200
Sel
ecte
d fre
quen
cy
AGP-u AGP-d AG1-u AG1-d AG2-u AG2-d AGS-u AGS-d GP-u GP-d G1-u G1-d G2-u G2-d GS-u GS-d
Fig 9 Variable selected frequency of over 190 runs for the TG ratio model of the lsquouprsquo
area of the soaking zone
0 2 4 6 8 10 12 14 16 18 20-05
05
AG
P-u
0 2 4 6 8 10 12 14 16 18 20-05
05
GP
-u
0 2 4 6 8 10 12 14 16 18 20-1
0
AG
P-d
0 2 4 6 8 10 12 14 16 18 20-1
0
GP
-d
0 2 4 6 8 10 12 14 16 18 20-05
05
AG
1-u
0 2 4 6 8 10 12 14 16 18 20-05
05
G1-
u
0 2 4 6 8 10 12 14 16 18 20-05
05
AG
1-d
0 2 4 6 8 10 12 14 16 18 20-05
05
G1-
d
0 2 4 6 8 10 12 14 16 18 20-1
0
AG
2-u
0 2 4 6 8 10 12 14 16 18 200
1
G2-
u
0 2 4 6 8 10 12 14 16 18 20-1
0
AG
2-d
0 2 4 6 8 10 12 14 16 18 20-1
0
G2-
d
0 2 4 6 8 10 12 14 16 18 200
1
AG
S-u
0 2 4 6 8 10 12 14 16 18 20-1
0
GS
-u
0 2 4 6 8 10 12 14 16 18 20-05
05
AG
S-d
0 2 4 6 8 10 12 14 16 18 20-05
05
GS
-d
Fig 10 Part of the NNG regression coefficients of each variable over 190 runs for the
TG ratio model of the lsquouprsquo area of the soaking zone
It is remarkable that the NNG regression coefficients corresponding to the AG ratio
in the lsquouprsquo area of the soaking zone are consistently positive On the contrary the
coefficients of the gas flow rate in the same area are consistently negative This
indicates that under this condition the rise of the AG leads to the increase of the TG
17
283
284
285
286
287
288
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
while the rise of the G leads to the drop of the TG This property is helpful for
performing optimization and improve combustion efficiency
For comparison the modeling results for two other algorithms artificial neural
network (ANN) and partial least squares (PLS) [26] are compared with the NNG
algorithm in the following three aspects [27]
(1) Model Size the number of variables selected for modeling
(2) Model Magnitude the mean of the L1 norm of the regression coefficients
(3) Prediction Precision the mean squared prediction error (MSPE)
Summary of the algorithm comparison is shown in Table 2 It can be seen that the
superiorities of the NNG regression in model size model magnitude and model
precision are remarkable
Table 2 Summary of algorithm comparison
Index NNG ANN PLS
Model Size 849 16 16
Model Magnitude 10075 -- 12150
MSPE 00093 00108 00140
5 Model-based optimization
51 Implemention of model-based optimization operation
The goal of optimization is to seek an optimal combination of AG and G in the lsquouprsquo
area of the soaking zone so as to minimize the gas consumption while keeping the
temperature at the target value According to the heating schedule the most expected
temperature in the soaking zone is 1200 Therefore the target temperature
is set at this value to achieve the desired heating effect As shown in Fig 11 in the real
18
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
operation the temperature fluctuates around the target value because of the imperfect
control performance However in order to simplify the analysis and compute the
maximum possible energy saving perfect temperature control is assumed when
adopting the model-based optimization strategy In other words it is assumed that the
actual temperature in the lsquouprsquo area of the soaking zone can be adjusted to the expected
temperature ie 1200
0 02 04 06 08 1 12 14 16 18 2
x 104
1150
1200
1250
1300
Sampling intervals
T S-u (
)
OriginalTarget
Fig 11 Original and target temperature in the lsquouprsquo area of the soaking zone
The adjustment scheme takes the maximum value and minimum value of the original
operation data as the upper and lower bounds for the adjustment Moreover in order to
assure the validity of the linear model the increment or decrement of AG and G should
not beyond of the original value (considered as 10 for the purposes of this paper)
The adjustment strategy can be formulated as
19
307
308
309
310
311
312
313
314
315
316
317
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
(3)
where NNG() denotes the NNG regression model of the TG ratio and are the
NNG regression coefficients of the AG and G in the lsquouprsquo area of the soaking zone for
modeling TG in same area
With the model-based strategy and adaptive model of the TG ratio given in Section
42 the optimization adjustment results can be obtained as follows
The adjustment of gas flow in the lsquouprsquo area of the soaking zone shown in Fig 12
illustrates that the implementation of the model-based optimization operation can reduce
the consumption of the gas flow Compared with the original case 938 of the gas can
be saved on average by utilizing the model-based adjustment
0 02 04 06 08 1 12 14 16 18 2
x 104
-800
-700
-600
-500
-400
-300
-200
-100
0
Sampling intervals
Adj
ustm
ent o
f gas
flow
m(
3 h)
20
318
319
320
321
322
323
324
325
326
327
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
(a) Adjustment amount of gas flow
0 02 04 06 08 1 12 14 16 18 2
x 104
500
1000
1500
2000
2500
3000
3500
4000
4500
5000G
as fl
owm
(3 h
)
Sampling intervals
OriginalAdjusted
(b) Comparison of original and adjusted gas flow
095 096 097 098 099 1 101 102 103 104 105
x 104
500
1000
1500
2000
2500
3000
3500
4000
4500
5000
Gas
flow
m(
3 h)
Sampling intervals
OriginalAdjusted
(c) Comparison of original and adjusted gas flow in interval [9501-10500]
Fig 12 Adjustment of gas flow in the lsquouprsquo area of the soaking zone
The adjustment of the air flow in the lsquouprsquo area of the soaking zone given in Fig 13
shows that the air flow is reduced in most cases and only increased for a few cases In
21
328
329
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
general 681 of the air flow is reduced after the implementation of the model-based
optimization
0 02 04 06 08 1 12 14 16 18 2
x 104
-2000
-1500
-1000
-500
0
500
1000
Sampling intervals
Adj
ustm
ent o
f air
flow
(m3 h
)
(a) Adjustment amount of air flow
0 02 04 06 08 1 12 14 16 18 2
x 104
4000
5000
6000
7000
8000
9000
10000
Air
flow
m(
3 h)
Sampling intervals
OriginalAdjusted
(b) Comparison of original and adjusted air flow
22
330
331
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
095 096 097 098 099 1 101 102 103 104 105
x 104
4000
5000
6000
7000
8000
9000
10000
Air
flow
m(
3 h)
Sampling intervals
OriginalAdjusted
(c) Comparison of original and adjusted air flow in interval [9501-10500]
Fig 13 Adjustment of air flow in the lsquouprsquo area of the soaking zone
52 Discussions
In this section detailed analysis for the optimization operation results is provided
The statistics of optimization at the boundary conditions shown in Table 3 illustrates
that the adjusted value for G reaches its lower limit in most cases (8045 for
and 687 for ) which is the most energy-efficient point In a number of cases
(1079 for and 001 for ) the adjusted AG reaches its upper
limit These results indicate that the optimization operations maximize the combustion
efficiency by decreasing G and increasing the AG ratio up to the boundary conditions
However the lower limit is also unexpectedly attained in a small number of cases
(110 for and 078 for ) for which a more detailed analysis is
given below
23
332
333
334
335
336
337
338
339
340
341
342
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
Table 3 Statistics of optimization at boundary conditions
Total
Amount 0 1374 0 16090 17464
Percentag
e 0 687 0 8045
8732
Total
Amount 2 220 2158 156 2536
Percentag
e 001 110 1079 078
1268
Table 4 Statistics for different optimization operations
Adjustment
Gdarr
Guarr TotalAGuarr AGdarr
Tdarr Tuarr Tdarr Tuarr
Amount 9548 3744 6328 380 0 20000
Percentage 4774 1872 3164 190 0 100
The statistics for different optimization operations is performed and shown in Table
4 It can be seen that all the adjusted operations result in the reduction of gas
consumption In over half the cases (4774 with decreased temperature and 1872
with increased temperature) AG is adjusted to higher levels Meanwhile a number of
the adjustments (3164) lower the temperature by reducing the G and AG at the same
time This result indicates that the original temperature cannot be adjusted to the target
by only reducing G in the constraint conditions The remaining cases (19) are
24
343
344
345
346
347
348
349
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
relatively special in which an excess of air is supplied in the original operation and the
adjusted operation thereby increasing the temperature by a smaller AG with less gas
consumption
6 Conclusion
Based on the actual operation data this paper aims to explore the improvement of the
combustion efficiency and the room for energy conservation Correlation analysis and
LDA show that it is of great significance to introduce two derived ratio variables which
are the AG ratio and the TG ratio A type of combustion efficiency utilizing an on-line
soft sensor is put forward by employing a NNG variable selection algorithm which
provides a good solution to the combustion efficiency real-time measurement problem
of a reheating furnace The implementation of the model-based optimization is studied
based on the actual operational data Detailed analysis for the optimization results is
given for the different cases The results show that significant energy conservation can
be achieved when the furnace operation is optimized by using the developed soft sensor
model
Acknowledgement
The authors would like to thank the financial support provided by the National Nature
Science Foundation of China under Grant 61171145 Y Yao was supported by Ministry
of Science amp Technology ROC under Grant number MOST 104-2221-E-007-129
References
25
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[1] Z J Wang Q D Wu and T Y Chai Optimal-setting control for complicated
industrial processes and its application studyControl Engineering Practice vol
12 pp 65-74 2004
[2] A Steinboeck K Graichen and A Kugi Dynamic Optimization of a Slab
Reheating Furnace With Consistent Approximation of Control VariablesIEEE
Transactions on Control Systems Technology vol 19 pp 1444-1456 2011
[3] B T Zhang C Y Wang Q Qin and L Li Influence of Boiler Combustion
Adjustment on NOxEmission and Boiler EfficiencyAdvanced Materials
Research vol 732-733 pp 234-237 2013
[4] C K Yoo and IB Lee Soft Sensor and Adaptive Model-Based Dissolved
Oxygen Control for Biological Wastewater Treatment ProcessesEnvironmental
Engineering Science vol 21 pp 331-340 2004
[5] S A Bhat D N Saraf S Gupta and S K Gupta Use of Agitator Power as a
Soft Sensor for Bulk Free-Radical Polymerization of Methyl Methacrylate in
Batch ReactorsIndustrial amp Engineering Chemistry Research vol 45 pp 4243-
4255 2006
[6] K Desai Y Badhe S S Tambe and B D Kulkarni Soft-sensor development
for fed-batch bioreactors using support vector regressionBiochemical
Engineering Journal vol 27 pp 225-239 2006
[7] Y P Badhe Lonari J Tambe S S amp Kulkarni B D Improve polyethylene
process control and product qualityHydrocarbon Processing vol 86 pp 53-60
2007
26
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[8] N K Nath K Mandal A K Singh B Basu C Bhanu S Kumar et al Ladle
furnace on-line reckoner for prediction and control of steel temperature and
compositionIronmaking amp Steelmaking vol 33 pp 140-150 2006
[9] A J Yan T Y Chai F H Wu and P Wang Hybrid intelligent control of
combustion process for ore-roasting furnaceJournal of Control Theory and
Applications vol 6 pp 80-85 2008
[10] J Li W M Zhong H Cheng X D Kong and F Qian A data-driven soft
sensor modeling for furnace temperature of Opposed Multi-Burner gasifier pp
705-710 2011
[11] Y H Yang Y H Liu X Z Liu and S K Qin Billet temperature soft sensor
model of reheating furnace based on RVM method pp 4003-4006 2011
[12] J H Wang C Wang X F Zhu and X K Fang Application of soft sensor in
welding seam tracking prediction based on LSSVM and PSO with compression
factor pp 2441-2446 2013
[13] L Balbis J Balderud and M J Grimble Nonlinear predictive control of steel
slab reheating furnace pp 1679-1684 2008
[14] A Steinboeck D Wild T Kiefer and A Kugi A mathematical model of a slab
reheating furnace with radiative heat transfer and non-participating gaseous
media International Journal Of Heat And Mass Transfer vol 53 pp 5933-
5946 Dec 2010
[15] A Steinboeck D Wild and A Kugi Nonlinear model predictive control of a
continuous slab reheating furnace Control Engineering Practice vol 21 pp
495-508 2013
27
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[16] C Zhang T Ishii and S Sugiyama Numerical Modeling Of the Thermal
Performance Of Regenerative Slab Reheat Furnaces Numerical Heat Transfer
Part A Applications vol 32 pp 613-631 1997
[17] J G Kim and K Y Huh Prediction of Transient Slab Temperature Distribution
in the Re-heating Furnace of a Walking-beam Type for Rolling of Steel Slabs
ISIJ International vol 40 pp 1115-1123 2000
[18] J G Kim K Y and H I T K Three-Dimensional Analysis Of the Walking-
Beam-Type Slab Reheating Furnace In Hot Strip Mills Numerical Heat
Transfer Part A Applications vol 38 pp 589-609 2000
[19] CT Hsieh MJ Huang ST Lee and CH Wang Numerical Modeling of a
Walking-Beam-Type Slab Reheating Furnace Numerical Heat Transfer Part A
Applications vol 53 pp 966-981 2008
[20] MJ Huang CT Hsieh ST Lee and CH Wang A Coupled Numerical Study
of Slab Temperature and Gas Temperature in the Walking-Beam-Type Slab
Reheating Furnace Numerical Heat Transfer Part A Applications vol 54 pp
625-646 2008
[21] Z Song and A Kusiak Constraint-Based Control of Boiler Efficiency A Data-
Mining Approach IEEE Transactions on Industrial Informatics vol 3 pp 73-
83 2007
[22] J Q Li J J Gu and C L Niu The Operation Optimization based on
Correlation Analysis of Operation Parameters in Power Plant pp 138-141
2008
28
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[23] J G Wang S S Shieh S S Jang D S H Wong and C W Wu A two-tier
approach to the data-driven modeling on thermal efficiency of a BFGcoal co-
firing boiler Fuel vol 111 pp 528-534 Sep 2013
[24] L Breiman Better Subset Regression Using the Nonnegative Garrote
Technometrics vol 37 pp 373-384 1995
[25] R A Fisher The use of multiple measurements in taxonomic problems
AnnHum Genet vol 7 pp 179-188 1936
[26] J Liu Developing a soft sensor based on sparse partial least squares with
variable selection Journal of Process Control vol 24 pp 1046-1056 2014
[27] K Sun J Liu JL Kang SS Jang D SH Wong and DS Chen
Development of a variable selection method for soft sensor using artificial
neural network and nonnegative garrote Journal of Process Control vol 24 pp
1068-1075 2014
29
439
440
441
442
443
444
445
446
447
448
449
450
451
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
AG_S-u-4
-2
0
2
Wei
ghtin
g fa
ctor
for y
1
AG_S-u
-101234
Wei
ghtin
g fa
ctor
for y
2
Fig 6 Weight factors of various input variables for projections y1 and y2
The two figures reveal that the three groups are clearly discriminated by the LDA
projection and the most significant variables for the different efficiency levels are the
derived variables ie the AG in the lsquouprsquo area of the soaking zone A similar result can
be obtained in the lsquodownrsquo area or any area of the other zones Therefore AG in each
heating area is the key manipulated variable that determines the different combustion
efficiency
4 Modeling and prediction of temperature and temperature-gas ratio
For the model-based operation optimization the models for the temperature and TG
ratio based on the NNG algorithm will be developed and compared
41 NNG variable selection algorithm
The NNG method can be generalized into a two-stage shrinkage method In the first
stage the sign for each variable is determined using the ordinary least square procedure
13
217
218
219
220
221
222
223
224
225
226
227
228
229
230
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
and in the second stage the corresponding magnitudes are computed by solving a series
of constrained quadratic programming
A set of observation data is provided where is the input matrix whose
columns represent the measured candidate variables and is the corresponding
vector of the response data The following expression is given with the number of the
response variable being equal to 1 but a similar procedure can be generalized to any
number of variables Let and be normalized to the zero-mean and unit standard
deviation Additionally let be a set of the ordinary least square estimates of the
coefficients of the following linear model then
(1)
The second stage shrinkage can be formulated as the following optimization problem
subject to
(2)
As decreased and the NNG is tightened more of the become zero and the
remaining nonzero coefficients are shrunk A solution path exists with on which
the appropriate shrinkage can be selected Conventionally the v-fold cross-validation is
used to estimate the prediction error and to select the best solution in the solution path
so as to minimize the prediction or model error
42 Modeling and prediction of temperature and TG ratio
The input-output relations change with time as the reheating process develops An
adaptive modeling strategy is often used to resolve time-varying characteristics of
14
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
industrial processes In this paper the NNG-based regression modeling prediction and
optimization is implemented in a moving window manner where the size of the
window and the length of the moving step are selected as 1000 and 100 respectively In
addition the prediction horizon is also set as 100 In each step the NNG regression
model is built based on the data in the current window which is then used for the
prediction and optimization in the subsequent prediction horizon Next the window
moves forward by replacing the oldest 100 samples for model training by the
observations collected in the previous prediction horizon This moving window strategy
is workable because in the reheating process the input-output relation is slowly time-
varying and the model is still valid for the prediction and optimization in the subsequent
short time
The modeling of the temperature T is based on the air flow rates and gas flow rates
while the modeling for the TG ratio is based on the AG ratios and gas flow rates
Taking the lsquouprsquo area of the soaking zone as an example the prediction result for T and
the TG ratio in the lsquouprsquo area of the soaking zone is shown in Figs 7 and 8 respectively
0 02 04 06 08 1 12 14 16 18 2
x 104
1140
1160
1180
1200
1220
1240
1260
1280
1300
1320
Sampling intervals
T S-u (
)
OriginalPredicted
Fig 7 Prediction of temperature in the lsquouprsquo area of the soaking zone
15
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
0 02 04 06 08 1 12 14 16 18 2
x 104
02
04
06
08
1
12
14
16
18
2
Sampling intervals
TG
S-u
OriginalPredicted
Fig 8 Prediction of TG ratio in the lsquouprsquo area of the soaking zone
A comparison between Figs 7 and 8 show that the prediction of the TG ratio is more
accurate than the prediction of temperature The average R2 in 190 NNG regressions is
0935 for the TG ratio while the average R2 is only 0814 for temperature This is
consistent with the statistical analysis which reveals that the correlation between AG
and TG is higher than that between A and T These results indicate that derived
variables are more meaningful for the purpose of prediction and the modeling of the
TG ratio is more appropriate for implementing optimization
The selected frequency for each variable in 190 NNG regressions and the coefficients
of each variable in 20 NNG regressions for the TG ratio modeling of the lsquouprsquo area of
the soaking zone are shown in Figs 9 and 10 respectively Fig 9 shows that the selected
frequency of the variables in the lsquouprsquo area of the soaking zone is much higher than
variables in other zones Similarly as is shown in Fig 10 the NNG regression
coefficients of the variables in the lsquouprsquo area of the soaking zone are much larger than the
variables in the other areas These results indicated that variables in the lsquouprsquo area of
16
269
270
271
272
273
274
275
276
277
278
279
280
281
282
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
soaking zone are mostly contributed to the modeling of the TG ratio in same area
which is quite reasonable Similar results can be obtained for the other areas
0
20
40
60
80
100
120
140
160
180
200
Sel
ecte
d fre
quen
cy
AGP-u AGP-d AG1-u AG1-d AG2-u AG2-d AGS-u AGS-d GP-u GP-d G1-u G1-d G2-u G2-d GS-u GS-d
Fig 9 Variable selected frequency of over 190 runs for the TG ratio model of the lsquouprsquo
area of the soaking zone
0 2 4 6 8 10 12 14 16 18 20-05
05
AG
P-u
0 2 4 6 8 10 12 14 16 18 20-05
05
GP
-u
0 2 4 6 8 10 12 14 16 18 20-1
0
AG
P-d
0 2 4 6 8 10 12 14 16 18 20-1
0
GP
-d
0 2 4 6 8 10 12 14 16 18 20-05
05
AG
1-u
0 2 4 6 8 10 12 14 16 18 20-05
05
G1-
u
0 2 4 6 8 10 12 14 16 18 20-05
05
AG
1-d
0 2 4 6 8 10 12 14 16 18 20-05
05
G1-
d
0 2 4 6 8 10 12 14 16 18 20-1
0
AG
2-u
0 2 4 6 8 10 12 14 16 18 200
1
G2-
u
0 2 4 6 8 10 12 14 16 18 20-1
0
AG
2-d
0 2 4 6 8 10 12 14 16 18 20-1
0
G2-
d
0 2 4 6 8 10 12 14 16 18 200
1
AG
S-u
0 2 4 6 8 10 12 14 16 18 20-1
0
GS
-u
0 2 4 6 8 10 12 14 16 18 20-05
05
AG
S-d
0 2 4 6 8 10 12 14 16 18 20-05
05
GS
-d
Fig 10 Part of the NNG regression coefficients of each variable over 190 runs for the
TG ratio model of the lsquouprsquo area of the soaking zone
It is remarkable that the NNG regression coefficients corresponding to the AG ratio
in the lsquouprsquo area of the soaking zone are consistently positive On the contrary the
coefficients of the gas flow rate in the same area are consistently negative This
indicates that under this condition the rise of the AG leads to the increase of the TG
17
283
284
285
286
287
288
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
while the rise of the G leads to the drop of the TG This property is helpful for
performing optimization and improve combustion efficiency
For comparison the modeling results for two other algorithms artificial neural
network (ANN) and partial least squares (PLS) [26] are compared with the NNG
algorithm in the following three aspects [27]
(1) Model Size the number of variables selected for modeling
(2) Model Magnitude the mean of the L1 norm of the regression coefficients
(3) Prediction Precision the mean squared prediction error (MSPE)
Summary of the algorithm comparison is shown in Table 2 It can be seen that the
superiorities of the NNG regression in model size model magnitude and model
precision are remarkable
Table 2 Summary of algorithm comparison
Index NNG ANN PLS
Model Size 849 16 16
Model Magnitude 10075 -- 12150
MSPE 00093 00108 00140
5 Model-based optimization
51 Implemention of model-based optimization operation
The goal of optimization is to seek an optimal combination of AG and G in the lsquouprsquo
area of the soaking zone so as to minimize the gas consumption while keeping the
temperature at the target value According to the heating schedule the most expected
temperature in the soaking zone is 1200 Therefore the target temperature
is set at this value to achieve the desired heating effect As shown in Fig 11 in the real
18
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
operation the temperature fluctuates around the target value because of the imperfect
control performance However in order to simplify the analysis and compute the
maximum possible energy saving perfect temperature control is assumed when
adopting the model-based optimization strategy In other words it is assumed that the
actual temperature in the lsquouprsquo area of the soaking zone can be adjusted to the expected
temperature ie 1200
0 02 04 06 08 1 12 14 16 18 2
x 104
1150
1200
1250
1300
Sampling intervals
T S-u (
)
OriginalTarget
Fig 11 Original and target temperature in the lsquouprsquo area of the soaking zone
The adjustment scheme takes the maximum value and minimum value of the original
operation data as the upper and lower bounds for the adjustment Moreover in order to
assure the validity of the linear model the increment or decrement of AG and G should
not beyond of the original value (considered as 10 for the purposes of this paper)
The adjustment strategy can be formulated as
19
307
308
309
310
311
312
313
314
315
316
317
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
(3)
where NNG() denotes the NNG regression model of the TG ratio and are the
NNG regression coefficients of the AG and G in the lsquouprsquo area of the soaking zone for
modeling TG in same area
With the model-based strategy and adaptive model of the TG ratio given in Section
42 the optimization adjustment results can be obtained as follows
The adjustment of gas flow in the lsquouprsquo area of the soaking zone shown in Fig 12
illustrates that the implementation of the model-based optimization operation can reduce
the consumption of the gas flow Compared with the original case 938 of the gas can
be saved on average by utilizing the model-based adjustment
0 02 04 06 08 1 12 14 16 18 2
x 104
-800
-700
-600
-500
-400
-300
-200
-100
0
Sampling intervals
Adj
ustm
ent o
f gas
flow
m(
3 h)
20
318
319
320
321
322
323
324
325
326
327
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
(a) Adjustment amount of gas flow
0 02 04 06 08 1 12 14 16 18 2
x 104
500
1000
1500
2000
2500
3000
3500
4000
4500
5000G
as fl
owm
(3 h
)
Sampling intervals
OriginalAdjusted
(b) Comparison of original and adjusted gas flow
095 096 097 098 099 1 101 102 103 104 105
x 104
500
1000
1500
2000
2500
3000
3500
4000
4500
5000
Gas
flow
m(
3 h)
Sampling intervals
OriginalAdjusted
(c) Comparison of original and adjusted gas flow in interval [9501-10500]
Fig 12 Adjustment of gas flow in the lsquouprsquo area of the soaking zone
The adjustment of the air flow in the lsquouprsquo area of the soaking zone given in Fig 13
shows that the air flow is reduced in most cases and only increased for a few cases In
21
328
329
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
general 681 of the air flow is reduced after the implementation of the model-based
optimization
0 02 04 06 08 1 12 14 16 18 2
x 104
-2000
-1500
-1000
-500
0
500
1000
Sampling intervals
Adj
ustm
ent o
f air
flow
(m3 h
)
(a) Adjustment amount of air flow
0 02 04 06 08 1 12 14 16 18 2
x 104
4000
5000
6000
7000
8000
9000
10000
Air
flow
m(
3 h)
Sampling intervals
OriginalAdjusted
(b) Comparison of original and adjusted air flow
22
330
331
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
095 096 097 098 099 1 101 102 103 104 105
x 104
4000
5000
6000
7000
8000
9000
10000
Air
flow
m(
3 h)
Sampling intervals
OriginalAdjusted
(c) Comparison of original and adjusted air flow in interval [9501-10500]
Fig 13 Adjustment of air flow in the lsquouprsquo area of the soaking zone
52 Discussions
In this section detailed analysis for the optimization operation results is provided
The statistics of optimization at the boundary conditions shown in Table 3 illustrates
that the adjusted value for G reaches its lower limit in most cases (8045 for
and 687 for ) which is the most energy-efficient point In a number of cases
(1079 for and 001 for ) the adjusted AG reaches its upper
limit These results indicate that the optimization operations maximize the combustion
efficiency by decreasing G and increasing the AG ratio up to the boundary conditions
However the lower limit is also unexpectedly attained in a small number of cases
(110 for and 078 for ) for which a more detailed analysis is
given below
23
332
333
334
335
336
337
338
339
340
341
342
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
Table 3 Statistics of optimization at boundary conditions
Total
Amount 0 1374 0 16090 17464
Percentag
e 0 687 0 8045
8732
Total
Amount 2 220 2158 156 2536
Percentag
e 001 110 1079 078
1268
Table 4 Statistics for different optimization operations
Adjustment
Gdarr
Guarr TotalAGuarr AGdarr
Tdarr Tuarr Tdarr Tuarr
Amount 9548 3744 6328 380 0 20000
Percentage 4774 1872 3164 190 0 100
The statistics for different optimization operations is performed and shown in Table
4 It can be seen that all the adjusted operations result in the reduction of gas
consumption In over half the cases (4774 with decreased temperature and 1872
with increased temperature) AG is adjusted to higher levels Meanwhile a number of
the adjustments (3164) lower the temperature by reducing the G and AG at the same
time This result indicates that the original temperature cannot be adjusted to the target
by only reducing G in the constraint conditions The remaining cases (19) are
24
343
344
345
346
347
348
349
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
relatively special in which an excess of air is supplied in the original operation and the
adjusted operation thereby increasing the temperature by a smaller AG with less gas
consumption
6 Conclusion
Based on the actual operation data this paper aims to explore the improvement of the
combustion efficiency and the room for energy conservation Correlation analysis and
LDA show that it is of great significance to introduce two derived ratio variables which
are the AG ratio and the TG ratio A type of combustion efficiency utilizing an on-line
soft sensor is put forward by employing a NNG variable selection algorithm which
provides a good solution to the combustion efficiency real-time measurement problem
of a reheating furnace The implementation of the model-based optimization is studied
based on the actual operational data Detailed analysis for the optimization results is
given for the different cases The results show that significant energy conservation can
be achieved when the furnace operation is optimized by using the developed soft sensor
model
Acknowledgement
The authors would like to thank the financial support provided by the National Nature
Science Foundation of China under Grant 61171145 Y Yao was supported by Ministry
of Science amp Technology ROC under Grant number MOST 104-2221-E-007-129
References
25
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[1] Z J Wang Q D Wu and T Y Chai Optimal-setting control for complicated
industrial processes and its application studyControl Engineering Practice vol
12 pp 65-74 2004
[2] A Steinboeck K Graichen and A Kugi Dynamic Optimization of a Slab
Reheating Furnace With Consistent Approximation of Control VariablesIEEE
Transactions on Control Systems Technology vol 19 pp 1444-1456 2011
[3] B T Zhang C Y Wang Q Qin and L Li Influence of Boiler Combustion
Adjustment on NOxEmission and Boiler EfficiencyAdvanced Materials
Research vol 732-733 pp 234-237 2013
[4] C K Yoo and IB Lee Soft Sensor and Adaptive Model-Based Dissolved
Oxygen Control for Biological Wastewater Treatment ProcessesEnvironmental
Engineering Science vol 21 pp 331-340 2004
[5] S A Bhat D N Saraf S Gupta and S K Gupta Use of Agitator Power as a
Soft Sensor for Bulk Free-Radical Polymerization of Methyl Methacrylate in
Batch ReactorsIndustrial amp Engineering Chemistry Research vol 45 pp 4243-
4255 2006
[6] K Desai Y Badhe S S Tambe and B D Kulkarni Soft-sensor development
for fed-batch bioreactors using support vector regressionBiochemical
Engineering Journal vol 27 pp 225-239 2006
[7] Y P Badhe Lonari J Tambe S S amp Kulkarni B D Improve polyethylene
process control and product qualityHydrocarbon Processing vol 86 pp 53-60
2007
26
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[8] N K Nath K Mandal A K Singh B Basu C Bhanu S Kumar et al Ladle
furnace on-line reckoner for prediction and control of steel temperature and
compositionIronmaking amp Steelmaking vol 33 pp 140-150 2006
[9] A J Yan T Y Chai F H Wu and P Wang Hybrid intelligent control of
combustion process for ore-roasting furnaceJournal of Control Theory and
Applications vol 6 pp 80-85 2008
[10] J Li W M Zhong H Cheng X D Kong and F Qian A data-driven soft
sensor modeling for furnace temperature of Opposed Multi-Burner gasifier pp
705-710 2011
[11] Y H Yang Y H Liu X Z Liu and S K Qin Billet temperature soft sensor
model of reheating furnace based on RVM method pp 4003-4006 2011
[12] J H Wang C Wang X F Zhu and X K Fang Application of soft sensor in
welding seam tracking prediction based on LSSVM and PSO with compression
factor pp 2441-2446 2013
[13] L Balbis J Balderud and M J Grimble Nonlinear predictive control of steel
slab reheating furnace pp 1679-1684 2008
[14] A Steinboeck D Wild T Kiefer and A Kugi A mathematical model of a slab
reheating furnace with radiative heat transfer and non-participating gaseous
media International Journal Of Heat And Mass Transfer vol 53 pp 5933-
5946 Dec 2010
[15] A Steinboeck D Wild and A Kugi Nonlinear model predictive control of a
continuous slab reheating furnace Control Engineering Practice vol 21 pp
495-508 2013
27
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[16] C Zhang T Ishii and S Sugiyama Numerical Modeling Of the Thermal
Performance Of Regenerative Slab Reheat Furnaces Numerical Heat Transfer
Part A Applications vol 32 pp 613-631 1997
[17] J G Kim and K Y Huh Prediction of Transient Slab Temperature Distribution
in the Re-heating Furnace of a Walking-beam Type for Rolling of Steel Slabs
ISIJ International vol 40 pp 1115-1123 2000
[18] J G Kim K Y and H I T K Three-Dimensional Analysis Of the Walking-
Beam-Type Slab Reheating Furnace In Hot Strip Mills Numerical Heat
Transfer Part A Applications vol 38 pp 589-609 2000
[19] CT Hsieh MJ Huang ST Lee and CH Wang Numerical Modeling of a
Walking-Beam-Type Slab Reheating Furnace Numerical Heat Transfer Part A
Applications vol 53 pp 966-981 2008
[20] MJ Huang CT Hsieh ST Lee and CH Wang A Coupled Numerical Study
of Slab Temperature and Gas Temperature in the Walking-Beam-Type Slab
Reheating Furnace Numerical Heat Transfer Part A Applications vol 54 pp
625-646 2008
[21] Z Song and A Kusiak Constraint-Based Control of Boiler Efficiency A Data-
Mining Approach IEEE Transactions on Industrial Informatics vol 3 pp 73-
83 2007
[22] J Q Li J J Gu and C L Niu The Operation Optimization based on
Correlation Analysis of Operation Parameters in Power Plant pp 138-141
2008
28
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[23] J G Wang S S Shieh S S Jang D S H Wong and C W Wu A two-tier
approach to the data-driven modeling on thermal efficiency of a BFGcoal co-
firing boiler Fuel vol 111 pp 528-534 Sep 2013
[24] L Breiman Better Subset Regression Using the Nonnegative Garrote
Technometrics vol 37 pp 373-384 1995
[25] R A Fisher The use of multiple measurements in taxonomic problems
AnnHum Genet vol 7 pp 179-188 1936
[26] J Liu Developing a soft sensor based on sparse partial least squares with
variable selection Journal of Process Control vol 24 pp 1046-1056 2014
[27] K Sun J Liu JL Kang SS Jang D SH Wong and DS Chen
Development of a variable selection method for soft sensor using artificial
neural network and nonnegative garrote Journal of Process Control vol 24 pp
1068-1075 2014
29
439
440
441
442
443
444
445
446
447
448
449
450
451
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
and in the second stage the corresponding magnitudes are computed by solving a series
of constrained quadratic programming
A set of observation data is provided where is the input matrix whose
columns represent the measured candidate variables and is the corresponding
vector of the response data The following expression is given with the number of the
response variable being equal to 1 but a similar procedure can be generalized to any
number of variables Let and be normalized to the zero-mean and unit standard
deviation Additionally let be a set of the ordinary least square estimates of the
coefficients of the following linear model then
(1)
The second stage shrinkage can be formulated as the following optimization problem
subject to
(2)
As decreased and the NNG is tightened more of the become zero and the
remaining nonzero coefficients are shrunk A solution path exists with on which
the appropriate shrinkage can be selected Conventionally the v-fold cross-validation is
used to estimate the prediction error and to select the best solution in the solution path
so as to minimize the prediction or model error
42 Modeling and prediction of temperature and TG ratio
The input-output relations change with time as the reheating process develops An
adaptive modeling strategy is often used to resolve time-varying characteristics of
14
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
industrial processes In this paper the NNG-based regression modeling prediction and
optimization is implemented in a moving window manner where the size of the
window and the length of the moving step are selected as 1000 and 100 respectively In
addition the prediction horizon is also set as 100 In each step the NNG regression
model is built based on the data in the current window which is then used for the
prediction and optimization in the subsequent prediction horizon Next the window
moves forward by replacing the oldest 100 samples for model training by the
observations collected in the previous prediction horizon This moving window strategy
is workable because in the reheating process the input-output relation is slowly time-
varying and the model is still valid for the prediction and optimization in the subsequent
short time
The modeling of the temperature T is based on the air flow rates and gas flow rates
while the modeling for the TG ratio is based on the AG ratios and gas flow rates
Taking the lsquouprsquo area of the soaking zone as an example the prediction result for T and
the TG ratio in the lsquouprsquo area of the soaking zone is shown in Figs 7 and 8 respectively
0 02 04 06 08 1 12 14 16 18 2
x 104
1140
1160
1180
1200
1220
1240
1260
1280
1300
1320
Sampling intervals
T S-u (
)
OriginalPredicted
Fig 7 Prediction of temperature in the lsquouprsquo area of the soaking zone
15
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
0 02 04 06 08 1 12 14 16 18 2
x 104
02
04
06
08
1
12
14
16
18
2
Sampling intervals
TG
S-u
OriginalPredicted
Fig 8 Prediction of TG ratio in the lsquouprsquo area of the soaking zone
A comparison between Figs 7 and 8 show that the prediction of the TG ratio is more
accurate than the prediction of temperature The average R2 in 190 NNG regressions is
0935 for the TG ratio while the average R2 is only 0814 for temperature This is
consistent with the statistical analysis which reveals that the correlation between AG
and TG is higher than that between A and T These results indicate that derived
variables are more meaningful for the purpose of prediction and the modeling of the
TG ratio is more appropriate for implementing optimization
The selected frequency for each variable in 190 NNG regressions and the coefficients
of each variable in 20 NNG regressions for the TG ratio modeling of the lsquouprsquo area of
the soaking zone are shown in Figs 9 and 10 respectively Fig 9 shows that the selected
frequency of the variables in the lsquouprsquo area of the soaking zone is much higher than
variables in other zones Similarly as is shown in Fig 10 the NNG regression
coefficients of the variables in the lsquouprsquo area of the soaking zone are much larger than the
variables in the other areas These results indicated that variables in the lsquouprsquo area of
16
269
270
271
272
273
274
275
276
277
278
279
280
281
282
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
soaking zone are mostly contributed to the modeling of the TG ratio in same area
which is quite reasonable Similar results can be obtained for the other areas
0
20
40
60
80
100
120
140
160
180
200
Sel
ecte
d fre
quen
cy
AGP-u AGP-d AG1-u AG1-d AG2-u AG2-d AGS-u AGS-d GP-u GP-d G1-u G1-d G2-u G2-d GS-u GS-d
Fig 9 Variable selected frequency of over 190 runs for the TG ratio model of the lsquouprsquo
area of the soaking zone
0 2 4 6 8 10 12 14 16 18 20-05
05
AG
P-u
0 2 4 6 8 10 12 14 16 18 20-05
05
GP
-u
0 2 4 6 8 10 12 14 16 18 20-1
0
AG
P-d
0 2 4 6 8 10 12 14 16 18 20-1
0
GP
-d
0 2 4 6 8 10 12 14 16 18 20-05
05
AG
1-u
0 2 4 6 8 10 12 14 16 18 20-05
05
G1-
u
0 2 4 6 8 10 12 14 16 18 20-05
05
AG
1-d
0 2 4 6 8 10 12 14 16 18 20-05
05
G1-
d
0 2 4 6 8 10 12 14 16 18 20-1
0
AG
2-u
0 2 4 6 8 10 12 14 16 18 200
1
G2-
u
0 2 4 6 8 10 12 14 16 18 20-1
0
AG
2-d
0 2 4 6 8 10 12 14 16 18 20-1
0
G2-
d
0 2 4 6 8 10 12 14 16 18 200
1
AG
S-u
0 2 4 6 8 10 12 14 16 18 20-1
0
GS
-u
0 2 4 6 8 10 12 14 16 18 20-05
05
AG
S-d
0 2 4 6 8 10 12 14 16 18 20-05
05
GS
-d
Fig 10 Part of the NNG regression coefficients of each variable over 190 runs for the
TG ratio model of the lsquouprsquo area of the soaking zone
It is remarkable that the NNG regression coefficients corresponding to the AG ratio
in the lsquouprsquo area of the soaking zone are consistently positive On the contrary the
coefficients of the gas flow rate in the same area are consistently negative This
indicates that under this condition the rise of the AG leads to the increase of the TG
17
283
284
285
286
287
288
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
while the rise of the G leads to the drop of the TG This property is helpful for
performing optimization and improve combustion efficiency
For comparison the modeling results for two other algorithms artificial neural
network (ANN) and partial least squares (PLS) [26] are compared with the NNG
algorithm in the following three aspects [27]
(1) Model Size the number of variables selected for modeling
(2) Model Magnitude the mean of the L1 norm of the regression coefficients
(3) Prediction Precision the mean squared prediction error (MSPE)
Summary of the algorithm comparison is shown in Table 2 It can be seen that the
superiorities of the NNG regression in model size model magnitude and model
precision are remarkable
Table 2 Summary of algorithm comparison
Index NNG ANN PLS
Model Size 849 16 16
Model Magnitude 10075 -- 12150
MSPE 00093 00108 00140
5 Model-based optimization
51 Implemention of model-based optimization operation
The goal of optimization is to seek an optimal combination of AG and G in the lsquouprsquo
area of the soaking zone so as to minimize the gas consumption while keeping the
temperature at the target value According to the heating schedule the most expected
temperature in the soaking zone is 1200 Therefore the target temperature
is set at this value to achieve the desired heating effect As shown in Fig 11 in the real
18
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
operation the temperature fluctuates around the target value because of the imperfect
control performance However in order to simplify the analysis and compute the
maximum possible energy saving perfect temperature control is assumed when
adopting the model-based optimization strategy In other words it is assumed that the
actual temperature in the lsquouprsquo area of the soaking zone can be adjusted to the expected
temperature ie 1200
0 02 04 06 08 1 12 14 16 18 2
x 104
1150
1200
1250
1300
Sampling intervals
T S-u (
)
OriginalTarget
Fig 11 Original and target temperature in the lsquouprsquo area of the soaking zone
The adjustment scheme takes the maximum value and minimum value of the original
operation data as the upper and lower bounds for the adjustment Moreover in order to
assure the validity of the linear model the increment or decrement of AG and G should
not beyond of the original value (considered as 10 for the purposes of this paper)
The adjustment strategy can be formulated as
19
307
308
309
310
311
312
313
314
315
316
317
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
(3)
where NNG() denotes the NNG regression model of the TG ratio and are the
NNG regression coefficients of the AG and G in the lsquouprsquo area of the soaking zone for
modeling TG in same area
With the model-based strategy and adaptive model of the TG ratio given in Section
42 the optimization adjustment results can be obtained as follows
The adjustment of gas flow in the lsquouprsquo area of the soaking zone shown in Fig 12
illustrates that the implementation of the model-based optimization operation can reduce
the consumption of the gas flow Compared with the original case 938 of the gas can
be saved on average by utilizing the model-based adjustment
0 02 04 06 08 1 12 14 16 18 2
x 104
-800
-700
-600
-500
-400
-300
-200
-100
0
Sampling intervals
Adj
ustm
ent o
f gas
flow
m(
3 h)
20
318
319
320
321
322
323
324
325
326
327
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
(a) Adjustment amount of gas flow
0 02 04 06 08 1 12 14 16 18 2
x 104
500
1000
1500
2000
2500
3000
3500
4000
4500
5000G
as fl
owm
(3 h
)
Sampling intervals
OriginalAdjusted
(b) Comparison of original and adjusted gas flow
095 096 097 098 099 1 101 102 103 104 105
x 104
500
1000
1500
2000
2500
3000
3500
4000
4500
5000
Gas
flow
m(
3 h)
Sampling intervals
OriginalAdjusted
(c) Comparison of original and adjusted gas flow in interval [9501-10500]
Fig 12 Adjustment of gas flow in the lsquouprsquo area of the soaking zone
The adjustment of the air flow in the lsquouprsquo area of the soaking zone given in Fig 13
shows that the air flow is reduced in most cases and only increased for a few cases In
21
328
329
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
general 681 of the air flow is reduced after the implementation of the model-based
optimization
0 02 04 06 08 1 12 14 16 18 2
x 104
-2000
-1500
-1000
-500
0
500
1000
Sampling intervals
Adj
ustm
ent o
f air
flow
(m3 h
)
(a) Adjustment amount of air flow
0 02 04 06 08 1 12 14 16 18 2
x 104
4000
5000
6000
7000
8000
9000
10000
Air
flow
m(
3 h)
Sampling intervals
OriginalAdjusted
(b) Comparison of original and adjusted air flow
22
330
331
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
095 096 097 098 099 1 101 102 103 104 105
x 104
4000
5000
6000
7000
8000
9000
10000
Air
flow
m(
3 h)
Sampling intervals
OriginalAdjusted
(c) Comparison of original and adjusted air flow in interval [9501-10500]
Fig 13 Adjustment of air flow in the lsquouprsquo area of the soaking zone
52 Discussions
In this section detailed analysis for the optimization operation results is provided
The statistics of optimization at the boundary conditions shown in Table 3 illustrates
that the adjusted value for G reaches its lower limit in most cases (8045 for
and 687 for ) which is the most energy-efficient point In a number of cases
(1079 for and 001 for ) the adjusted AG reaches its upper
limit These results indicate that the optimization operations maximize the combustion
efficiency by decreasing G and increasing the AG ratio up to the boundary conditions
However the lower limit is also unexpectedly attained in a small number of cases
(110 for and 078 for ) for which a more detailed analysis is
given below
23
332
333
334
335
336
337
338
339
340
341
342
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
Table 3 Statistics of optimization at boundary conditions
Total
Amount 0 1374 0 16090 17464
Percentag
e 0 687 0 8045
8732
Total
Amount 2 220 2158 156 2536
Percentag
e 001 110 1079 078
1268
Table 4 Statistics for different optimization operations
Adjustment
Gdarr
Guarr TotalAGuarr AGdarr
Tdarr Tuarr Tdarr Tuarr
Amount 9548 3744 6328 380 0 20000
Percentage 4774 1872 3164 190 0 100
The statistics for different optimization operations is performed and shown in Table
4 It can be seen that all the adjusted operations result in the reduction of gas
consumption In over half the cases (4774 with decreased temperature and 1872
with increased temperature) AG is adjusted to higher levels Meanwhile a number of
the adjustments (3164) lower the temperature by reducing the G and AG at the same
time This result indicates that the original temperature cannot be adjusted to the target
by only reducing G in the constraint conditions The remaining cases (19) are
24
343
344
345
346
347
348
349
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
relatively special in which an excess of air is supplied in the original operation and the
adjusted operation thereby increasing the temperature by a smaller AG with less gas
consumption
6 Conclusion
Based on the actual operation data this paper aims to explore the improvement of the
combustion efficiency and the room for energy conservation Correlation analysis and
LDA show that it is of great significance to introduce two derived ratio variables which
are the AG ratio and the TG ratio A type of combustion efficiency utilizing an on-line
soft sensor is put forward by employing a NNG variable selection algorithm which
provides a good solution to the combustion efficiency real-time measurement problem
of a reheating furnace The implementation of the model-based optimization is studied
based on the actual operational data Detailed analysis for the optimization results is
given for the different cases The results show that significant energy conservation can
be achieved when the furnace operation is optimized by using the developed soft sensor
model
Acknowledgement
The authors would like to thank the financial support provided by the National Nature
Science Foundation of China under Grant 61171145 Y Yao was supported by Ministry
of Science amp Technology ROC under Grant number MOST 104-2221-E-007-129
References
25
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[1] Z J Wang Q D Wu and T Y Chai Optimal-setting control for complicated
industrial processes and its application studyControl Engineering Practice vol
12 pp 65-74 2004
[2] A Steinboeck K Graichen and A Kugi Dynamic Optimization of a Slab
Reheating Furnace With Consistent Approximation of Control VariablesIEEE
Transactions on Control Systems Technology vol 19 pp 1444-1456 2011
[3] B T Zhang C Y Wang Q Qin and L Li Influence of Boiler Combustion
Adjustment on NOxEmission and Boiler EfficiencyAdvanced Materials
Research vol 732-733 pp 234-237 2013
[4] C K Yoo and IB Lee Soft Sensor and Adaptive Model-Based Dissolved
Oxygen Control for Biological Wastewater Treatment ProcessesEnvironmental
Engineering Science vol 21 pp 331-340 2004
[5] S A Bhat D N Saraf S Gupta and S K Gupta Use of Agitator Power as a
Soft Sensor for Bulk Free-Radical Polymerization of Methyl Methacrylate in
Batch ReactorsIndustrial amp Engineering Chemistry Research vol 45 pp 4243-
4255 2006
[6] K Desai Y Badhe S S Tambe and B D Kulkarni Soft-sensor development
for fed-batch bioreactors using support vector regressionBiochemical
Engineering Journal vol 27 pp 225-239 2006
[7] Y P Badhe Lonari J Tambe S S amp Kulkarni B D Improve polyethylene
process control and product qualityHydrocarbon Processing vol 86 pp 53-60
2007
26
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[8] N K Nath K Mandal A K Singh B Basu C Bhanu S Kumar et al Ladle
furnace on-line reckoner for prediction and control of steel temperature and
compositionIronmaking amp Steelmaking vol 33 pp 140-150 2006
[9] A J Yan T Y Chai F H Wu and P Wang Hybrid intelligent control of
combustion process for ore-roasting furnaceJournal of Control Theory and
Applications vol 6 pp 80-85 2008
[10] J Li W M Zhong H Cheng X D Kong and F Qian A data-driven soft
sensor modeling for furnace temperature of Opposed Multi-Burner gasifier pp
705-710 2011
[11] Y H Yang Y H Liu X Z Liu and S K Qin Billet temperature soft sensor
model of reheating furnace based on RVM method pp 4003-4006 2011
[12] J H Wang C Wang X F Zhu and X K Fang Application of soft sensor in
welding seam tracking prediction based on LSSVM and PSO with compression
factor pp 2441-2446 2013
[13] L Balbis J Balderud and M J Grimble Nonlinear predictive control of steel
slab reheating furnace pp 1679-1684 2008
[14] A Steinboeck D Wild T Kiefer and A Kugi A mathematical model of a slab
reheating furnace with radiative heat transfer and non-participating gaseous
media International Journal Of Heat And Mass Transfer vol 53 pp 5933-
5946 Dec 2010
[15] A Steinboeck D Wild and A Kugi Nonlinear model predictive control of a
continuous slab reheating furnace Control Engineering Practice vol 21 pp
495-508 2013
27
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[16] C Zhang T Ishii and S Sugiyama Numerical Modeling Of the Thermal
Performance Of Regenerative Slab Reheat Furnaces Numerical Heat Transfer
Part A Applications vol 32 pp 613-631 1997
[17] J G Kim and K Y Huh Prediction of Transient Slab Temperature Distribution
in the Re-heating Furnace of a Walking-beam Type for Rolling of Steel Slabs
ISIJ International vol 40 pp 1115-1123 2000
[18] J G Kim K Y and H I T K Three-Dimensional Analysis Of the Walking-
Beam-Type Slab Reheating Furnace In Hot Strip Mills Numerical Heat
Transfer Part A Applications vol 38 pp 589-609 2000
[19] CT Hsieh MJ Huang ST Lee and CH Wang Numerical Modeling of a
Walking-Beam-Type Slab Reheating Furnace Numerical Heat Transfer Part A
Applications vol 53 pp 966-981 2008
[20] MJ Huang CT Hsieh ST Lee and CH Wang A Coupled Numerical Study
of Slab Temperature and Gas Temperature in the Walking-Beam-Type Slab
Reheating Furnace Numerical Heat Transfer Part A Applications vol 54 pp
625-646 2008
[21] Z Song and A Kusiak Constraint-Based Control of Boiler Efficiency A Data-
Mining Approach IEEE Transactions on Industrial Informatics vol 3 pp 73-
83 2007
[22] J Q Li J J Gu and C L Niu The Operation Optimization based on
Correlation Analysis of Operation Parameters in Power Plant pp 138-141
2008
28
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[23] J G Wang S S Shieh S S Jang D S H Wong and C W Wu A two-tier
approach to the data-driven modeling on thermal efficiency of a BFGcoal co-
firing boiler Fuel vol 111 pp 528-534 Sep 2013
[24] L Breiman Better Subset Regression Using the Nonnegative Garrote
Technometrics vol 37 pp 373-384 1995
[25] R A Fisher The use of multiple measurements in taxonomic problems
AnnHum Genet vol 7 pp 179-188 1936
[26] J Liu Developing a soft sensor based on sparse partial least squares with
variable selection Journal of Process Control vol 24 pp 1046-1056 2014
[27] K Sun J Liu JL Kang SS Jang D SH Wong and DS Chen
Development of a variable selection method for soft sensor using artificial
neural network and nonnegative garrote Journal of Process Control vol 24 pp
1068-1075 2014
29
439
440
441
442
443
444
445
446
447
448
449
450
451
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
industrial processes In this paper the NNG-based regression modeling prediction and
optimization is implemented in a moving window manner where the size of the
window and the length of the moving step are selected as 1000 and 100 respectively In
addition the prediction horizon is also set as 100 In each step the NNG regression
model is built based on the data in the current window which is then used for the
prediction and optimization in the subsequent prediction horizon Next the window
moves forward by replacing the oldest 100 samples for model training by the
observations collected in the previous prediction horizon This moving window strategy
is workable because in the reheating process the input-output relation is slowly time-
varying and the model is still valid for the prediction and optimization in the subsequent
short time
The modeling of the temperature T is based on the air flow rates and gas flow rates
while the modeling for the TG ratio is based on the AG ratios and gas flow rates
Taking the lsquouprsquo area of the soaking zone as an example the prediction result for T and
the TG ratio in the lsquouprsquo area of the soaking zone is shown in Figs 7 and 8 respectively
0 02 04 06 08 1 12 14 16 18 2
x 104
1140
1160
1180
1200
1220
1240
1260
1280
1300
1320
Sampling intervals
T S-u (
)
OriginalPredicted
Fig 7 Prediction of temperature in the lsquouprsquo area of the soaking zone
15
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
0 02 04 06 08 1 12 14 16 18 2
x 104
02
04
06
08
1
12
14
16
18
2
Sampling intervals
TG
S-u
OriginalPredicted
Fig 8 Prediction of TG ratio in the lsquouprsquo area of the soaking zone
A comparison between Figs 7 and 8 show that the prediction of the TG ratio is more
accurate than the prediction of temperature The average R2 in 190 NNG regressions is
0935 for the TG ratio while the average R2 is only 0814 for temperature This is
consistent with the statistical analysis which reveals that the correlation between AG
and TG is higher than that between A and T These results indicate that derived
variables are more meaningful for the purpose of prediction and the modeling of the
TG ratio is more appropriate for implementing optimization
The selected frequency for each variable in 190 NNG regressions and the coefficients
of each variable in 20 NNG regressions for the TG ratio modeling of the lsquouprsquo area of
the soaking zone are shown in Figs 9 and 10 respectively Fig 9 shows that the selected
frequency of the variables in the lsquouprsquo area of the soaking zone is much higher than
variables in other zones Similarly as is shown in Fig 10 the NNG regression
coefficients of the variables in the lsquouprsquo area of the soaking zone are much larger than the
variables in the other areas These results indicated that variables in the lsquouprsquo area of
16
269
270
271
272
273
274
275
276
277
278
279
280
281
282
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
soaking zone are mostly contributed to the modeling of the TG ratio in same area
which is quite reasonable Similar results can be obtained for the other areas
0
20
40
60
80
100
120
140
160
180
200
Sel
ecte
d fre
quen
cy
AGP-u AGP-d AG1-u AG1-d AG2-u AG2-d AGS-u AGS-d GP-u GP-d G1-u G1-d G2-u G2-d GS-u GS-d
Fig 9 Variable selected frequency of over 190 runs for the TG ratio model of the lsquouprsquo
area of the soaking zone
0 2 4 6 8 10 12 14 16 18 20-05
05
AG
P-u
0 2 4 6 8 10 12 14 16 18 20-05
05
GP
-u
0 2 4 6 8 10 12 14 16 18 20-1
0
AG
P-d
0 2 4 6 8 10 12 14 16 18 20-1
0
GP
-d
0 2 4 6 8 10 12 14 16 18 20-05
05
AG
1-u
0 2 4 6 8 10 12 14 16 18 20-05
05
G1-
u
0 2 4 6 8 10 12 14 16 18 20-05
05
AG
1-d
0 2 4 6 8 10 12 14 16 18 20-05
05
G1-
d
0 2 4 6 8 10 12 14 16 18 20-1
0
AG
2-u
0 2 4 6 8 10 12 14 16 18 200
1
G2-
u
0 2 4 6 8 10 12 14 16 18 20-1
0
AG
2-d
0 2 4 6 8 10 12 14 16 18 20-1
0
G2-
d
0 2 4 6 8 10 12 14 16 18 200
1
AG
S-u
0 2 4 6 8 10 12 14 16 18 20-1
0
GS
-u
0 2 4 6 8 10 12 14 16 18 20-05
05
AG
S-d
0 2 4 6 8 10 12 14 16 18 20-05
05
GS
-d
Fig 10 Part of the NNG regression coefficients of each variable over 190 runs for the
TG ratio model of the lsquouprsquo area of the soaking zone
It is remarkable that the NNG regression coefficients corresponding to the AG ratio
in the lsquouprsquo area of the soaking zone are consistently positive On the contrary the
coefficients of the gas flow rate in the same area are consistently negative This
indicates that under this condition the rise of the AG leads to the increase of the TG
17
283
284
285
286
287
288
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
while the rise of the G leads to the drop of the TG This property is helpful for
performing optimization and improve combustion efficiency
For comparison the modeling results for two other algorithms artificial neural
network (ANN) and partial least squares (PLS) [26] are compared with the NNG
algorithm in the following three aspects [27]
(1) Model Size the number of variables selected for modeling
(2) Model Magnitude the mean of the L1 norm of the regression coefficients
(3) Prediction Precision the mean squared prediction error (MSPE)
Summary of the algorithm comparison is shown in Table 2 It can be seen that the
superiorities of the NNG regression in model size model magnitude and model
precision are remarkable
Table 2 Summary of algorithm comparison
Index NNG ANN PLS
Model Size 849 16 16
Model Magnitude 10075 -- 12150
MSPE 00093 00108 00140
5 Model-based optimization
51 Implemention of model-based optimization operation
The goal of optimization is to seek an optimal combination of AG and G in the lsquouprsquo
area of the soaking zone so as to minimize the gas consumption while keeping the
temperature at the target value According to the heating schedule the most expected
temperature in the soaking zone is 1200 Therefore the target temperature
is set at this value to achieve the desired heating effect As shown in Fig 11 in the real
18
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
operation the temperature fluctuates around the target value because of the imperfect
control performance However in order to simplify the analysis and compute the
maximum possible energy saving perfect temperature control is assumed when
adopting the model-based optimization strategy In other words it is assumed that the
actual temperature in the lsquouprsquo area of the soaking zone can be adjusted to the expected
temperature ie 1200
0 02 04 06 08 1 12 14 16 18 2
x 104
1150
1200
1250
1300
Sampling intervals
T S-u (
)
OriginalTarget
Fig 11 Original and target temperature in the lsquouprsquo area of the soaking zone
The adjustment scheme takes the maximum value and minimum value of the original
operation data as the upper and lower bounds for the adjustment Moreover in order to
assure the validity of the linear model the increment or decrement of AG and G should
not beyond of the original value (considered as 10 for the purposes of this paper)
The adjustment strategy can be formulated as
19
307
308
309
310
311
312
313
314
315
316
317
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
(3)
where NNG() denotes the NNG regression model of the TG ratio and are the
NNG regression coefficients of the AG and G in the lsquouprsquo area of the soaking zone for
modeling TG in same area
With the model-based strategy and adaptive model of the TG ratio given in Section
42 the optimization adjustment results can be obtained as follows
The adjustment of gas flow in the lsquouprsquo area of the soaking zone shown in Fig 12
illustrates that the implementation of the model-based optimization operation can reduce
the consumption of the gas flow Compared with the original case 938 of the gas can
be saved on average by utilizing the model-based adjustment
0 02 04 06 08 1 12 14 16 18 2
x 104
-800
-700
-600
-500
-400
-300
-200
-100
0
Sampling intervals
Adj
ustm
ent o
f gas
flow
m(
3 h)
20
318
319
320
321
322
323
324
325
326
327
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
(a) Adjustment amount of gas flow
0 02 04 06 08 1 12 14 16 18 2
x 104
500
1000
1500
2000
2500
3000
3500
4000
4500
5000G
as fl
owm
(3 h
)
Sampling intervals
OriginalAdjusted
(b) Comparison of original and adjusted gas flow
095 096 097 098 099 1 101 102 103 104 105
x 104
500
1000
1500
2000
2500
3000
3500
4000
4500
5000
Gas
flow
m(
3 h)
Sampling intervals
OriginalAdjusted
(c) Comparison of original and adjusted gas flow in interval [9501-10500]
Fig 12 Adjustment of gas flow in the lsquouprsquo area of the soaking zone
The adjustment of the air flow in the lsquouprsquo area of the soaking zone given in Fig 13
shows that the air flow is reduced in most cases and only increased for a few cases In
21
328
329
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
general 681 of the air flow is reduced after the implementation of the model-based
optimization
0 02 04 06 08 1 12 14 16 18 2
x 104
-2000
-1500
-1000
-500
0
500
1000
Sampling intervals
Adj
ustm
ent o
f air
flow
(m3 h
)
(a) Adjustment amount of air flow
0 02 04 06 08 1 12 14 16 18 2
x 104
4000
5000
6000
7000
8000
9000
10000
Air
flow
m(
3 h)
Sampling intervals
OriginalAdjusted
(b) Comparison of original and adjusted air flow
22
330
331
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
095 096 097 098 099 1 101 102 103 104 105
x 104
4000
5000
6000
7000
8000
9000
10000
Air
flow
m(
3 h)
Sampling intervals
OriginalAdjusted
(c) Comparison of original and adjusted air flow in interval [9501-10500]
Fig 13 Adjustment of air flow in the lsquouprsquo area of the soaking zone
52 Discussions
In this section detailed analysis for the optimization operation results is provided
The statistics of optimization at the boundary conditions shown in Table 3 illustrates
that the adjusted value for G reaches its lower limit in most cases (8045 for
and 687 for ) which is the most energy-efficient point In a number of cases
(1079 for and 001 for ) the adjusted AG reaches its upper
limit These results indicate that the optimization operations maximize the combustion
efficiency by decreasing G and increasing the AG ratio up to the boundary conditions
However the lower limit is also unexpectedly attained in a small number of cases
(110 for and 078 for ) for which a more detailed analysis is
given below
23
332
333
334
335
336
337
338
339
340
341
342
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
Table 3 Statistics of optimization at boundary conditions
Total
Amount 0 1374 0 16090 17464
Percentag
e 0 687 0 8045
8732
Total
Amount 2 220 2158 156 2536
Percentag
e 001 110 1079 078
1268
Table 4 Statistics for different optimization operations
Adjustment
Gdarr
Guarr TotalAGuarr AGdarr
Tdarr Tuarr Tdarr Tuarr
Amount 9548 3744 6328 380 0 20000
Percentage 4774 1872 3164 190 0 100
The statistics for different optimization operations is performed and shown in Table
4 It can be seen that all the adjusted operations result in the reduction of gas
consumption In over half the cases (4774 with decreased temperature and 1872
with increased temperature) AG is adjusted to higher levels Meanwhile a number of
the adjustments (3164) lower the temperature by reducing the G and AG at the same
time This result indicates that the original temperature cannot be adjusted to the target
by only reducing G in the constraint conditions The remaining cases (19) are
24
343
344
345
346
347
348
349
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
relatively special in which an excess of air is supplied in the original operation and the
adjusted operation thereby increasing the temperature by a smaller AG with less gas
consumption
6 Conclusion
Based on the actual operation data this paper aims to explore the improvement of the
combustion efficiency and the room for energy conservation Correlation analysis and
LDA show that it is of great significance to introduce two derived ratio variables which
are the AG ratio and the TG ratio A type of combustion efficiency utilizing an on-line
soft sensor is put forward by employing a NNG variable selection algorithm which
provides a good solution to the combustion efficiency real-time measurement problem
of a reheating furnace The implementation of the model-based optimization is studied
based on the actual operational data Detailed analysis for the optimization results is
given for the different cases The results show that significant energy conservation can
be achieved when the furnace operation is optimized by using the developed soft sensor
model
Acknowledgement
The authors would like to thank the financial support provided by the National Nature
Science Foundation of China under Grant 61171145 Y Yao was supported by Ministry
of Science amp Technology ROC under Grant number MOST 104-2221-E-007-129
References
25
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[1] Z J Wang Q D Wu and T Y Chai Optimal-setting control for complicated
industrial processes and its application studyControl Engineering Practice vol
12 pp 65-74 2004
[2] A Steinboeck K Graichen and A Kugi Dynamic Optimization of a Slab
Reheating Furnace With Consistent Approximation of Control VariablesIEEE
Transactions on Control Systems Technology vol 19 pp 1444-1456 2011
[3] B T Zhang C Y Wang Q Qin and L Li Influence of Boiler Combustion
Adjustment on NOxEmission and Boiler EfficiencyAdvanced Materials
Research vol 732-733 pp 234-237 2013
[4] C K Yoo and IB Lee Soft Sensor and Adaptive Model-Based Dissolved
Oxygen Control for Biological Wastewater Treatment ProcessesEnvironmental
Engineering Science vol 21 pp 331-340 2004
[5] S A Bhat D N Saraf S Gupta and S K Gupta Use of Agitator Power as a
Soft Sensor for Bulk Free-Radical Polymerization of Methyl Methacrylate in
Batch ReactorsIndustrial amp Engineering Chemistry Research vol 45 pp 4243-
4255 2006
[6] K Desai Y Badhe S S Tambe and B D Kulkarni Soft-sensor development
for fed-batch bioreactors using support vector regressionBiochemical
Engineering Journal vol 27 pp 225-239 2006
[7] Y P Badhe Lonari J Tambe S S amp Kulkarni B D Improve polyethylene
process control and product qualityHydrocarbon Processing vol 86 pp 53-60
2007
26
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[8] N K Nath K Mandal A K Singh B Basu C Bhanu S Kumar et al Ladle
furnace on-line reckoner for prediction and control of steel temperature and
compositionIronmaking amp Steelmaking vol 33 pp 140-150 2006
[9] A J Yan T Y Chai F H Wu and P Wang Hybrid intelligent control of
combustion process for ore-roasting furnaceJournal of Control Theory and
Applications vol 6 pp 80-85 2008
[10] J Li W M Zhong H Cheng X D Kong and F Qian A data-driven soft
sensor modeling for furnace temperature of Opposed Multi-Burner gasifier pp
705-710 2011
[11] Y H Yang Y H Liu X Z Liu and S K Qin Billet temperature soft sensor
model of reheating furnace based on RVM method pp 4003-4006 2011
[12] J H Wang C Wang X F Zhu and X K Fang Application of soft sensor in
welding seam tracking prediction based on LSSVM and PSO with compression
factor pp 2441-2446 2013
[13] L Balbis J Balderud and M J Grimble Nonlinear predictive control of steel
slab reheating furnace pp 1679-1684 2008
[14] A Steinboeck D Wild T Kiefer and A Kugi A mathematical model of a slab
reheating furnace with radiative heat transfer and non-participating gaseous
media International Journal Of Heat And Mass Transfer vol 53 pp 5933-
5946 Dec 2010
[15] A Steinboeck D Wild and A Kugi Nonlinear model predictive control of a
continuous slab reheating furnace Control Engineering Practice vol 21 pp
495-508 2013
27
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[16] C Zhang T Ishii and S Sugiyama Numerical Modeling Of the Thermal
Performance Of Regenerative Slab Reheat Furnaces Numerical Heat Transfer
Part A Applications vol 32 pp 613-631 1997
[17] J G Kim and K Y Huh Prediction of Transient Slab Temperature Distribution
in the Re-heating Furnace of a Walking-beam Type for Rolling of Steel Slabs
ISIJ International vol 40 pp 1115-1123 2000
[18] J G Kim K Y and H I T K Three-Dimensional Analysis Of the Walking-
Beam-Type Slab Reheating Furnace In Hot Strip Mills Numerical Heat
Transfer Part A Applications vol 38 pp 589-609 2000
[19] CT Hsieh MJ Huang ST Lee and CH Wang Numerical Modeling of a
Walking-Beam-Type Slab Reheating Furnace Numerical Heat Transfer Part A
Applications vol 53 pp 966-981 2008
[20] MJ Huang CT Hsieh ST Lee and CH Wang A Coupled Numerical Study
of Slab Temperature and Gas Temperature in the Walking-Beam-Type Slab
Reheating Furnace Numerical Heat Transfer Part A Applications vol 54 pp
625-646 2008
[21] Z Song and A Kusiak Constraint-Based Control of Boiler Efficiency A Data-
Mining Approach IEEE Transactions on Industrial Informatics vol 3 pp 73-
83 2007
[22] J Q Li J J Gu and C L Niu The Operation Optimization based on
Correlation Analysis of Operation Parameters in Power Plant pp 138-141
2008
28
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[23] J G Wang S S Shieh S S Jang D S H Wong and C W Wu A two-tier
approach to the data-driven modeling on thermal efficiency of a BFGcoal co-
firing boiler Fuel vol 111 pp 528-534 Sep 2013
[24] L Breiman Better Subset Regression Using the Nonnegative Garrote
Technometrics vol 37 pp 373-384 1995
[25] R A Fisher The use of multiple measurements in taxonomic problems
AnnHum Genet vol 7 pp 179-188 1936
[26] J Liu Developing a soft sensor based on sparse partial least squares with
variable selection Journal of Process Control vol 24 pp 1046-1056 2014
[27] K Sun J Liu JL Kang SS Jang D SH Wong and DS Chen
Development of a variable selection method for soft sensor using artificial
neural network and nonnegative garrote Journal of Process Control vol 24 pp
1068-1075 2014
29
439
440
441
442
443
444
445
446
447
448
449
450
451
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
0 02 04 06 08 1 12 14 16 18 2
x 104
02
04
06
08
1
12
14
16
18
2
Sampling intervals
TG
S-u
OriginalPredicted
Fig 8 Prediction of TG ratio in the lsquouprsquo area of the soaking zone
A comparison between Figs 7 and 8 show that the prediction of the TG ratio is more
accurate than the prediction of temperature The average R2 in 190 NNG regressions is
0935 for the TG ratio while the average R2 is only 0814 for temperature This is
consistent with the statistical analysis which reveals that the correlation between AG
and TG is higher than that between A and T These results indicate that derived
variables are more meaningful for the purpose of prediction and the modeling of the
TG ratio is more appropriate for implementing optimization
The selected frequency for each variable in 190 NNG regressions and the coefficients
of each variable in 20 NNG regressions for the TG ratio modeling of the lsquouprsquo area of
the soaking zone are shown in Figs 9 and 10 respectively Fig 9 shows that the selected
frequency of the variables in the lsquouprsquo area of the soaking zone is much higher than
variables in other zones Similarly as is shown in Fig 10 the NNG regression
coefficients of the variables in the lsquouprsquo area of the soaking zone are much larger than the
variables in the other areas These results indicated that variables in the lsquouprsquo area of
16
269
270
271
272
273
274
275
276
277
278
279
280
281
282
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
soaking zone are mostly contributed to the modeling of the TG ratio in same area
which is quite reasonable Similar results can be obtained for the other areas
0
20
40
60
80
100
120
140
160
180
200
Sel
ecte
d fre
quen
cy
AGP-u AGP-d AG1-u AG1-d AG2-u AG2-d AGS-u AGS-d GP-u GP-d G1-u G1-d G2-u G2-d GS-u GS-d
Fig 9 Variable selected frequency of over 190 runs for the TG ratio model of the lsquouprsquo
area of the soaking zone
0 2 4 6 8 10 12 14 16 18 20-05
05
AG
P-u
0 2 4 6 8 10 12 14 16 18 20-05
05
GP
-u
0 2 4 6 8 10 12 14 16 18 20-1
0
AG
P-d
0 2 4 6 8 10 12 14 16 18 20-1
0
GP
-d
0 2 4 6 8 10 12 14 16 18 20-05
05
AG
1-u
0 2 4 6 8 10 12 14 16 18 20-05
05
G1-
u
0 2 4 6 8 10 12 14 16 18 20-05
05
AG
1-d
0 2 4 6 8 10 12 14 16 18 20-05
05
G1-
d
0 2 4 6 8 10 12 14 16 18 20-1
0
AG
2-u
0 2 4 6 8 10 12 14 16 18 200
1
G2-
u
0 2 4 6 8 10 12 14 16 18 20-1
0
AG
2-d
0 2 4 6 8 10 12 14 16 18 20-1
0
G2-
d
0 2 4 6 8 10 12 14 16 18 200
1
AG
S-u
0 2 4 6 8 10 12 14 16 18 20-1
0
GS
-u
0 2 4 6 8 10 12 14 16 18 20-05
05
AG
S-d
0 2 4 6 8 10 12 14 16 18 20-05
05
GS
-d
Fig 10 Part of the NNG regression coefficients of each variable over 190 runs for the
TG ratio model of the lsquouprsquo area of the soaking zone
It is remarkable that the NNG regression coefficients corresponding to the AG ratio
in the lsquouprsquo area of the soaking zone are consistently positive On the contrary the
coefficients of the gas flow rate in the same area are consistently negative This
indicates that under this condition the rise of the AG leads to the increase of the TG
17
283
284
285
286
287
288
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
while the rise of the G leads to the drop of the TG This property is helpful for
performing optimization and improve combustion efficiency
For comparison the modeling results for two other algorithms artificial neural
network (ANN) and partial least squares (PLS) [26] are compared with the NNG
algorithm in the following three aspects [27]
(1) Model Size the number of variables selected for modeling
(2) Model Magnitude the mean of the L1 norm of the regression coefficients
(3) Prediction Precision the mean squared prediction error (MSPE)
Summary of the algorithm comparison is shown in Table 2 It can be seen that the
superiorities of the NNG regression in model size model magnitude and model
precision are remarkable
Table 2 Summary of algorithm comparison
Index NNG ANN PLS
Model Size 849 16 16
Model Magnitude 10075 -- 12150
MSPE 00093 00108 00140
5 Model-based optimization
51 Implemention of model-based optimization operation
The goal of optimization is to seek an optimal combination of AG and G in the lsquouprsquo
area of the soaking zone so as to minimize the gas consumption while keeping the
temperature at the target value According to the heating schedule the most expected
temperature in the soaking zone is 1200 Therefore the target temperature
is set at this value to achieve the desired heating effect As shown in Fig 11 in the real
18
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
operation the temperature fluctuates around the target value because of the imperfect
control performance However in order to simplify the analysis and compute the
maximum possible energy saving perfect temperature control is assumed when
adopting the model-based optimization strategy In other words it is assumed that the
actual temperature in the lsquouprsquo area of the soaking zone can be adjusted to the expected
temperature ie 1200
0 02 04 06 08 1 12 14 16 18 2
x 104
1150
1200
1250
1300
Sampling intervals
T S-u (
)
OriginalTarget
Fig 11 Original and target temperature in the lsquouprsquo area of the soaking zone
The adjustment scheme takes the maximum value and minimum value of the original
operation data as the upper and lower bounds for the adjustment Moreover in order to
assure the validity of the linear model the increment or decrement of AG and G should
not beyond of the original value (considered as 10 for the purposes of this paper)
The adjustment strategy can be formulated as
19
307
308
309
310
311
312
313
314
315
316
317
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
(3)
where NNG() denotes the NNG regression model of the TG ratio and are the
NNG regression coefficients of the AG and G in the lsquouprsquo area of the soaking zone for
modeling TG in same area
With the model-based strategy and adaptive model of the TG ratio given in Section
42 the optimization adjustment results can be obtained as follows
The adjustment of gas flow in the lsquouprsquo area of the soaking zone shown in Fig 12
illustrates that the implementation of the model-based optimization operation can reduce
the consumption of the gas flow Compared with the original case 938 of the gas can
be saved on average by utilizing the model-based adjustment
0 02 04 06 08 1 12 14 16 18 2
x 104
-800
-700
-600
-500
-400
-300
-200
-100
0
Sampling intervals
Adj
ustm
ent o
f gas
flow
m(
3 h)
20
318
319
320
321
322
323
324
325
326
327
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
(a) Adjustment amount of gas flow
0 02 04 06 08 1 12 14 16 18 2
x 104
500
1000
1500
2000
2500
3000
3500
4000
4500
5000G
as fl
owm
(3 h
)
Sampling intervals
OriginalAdjusted
(b) Comparison of original and adjusted gas flow
095 096 097 098 099 1 101 102 103 104 105
x 104
500
1000
1500
2000
2500
3000
3500
4000
4500
5000
Gas
flow
m(
3 h)
Sampling intervals
OriginalAdjusted
(c) Comparison of original and adjusted gas flow in interval [9501-10500]
Fig 12 Adjustment of gas flow in the lsquouprsquo area of the soaking zone
The adjustment of the air flow in the lsquouprsquo area of the soaking zone given in Fig 13
shows that the air flow is reduced in most cases and only increased for a few cases In
21
328
329
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
general 681 of the air flow is reduced after the implementation of the model-based
optimization
0 02 04 06 08 1 12 14 16 18 2
x 104
-2000
-1500
-1000
-500
0
500
1000
Sampling intervals
Adj
ustm
ent o
f air
flow
(m3 h
)
(a) Adjustment amount of air flow
0 02 04 06 08 1 12 14 16 18 2
x 104
4000
5000
6000
7000
8000
9000
10000
Air
flow
m(
3 h)
Sampling intervals
OriginalAdjusted
(b) Comparison of original and adjusted air flow
22
330
331
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
095 096 097 098 099 1 101 102 103 104 105
x 104
4000
5000
6000
7000
8000
9000
10000
Air
flow
m(
3 h)
Sampling intervals
OriginalAdjusted
(c) Comparison of original and adjusted air flow in interval [9501-10500]
Fig 13 Adjustment of air flow in the lsquouprsquo area of the soaking zone
52 Discussions
In this section detailed analysis for the optimization operation results is provided
The statistics of optimization at the boundary conditions shown in Table 3 illustrates
that the adjusted value for G reaches its lower limit in most cases (8045 for
and 687 for ) which is the most energy-efficient point In a number of cases
(1079 for and 001 for ) the adjusted AG reaches its upper
limit These results indicate that the optimization operations maximize the combustion
efficiency by decreasing G and increasing the AG ratio up to the boundary conditions
However the lower limit is also unexpectedly attained in a small number of cases
(110 for and 078 for ) for which a more detailed analysis is
given below
23
332
333
334
335
336
337
338
339
340
341
342
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
Table 3 Statistics of optimization at boundary conditions
Total
Amount 0 1374 0 16090 17464
Percentag
e 0 687 0 8045
8732
Total
Amount 2 220 2158 156 2536
Percentag
e 001 110 1079 078
1268
Table 4 Statistics for different optimization operations
Adjustment
Gdarr
Guarr TotalAGuarr AGdarr
Tdarr Tuarr Tdarr Tuarr
Amount 9548 3744 6328 380 0 20000
Percentage 4774 1872 3164 190 0 100
The statistics for different optimization operations is performed and shown in Table
4 It can be seen that all the adjusted operations result in the reduction of gas
consumption In over half the cases (4774 with decreased temperature and 1872
with increased temperature) AG is adjusted to higher levels Meanwhile a number of
the adjustments (3164) lower the temperature by reducing the G and AG at the same
time This result indicates that the original temperature cannot be adjusted to the target
by only reducing G in the constraint conditions The remaining cases (19) are
24
343
344
345
346
347
348
349
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
relatively special in which an excess of air is supplied in the original operation and the
adjusted operation thereby increasing the temperature by a smaller AG with less gas
consumption
6 Conclusion
Based on the actual operation data this paper aims to explore the improvement of the
combustion efficiency and the room for energy conservation Correlation analysis and
LDA show that it is of great significance to introduce two derived ratio variables which
are the AG ratio and the TG ratio A type of combustion efficiency utilizing an on-line
soft sensor is put forward by employing a NNG variable selection algorithm which
provides a good solution to the combustion efficiency real-time measurement problem
of a reheating furnace The implementation of the model-based optimization is studied
based on the actual operational data Detailed analysis for the optimization results is
given for the different cases The results show that significant energy conservation can
be achieved when the furnace operation is optimized by using the developed soft sensor
model
Acknowledgement
The authors would like to thank the financial support provided by the National Nature
Science Foundation of China under Grant 61171145 Y Yao was supported by Ministry
of Science amp Technology ROC under Grant number MOST 104-2221-E-007-129
References
25
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[1] Z J Wang Q D Wu and T Y Chai Optimal-setting control for complicated
industrial processes and its application studyControl Engineering Practice vol
12 pp 65-74 2004
[2] A Steinboeck K Graichen and A Kugi Dynamic Optimization of a Slab
Reheating Furnace With Consistent Approximation of Control VariablesIEEE
Transactions on Control Systems Technology vol 19 pp 1444-1456 2011
[3] B T Zhang C Y Wang Q Qin and L Li Influence of Boiler Combustion
Adjustment on NOxEmission and Boiler EfficiencyAdvanced Materials
Research vol 732-733 pp 234-237 2013
[4] C K Yoo and IB Lee Soft Sensor and Adaptive Model-Based Dissolved
Oxygen Control for Biological Wastewater Treatment ProcessesEnvironmental
Engineering Science vol 21 pp 331-340 2004
[5] S A Bhat D N Saraf S Gupta and S K Gupta Use of Agitator Power as a
Soft Sensor for Bulk Free-Radical Polymerization of Methyl Methacrylate in
Batch ReactorsIndustrial amp Engineering Chemistry Research vol 45 pp 4243-
4255 2006
[6] K Desai Y Badhe S S Tambe and B D Kulkarni Soft-sensor development
for fed-batch bioreactors using support vector regressionBiochemical
Engineering Journal vol 27 pp 225-239 2006
[7] Y P Badhe Lonari J Tambe S S amp Kulkarni B D Improve polyethylene
process control and product qualityHydrocarbon Processing vol 86 pp 53-60
2007
26
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[8] N K Nath K Mandal A K Singh B Basu C Bhanu S Kumar et al Ladle
furnace on-line reckoner for prediction and control of steel temperature and
compositionIronmaking amp Steelmaking vol 33 pp 140-150 2006
[9] A J Yan T Y Chai F H Wu and P Wang Hybrid intelligent control of
combustion process for ore-roasting furnaceJournal of Control Theory and
Applications vol 6 pp 80-85 2008
[10] J Li W M Zhong H Cheng X D Kong and F Qian A data-driven soft
sensor modeling for furnace temperature of Opposed Multi-Burner gasifier pp
705-710 2011
[11] Y H Yang Y H Liu X Z Liu and S K Qin Billet temperature soft sensor
model of reheating furnace based on RVM method pp 4003-4006 2011
[12] J H Wang C Wang X F Zhu and X K Fang Application of soft sensor in
welding seam tracking prediction based on LSSVM and PSO with compression
factor pp 2441-2446 2013
[13] L Balbis J Balderud and M J Grimble Nonlinear predictive control of steel
slab reheating furnace pp 1679-1684 2008
[14] A Steinboeck D Wild T Kiefer and A Kugi A mathematical model of a slab
reheating furnace with radiative heat transfer and non-participating gaseous
media International Journal Of Heat And Mass Transfer vol 53 pp 5933-
5946 Dec 2010
[15] A Steinboeck D Wild and A Kugi Nonlinear model predictive control of a
continuous slab reheating furnace Control Engineering Practice vol 21 pp
495-508 2013
27
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[16] C Zhang T Ishii and S Sugiyama Numerical Modeling Of the Thermal
Performance Of Regenerative Slab Reheat Furnaces Numerical Heat Transfer
Part A Applications vol 32 pp 613-631 1997
[17] J G Kim and K Y Huh Prediction of Transient Slab Temperature Distribution
in the Re-heating Furnace of a Walking-beam Type for Rolling of Steel Slabs
ISIJ International vol 40 pp 1115-1123 2000
[18] J G Kim K Y and H I T K Three-Dimensional Analysis Of the Walking-
Beam-Type Slab Reheating Furnace In Hot Strip Mills Numerical Heat
Transfer Part A Applications vol 38 pp 589-609 2000
[19] CT Hsieh MJ Huang ST Lee and CH Wang Numerical Modeling of a
Walking-Beam-Type Slab Reheating Furnace Numerical Heat Transfer Part A
Applications vol 53 pp 966-981 2008
[20] MJ Huang CT Hsieh ST Lee and CH Wang A Coupled Numerical Study
of Slab Temperature and Gas Temperature in the Walking-Beam-Type Slab
Reheating Furnace Numerical Heat Transfer Part A Applications vol 54 pp
625-646 2008
[21] Z Song and A Kusiak Constraint-Based Control of Boiler Efficiency A Data-
Mining Approach IEEE Transactions on Industrial Informatics vol 3 pp 73-
83 2007
[22] J Q Li J J Gu and C L Niu The Operation Optimization based on
Correlation Analysis of Operation Parameters in Power Plant pp 138-141
2008
28
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[23] J G Wang S S Shieh S S Jang D S H Wong and C W Wu A two-tier
approach to the data-driven modeling on thermal efficiency of a BFGcoal co-
firing boiler Fuel vol 111 pp 528-534 Sep 2013
[24] L Breiman Better Subset Regression Using the Nonnegative Garrote
Technometrics vol 37 pp 373-384 1995
[25] R A Fisher The use of multiple measurements in taxonomic problems
AnnHum Genet vol 7 pp 179-188 1936
[26] J Liu Developing a soft sensor based on sparse partial least squares with
variable selection Journal of Process Control vol 24 pp 1046-1056 2014
[27] K Sun J Liu JL Kang SS Jang D SH Wong and DS Chen
Development of a variable selection method for soft sensor using artificial
neural network and nonnegative garrote Journal of Process Control vol 24 pp
1068-1075 2014
29
439
440
441
442
443
444
445
446
447
448
449
450
451
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
soaking zone are mostly contributed to the modeling of the TG ratio in same area
which is quite reasonable Similar results can be obtained for the other areas
0
20
40
60
80
100
120
140
160
180
200
Sel
ecte
d fre
quen
cy
AGP-u AGP-d AG1-u AG1-d AG2-u AG2-d AGS-u AGS-d GP-u GP-d G1-u G1-d G2-u G2-d GS-u GS-d
Fig 9 Variable selected frequency of over 190 runs for the TG ratio model of the lsquouprsquo
area of the soaking zone
0 2 4 6 8 10 12 14 16 18 20-05
05
AG
P-u
0 2 4 6 8 10 12 14 16 18 20-05
05
GP
-u
0 2 4 6 8 10 12 14 16 18 20-1
0
AG
P-d
0 2 4 6 8 10 12 14 16 18 20-1
0
GP
-d
0 2 4 6 8 10 12 14 16 18 20-05
05
AG
1-u
0 2 4 6 8 10 12 14 16 18 20-05
05
G1-
u
0 2 4 6 8 10 12 14 16 18 20-05
05
AG
1-d
0 2 4 6 8 10 12 14 16 18 20-05
05
G1-
d
0 2 4 6 8 10 12 14 16 18 20-1
0
AG
2-u
0 2 4 6 8 10 12 14 16 18 200
1
G2-
u
0 2 4 6 8 10 12 14 16 18 20-1
0
AG
2-d
0 2 4 6 8 10 12 14 16 18 20-1
0
G2-
d
0 2 4 6 8 10 12 14 16 18 200
1
AG
S-u
0 2 4 6 8 10 12 14 16 18 20-1
0
GS
-u
0 2 4 6 8 10 12 14 16 18 20-05
05
AG
S-d
0 2 4 6 8 10 12 14 16 18 20-05
05
GS
-d
Fig 10 Part of the NNG regression coefficients of each variable over 190 runs for the
TG ratio model of the lsquouprsquo area of the soaking zone
It is remarkable that the NNG regression coefficients corresponding to the AG ratio
in the lsquouprsquo area of the soaking zone are consistently positive On the contrary the
coefficients of the gas flow rate in the same area are consistently negative This
indicates that under this condition the rise of the AG leads to the increase of the TG
17
283
284
285
286
287
288
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
while the rise of the G leads to the drop of the TG This property is helpful for
performing optimization and improve combustion efficiency
For comparison the modeling results for two other algorithms artificial neural
network (ANN) and partial least squares (PLS) [26] are compared with the NNG
algorithm in the following three aspects [27]
(1) Model Size the number of variables selected for modeling
(2) Model Magnitude the mean of the L1 norm of the regression coefficients
(3) Prediction Precision the mean squared prediction error (MSPE)
Summary of the algorithm comparison is shown in Table 2 It can be seen that the
superiorities of the NNG regression in model size model magnitude and model
precision are remarkable
Table 2 Summary of algorithm comparison
Index NNG ANN PLS
Model Size 849 16 16
Model Magnitude 10075 -- 12150
MSPE 00093 00108 00140
5 Model-based optimization
51 Implemention of model-based optimization operation
The goal of optimization is to seek an optimal combination of AG and G in the lsquouprsquo
area of the soaking zone so as to minimize the gas consumption while keeping the
temperature at the target value According to the heating schedule the most expected
temperature in the soaking zone is 1200 Therefore the target temperature
is set at this value to achieve the desired heating effect As shown in Fig 11 in the real
18
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
operation the temperature fluctuates around the target value because of the imperfect
control performance However in order to simplify the analysis and compute the
maximum possible energy saving perfect temperature control is assumed when
adopting the model-based optimization strategy In other words it is assumed that the
actual temperature in the lsquouprsquo area of the soaking zone can be adjusted to the expected
temperature ie 1200
0 02 04 06 08 1 12 14 16 18 2
x 104
1150
1200
1250
1300
Sampling intervals
T S-u (
)
OriginalTarget
Fig 11 Original and target temperature in the lsquouprsquo area of the soaking zone
The adjustment scheme takes the maximum value and minimum value of the original
operation data as the upper and lower bounds for the adjustment Moreover in order to
assure the validity of the linear model the increment or decrement of AG and G should
not beyond of the original value (considered as 10 for the purposes of this paper)
The adjustment strategy can be formulated as
19
307
308
309
310
311
312
313
314
315
316
317
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
(3)
where NNG() denotes the NNG regression model of the TG ratio and are the
NNG regression coefficients of the AG and G in the lsquouprsquo area of the soaking zone for
modeling TG in same area
With the model-based strategy and adaptive model of the TG ratio given in Section
42 the optimization adjustment results can be obtained as follows
The adjustment of gas flow in the lsquouprsquo area of the soaking zone shown in Fig 12
illustrates that the implementation of the model-based optimization operation can reduce
the consumption of the gas flow Compared with the original case 938 of the gas can
be saved on average by utilizing the model-based adjustment
0 02 04 06 08 1 12 14 16 18 2
x 104
-800
-700
-600
-500
-400
-300
-200
-100
0
Sampling intervals
Adj
ustm
ent o
f gas
flow
m(
3 h)
20
318
319
320
321
322
323
324
325
326
327
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
(a) Adjustment amount of gas flow
0 02 04 06 08 1 12 14 16 18 2
x 104
500
1000
1500
2000
2500
3000
3500
4000
4500
5000G
as fl
owm
(3 h
)
Sampling intervals
OriginalAdjusted
(b) Comparison of original and adjusted gas flow
095 096 097 098 099 1 101 102 103 104 105
x 104
500
1000
1500
2000
2500
3000
3500
4000
4500
5000
Gas
flow
m(
3 h)
Sampling intervals
OriginalAdjusted
(c) Comparison of original and adjusted gas flow in interval [9501-10500]
Fig 12 Adjustment of gas flow in the lsquouprsquo area of the soaking zone
The adjustment of the air flow in the lsquouprsquo area of the soaking zone given in Fig 13
shows that the air flow is reduced in most cases and only increased for a few cases In
21
328
329
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
general 681 of the air flow is reduced after the implementation of the model-based
optimization
0 02 04 06 08 1 12 14 16 18 2
x 104
-2000
-1500
-1000
-500
0
500
1000
Sampling intervals
Adj
ustm
ent o
f air
flow
(m3 h
)
(a) Adjustment amount of air flow
0 02 04 06 08 1 12 14 16 18 2
x 104
4000
5000
6000
7000
8000
9000
10000
Air
flow
m(
3 h)
Sampling intervals
OriginalAdjusted
(b) Comparison of original and adjusted air flow
22
330
331
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
095 096 097 098 099 1 101 102 103 104 105
x 104
4000
5000
6000
7000
8000
9000
10000
Air
flow
m(
3 h)
Sampling intervals
OriginalAdjusted
(c) Comparison of original and adjusted air flow in interval [9501-10500]
Fig 13 Adjustment of air flow in the lsquouprsquo area of the soaking zone
52 Discussions
In this section detailed analysis for the optimization operation results is provided
The statistics of optimization at the boundary conditions shown in Table 3 illustrates
that the adjusted value for G reaches its lower limit in most cases (8045 for
and 687 for ) which is the most energy-efficient point In a number of cases
(1079 for and 001 for ) the adjusted AG reaches its upper
limit These results indicate that the optimization operations maximize the combustion
efficiency by decreasing G and increasing the AG ratio up to the boundary conditions
However the lower limit is also unexpectedly attained in a small number of cases
(110 for and 078 for ) for which a more detailed analysis is
given below
23
332
333
334
335
336
337
338
339
340
341
342
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
Table 3 Statistics of optimization at boundary conditions
Total
Amount 0 1374 0 16090 17464
Percentag
e 0 687 0 8045
8732
Total
Amount 2 220 2158 156 2536
Percentag
e 001 110 1079 078
1268
Table 4 Statistics for different optimization operations
Adjustment
Gdarr
Guarr TotalAGuarr AGdarr
Tdarr Tuarr Tdarr Tuarr
Amount 9548 3744 6328 380 0 20000
Percentage 4774 1872 3164 190 0 100
The statistics for different optimization operations is performed and shown in Table
4 It can be seen that all the adjusted operations result in the reduction of gas
consumption In over half the cases (4774 with decreased temperature and 1872
with increased temperature) AG is adjusted to higher levels Meanwhile a number of
the adjustments (3164) lower the temperature by reducing the G and AG at the same
time This result indicates that the original temperature cannot be adjusted to the target
by only reducing G in the constraint conditions The remaining cases (19) are
24
343
344
345
346
347
348
349
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
relatively special in which an excess of air is supplied in the original operation and the
adjusted operation thereby increasing the temperature by a smaller AG with less gas
consumption
6 Conclusion
Based on the actual operation data this paper aims to explore the improvement of the
combustion efficiency and the room for energy conservation Correlation analysis and
LDA show that it is of great significance to introduce two derived ratio variables which
are the AG ratio and the TG ratio A type of combustion efficiency utilizing an on-line
soft sensor is put forward by employing a NNG variable selection algorithm which
provides a good solution to the combustion efficiency real-time measurement problem
of a reheating furnace The implementation of the model-based optimization is studied
based on the actual operational data Detailed analysis for the optimization results is
given for the different cases The results show that significant energy conservation can
be achieved when the furnace operation is optimized by using the developed soft sensor
model
Acknowledgement
The authors would like to thank the financial support provided by the National Nature
Science Foundation of China under Grant 61171145 Y Yao was supported by Ministry
of Science amp Technology ROC under Grant number MOST 104-2221-E-007-129
References
25
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[1] Z J Wang Q D Wu and T Y Chai Optimal-setting control for complicated
industrial processes and its application studyControl Engineering Practice vol
12 pp 65-74 2004
[2] A Steinboeck K Graichen and A Kugi Dynamic Optimization of a Slab
Reheating Furnace With Consistent Approximation of Control VariablesIEEE
Transactions on Control Systems Technology vol 19 pp 1444-1456 2011
[3] B T Zhang C Y Wang Q Qin and L Li Influence of Boiler Combustion
Adjustment on NOxEmission and Boiler EfficiencyAdvanced Materials
Research vol 732-733 pp 234-237 2013
[4] C K Yoo and IB Lee Soft Sensor and Adaptive Model-Based Dissolved
Oxygen Control for Biological Wastewater Treatment ProcessesEnvironmental
Engineering Science vol 21 pp 331-340 2004
[5] S A Bhat D N Saraf S Gupta and S K Gupta Use of Agitator Power as a
Soft Sensor for Bulk Free-Radical Polymerization of Methyl Methacrylate in
Batch ReactorsIndustrial amp Engineering Chemistry Research vol 45 pp 4243-
4255 2006
[6] K Desai Y Badhe S S Tambe and B D Kulkarni Soft-sensor development
for fed-batch bioreactors using support vector regressionBiochemical
Engineering Journal vol 27 pp 225-239 2006
[7] Y P Badhe Lonari J Tambe S S amp Kulkarni B D Improve polyethylene
process control and product qualityHydrocarbon Processing vol 86 pp 53-60
2007
26
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[8] N K Nath K Mandal A K Singh B Basu C Bhanu S Kumar et al Ladle
furnace on-line reckoner for prediction and control of steel temperature and
compositionIronmaking amp Steelmaking vol 33 pp 140-150 2006
[9] A J Yan T Y Chai F H Wu and P Wang Hybrid intelligent control of
combustion process for ore-roasting furnaceJournal of Control Theory and
Applications vol 6 pp 80-85 2008
[10] J Li W M Zhong H Cheng X D Kong and F Qian A data-driven soft
sensor modeling for furnace temperature of Opposed Multi-Burner gasifier pp
705-710 2011
[11] Y H Yang Y H Liu X Z Liu and S K Qin Billet temperature soft sensor
model of reheating furnace based on RVM method pp 4003-4006 2011
[12] J H Wang C Wang X F Zhu and X K Fang Application of soft sensor in
welding seam tracking prediction based on LSSVM and PSO with compression
factor pp 2441-2446 2013
[13] L Balbis J Balderud and M J Grimble Nonlinear predictive control of steel
slab reheating furnace pp 1679-1684 2008
[14] A Steinboeck D Wild T Kiefer and A Kugi A mathematical model of a slab
reheating furnace with radiative heat transfer and non-participating gaseous
media International Journal Of Heat And Mass Transfer vol 53 pp 5933-
5946 Dec 2010
[15] A Steinboeck D Wild and A Kugi Nonlinear model predictive control of a
continuous slab reheating furnace Control Engineering Practice vol 21 pp
495-508 2013
27
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[16] C Zhang T Ishii and S Sugiyama Numerical Modeling Of the Thermal
Performance Of Regenerative Slab Reheat Furnaces Numerical Heat Transfer
Part A Applications vol 32 pp 613-631 1997
[17] J G Kim and K Y Huh Prediction of Transient Slab Temperature Distribution
in the Re-heating Furnace of a Walking-beam Type for Rolling of Steel Slabs
ISIJ International vol 40 pp 1115-1123 2000
[18] J G Kim K Y and H I T K Three-Dimensional Analysis Of the Walking-
Beam-Type Slab Reheating Furnace In Hot Strip Mills Numerical Heat
Transfer Part A Applications vol 38 pp 589-609 2000
[19] CT Hsieh MJ Huang ST Lee and CH Wang Numerical Modeling of a
Walking-Beam-Type Slab Reheating Furnace Numerical Heat Transfer Part A
Applications vol 53 pp 966-981 2008
[20] MJ Huang CT Hsieh ST Lee and CH Wang A Coupled Numerical Study
of Slab Temperature and Gas Temperature in the Walking-Beam-Type Slab
Reheating Furnace Numerical Heat Transfer Part A Applications vol 54 pp
625-646 2008
[21] Z Song and A Kusiak Constraint-Based Control of Boiler Efficiency A Data-
Mining Approach IEEE Transactions on Industrial Informatics vol 3 pp 73-
83 2007
[22] J Q Li J J Gu and C L Niu The Operation Optimization based on
Correlation Analysis of Operation Parameters in Power Plant pp 138-141
2008
28
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[23] J G Wang S S Shieh S S Jang D S H Wong and C W Wu A two-tier
approach to the data-driven modeling on thermal efficiency of a BFGcoal co-
firing boiler Fuel vol 111 pp 528-534 Sep 2013
[24] L Breiman Better Subset Regression Using the Nonnegative Garrote
Technometrics vol 37 pp 373-384 1995
[25] R A Fisher The use of multiple measurements in taxonomic problems
AnnHum Genet vol 7 pp 179-188 1936
[26] J Liu Developing a soft sensor based on sparse partial least squares with
variable selection Journal of Process Control vol 24 pp 1046-1056 2014
[27] K Sun J Liu JL Kang SS Jang D SH Wong and DS Chen
Development of a variable selection method for soft sensor using artificial
neural network and nonnegative garrote Journal of Process Control vol 24 pp
1068-1075 2014
29
439
440
441
442
443
444
445
446
447
448
449
450
451
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
while the rise of the G leads to the drop of the TG This property is helpful for
performing optimization and improve combustion efficiency
For comparison the modeling results for two other algorithms artificial neural
network (ANN) and partial least squares (PLS) [26] are compared with the NNG
algorithm in the following three aspects [27]
(1) Model Size the number of variables selected for modeling
(2) Model Magnitude the mean of the L1 norm of the regression coefficients
(3) Prediction Precision the mean squared prediction error (MSPE)
Summary of the algorithm comparison is shown in Table 2 It can be seen that the
superiorities of the NNG regression in model size model magnitude and model
precision are remarkable
Table 2 Summary of algorithm comparison
Index NNG ANN PLS
Model Size 849 16 16
Model Magnitude 10075 -- 12150
MSPE 00093 00108 00140
5 Model-based optimization
51 Implemention of model-based optimization operation
The goal of optimization is to seek an optimal combination of AG and G in the lsquouprsquo
area of the soaking zone so as to minimize the gas consumption while keeping the
temperature at the target value According to the heating schedule the most expected
temperature in the soaking zone is 1200 Therefore the target temperature
is set at this value to achieve the desired heating effect As shown in Fig 11 in the real
18
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
operation the temperature fluctuates around the target value because of the imperfect
control performance However in order to simplify the analysis and compute the
maximum possible energy saving perfect temperature control is assumed when
adopting the model-based optimization strategy In other words it is assumed that the
actual temperature in the lsquouprsquo area of the soaking zone can be adjusted to the expected
temperature ie 1200
0 02 04 06 08 1 12 14 16 18 2
x 104
1150
1200
1250
1300
Sampling intervals
T S-u (
)
OriginalTarget
Fig 11 Original and target temperature in the lsquouprsquo area of the soaking zone
The adjustment scheme takes the maximum value and minimum value of the original
operation data as the upper and lower bounds for the adjustment Moreover in order to
assure the validity of the linear model the increment or decrement of AG and G should
not beyond of the original value (considered as 10 for the purposes of this paper)
The adjustment strategy can be formulated as
19
307
308
309
310
311
312
313
314
315
316
317
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
(3)
where NNG() denotes the NNG regression model of the TG ratio and are the
NNG regression coefficients of the AG and G in the lsquouprsquo area of the soaking zone for
modeling TG in same area
With the model-based strategy and adaptive model of the TG ratio given in Section
42 the optimization adjustment results can be obtained as follows
The adjustment of gas flow in the lsquouprsquo area of the soaking zone shown in Fig 12
illustrates that the implementation of the model-based optimization operation can reduce
the consumption of the gas flow Compared with the original case 938 of the gas can
be saved on average by utilizing the model-based adjustment
0 02 04 06 08 1 12 14 16 18 2
x 104
-800
-700
-600
-500
-400
-300
-200
-100
0
Sampling intervals
Adj
ustm
ent o
f gas
flow
m(
3 h)
20
318
319
320
321
322
323
324
325
326
327
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
(a) Adjustment amount of gas flow
0 02 04 06 08 1 12 14 16 18 2
x 104
500
1000
1500
2000
2500
3000
3500
4000
4500
5000G
as fl
owm
(3 h
)
Sampling intervals
OriginalAdjusted
(b) Comparison of original and adjusted gas flow
095 096 097 098 099 1 101 102 103 104 105
x 104
500
1000
1500
2000
2500
3000
3500
4000
4500
5000
Gas
flow
m(
3 h)
Sampling intervals
OriginalAdjusted
(c) Comparison of original and adjusted gas flow in interval [9501-10500]
Fig 12 Adjustment of gas flow in the lsquouprsquo area of the soaking zone
The adjustment of the air flow in the lsquouprsquo area of the soaking zone given in Fig 13
shows that the air flow is reduced in most cases and only increased for a few cases In
21
328
329
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
general 681 of the air flow is reduced after the implementation of the model-based
optimization
0 02 04 06 08 1 12 14 16 18 2
x 104
-2000
-1500
-1000
-500
0
500
1000
Sampling intervals
Adj
ustm
ent o
f air
flow
(m3 h
)
(a) Adjustment amount of air flow
0 02 04 06 08 1 12 14 16 18 2
x 104
4000
5000
6000
7000
8000
9000
10000
Air
flow
m(
3 h)
Sampling intervals
OriginalAdjusted
(b) Comparison of original and adjusted air flow
22
330
331
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
095 096 097 098 099 1 101 102 103 104 105
x 104
4000
5000
6000
7000
8000
9000
10000
Air
flow
m(
3 h)
Sampling intervals
OriginalAdjusted
(c) Comparison of original and adjusted air flow in interval [9501-10500]
Fig 13 Adjustment of air flow in the lsquouprsquo area of the soaking zone
52 Discussions
In this section detailed analysis for the optimization operation results is provided
The statistics of optimization at the boundary conditions shown in Table 3 illustrates
that the adjusted value for G reaches its lower limit in most cases (8045 for
and 687 for ) which is the most energy-efficient point In a number of cases
(1079 for and 001 for ) the adjusted AG reaches its upper
limit These results indicate that the optimization operations maximize the combustion
efficiency by decreasing G and increasing the AG ratio up to the boundary conditions
However the lower limit is also unexpectedly attained in a small number of cases
(110 for and 078 for ) for which a more detailed analysis is
given below
23
332
333
334
335
336
337
338
339
340
341
342
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
Table 3 Statistics of optimization at boundary conditions
Total
Amount 0 1374 0 16090 17464
Percentag
e 0 687 0 8045
8732
Total
Amount 2 220 2158 156 2536
Percentag
e 001 110 1079 078
1268
Table 4 Statistics for different optimization operations
Adjustment
Gdarr
Guarr TotalAGuarr AGdarr
Tdarr Tuarr Tdarr Tuarr
Amount 9548 3744 6328 380 0 20000
Percentage 4774 1872 3164 190 0 100
The statistics for different optimization operations is performed and shown in Table
4 It can be seen that all the adjusted operations result in the reduction of gas
consumption In over half the cases (4774 with decreased temperature and 1872
with increased temperature) AG is adjusted to higher levels Meanwhile a number of
the adjustments (3164) lower the temperature by reducing the G and AG at the same
time This result indicates that the original temperature cannot be adjusted to the target
by only reducing G in the constraint conditions The remaining cases (19) are
24
343
344
345
346
347
348
349
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
relatively special in which an excess of air is supplied in the original operation and the
adjusted operation thereby increasing the temperature by a smaller AG with less gas
consumption
6 Conclusion
Based on the actual operation data this paper aims to explore the improvement of the
combustion efficiency and the room for energy conservation Correlation analysis and
LDA show that it is of great significance to introduce two derived ratio variables which
are the AG ratio and the TG ratio A type of combustion efficiency utilizing an on-line
soft sensor is put forward by employing a NNG variable selection algorithm which
provides a good solution to the combustion efficiency real-time measurement problem
of a reheating furnace The implementation of the model-based optimization is studied
based on the actual operational data Detailed analysis for the optimization results is
given for the different cases The results show that significant energy conservation can
be achieved when the furnace operation is optimized by using the developed soft sensor
model
Acknowledgement
The authors would like to thank the financial support provided by the National Nature
Science Foundation of China under Grant 61171145 Y Yao was supported by Ministry
of Science amp Technology ROC under Grant number MOST 104-2221-E-007-129
References
25
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[1] Z J Wang Q D Wu and T Y Chai Optimal-setting control for complicated
industrial processes and its application studyControl Engineering Practice vol
12 pp 65-74 2004
[2] A Steinboeck K Graichen and A Kugi Dynamic Optimization of a Slab
Reheating Furnace With Consistent Approximation of Control VariablesIEEE
Transactions on Control Systems Technology vol 19 pp 1444-1456 2011
[3] B T Zhang C Y Wang Q Qin and L Li Influence of Boiler Combustion
Adjustment on NOxEmission and Boiler EfficiencyAdvanced Materials
Research vol 732-733 pp 234-237 2013
[4] C K Yoo and IB Lee Soft Sensor and Adaptive Model-Based Dissolved
Oxygen Control for Biological Wastewater Treatment ProcessesEnvironmental
Engineering Science vol 21 pp 331-340 2004
[5] S A Bhat D N Saraf S Gupta and S K Gupta Use of Agitator Power as a
Soft Sensor for Bulk Free-Radical Polymerization of Methyl Methacrylate in
Batch ReactorsIndustrial amp Engineering Chemistry Research vol 45 pp 4243-
4255 2006
[6] K Desai Y Badhe S S Tambe and B D Kulkarni Soft-sensor development
for fed-batch bioreactors using support vector regressionBiochemical
Engineering Journal vol 27 pp 225-239 2006
[7] Y P Badhe Lonari J Tambe S S amp Kulkarni B D Improve polyethylene
process control and product qualityHydrocarbon Processing vol 86 pp 53-60
2007
26
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[8] N K Nath K Mandal A K Singh B Basu C Bhanu S Kumar et al Ladle
furnace on-line reckoner for prediction and control of steel temperature and
compositionIronmaking amp Steelmaking vol 33 pp 140-150 2006
[9] A J Yan T Y Chai F H Wu and P Wang Hybrid intelligent control of
combustion process for ore-roasting furnaceJournal of Control Theory and
Applications vol 6 pp 80-85 2008
[10] J Li W M Zhong H Cheng X D Kong and F Qian A data-driven soft
sensor modeling for furnace temperature of Opposed Multi-Burner gasifier pp
705-710 2011
[11] Y H Yang Y H Liu X Z Liu and S K Qin Billet temperature soft sensor
model of reheating furnace based on RVM method pp 4003-4006 2011
[12] J H Wang C Wang X F Zhu and X K Fang Application of soft sensor in
welding seam tracking prediction based on LSSVM and PSO with compression
factor pp 2441-2446 2013
[13] L Balbis J Balderud and M J Grimble Nonlinear predictive control of steel
slab reheating furnace pp 1679-1684 2008
[14] A Steinboeck D Wild T Kiefer and A Kugi A mathematical model of a slab
reheating furnace with radiative heat transfer and non-participating gaseous
media International Journal Of Heat And Mass Transfer vol 53 pp 5933-
5946 Dec 2010
[15] A Steinboeck D Wild and A Kugi Nonlinear model predictive control of a
continuous slab reheating furnace Control Engineering Practice vol 21 pp
495-508 2013
27
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[16] C Zhang T Ishii and S Sugiyama Numerical Modeling Of the Thermal
Performance Of Regenerative Slab Reheat Furnaces Numerical Heat Transfer
Part A Applications vol 32 pp 613-631 1997
[17] J G Kim and K Y Huh Prediction of Transient Slab Temperature Distribution
in the Re-heating Furnace of a Walking-beam Type for Rolling of Steel Slabs
ISIJ International vol 40 pp 1115-1123 2000
[18] J G Kim K Y and H I T K Three-Dimensional Analysis Of the Walking-
Beam-Type Slab Reheating Furnace In Hot Strip Mills Numerical Heat
Transfer Part A Applications vol 38 pp 589-609 2000
[19] CT Hsieh MJ Huang ST Lee and CH Wang Numerical Modeling of a
Walking-Beam-Type Slab Reheating Furnace Numerical Heat Transfer Part A
Applications vol 53 pp 966-981 2008
[20] MJ Huang CT Hsieh ST Lee and CH Wang A Coupled Numerical Study
of Slab Temperature and Gas Temperature in the Walking-Beam-Type Slab
Reheating Furnace Numerical Heat Transfer Part A Applications vol 54 pp
625-646 2008
[21] Z Song and A Kusiak Constraint-Based Control of Boiler Efficiency A Data-
Mining Approach IEEE Transactions on Industrial Informatics vol 3 pp 73-
83 2007
[22] J Q Li J J Gu and C L Niu The Operation Optimization based on
Correlation Analysis of Operation Parameters in Power Plant pp 138-141
2008
28
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[23] J G Wang S S Shieh S S Jang D S H Wong and C W Wu A two-tier
approach to the data-driven modeling on thermal efficiency of a BFGcoal co-
firing boiler Fuel vol 111 pp 528-534 Sep 2013
[24] L Breiman Better Subset Regression Using the Nonnegative Garrote
Technometrics vol 37 pp 373-384 1995
[25] R A Fisher The use of multiple measurements in taxonomic problems
AnnHum Genet vol 7 pp 179-188 1936
[26] J Liu Developing a soft sensor based on sparse partial least squares with
variable selection Journal of Process Control vol 24 pp 1046-1056 2014
[27] K Sun J Liu JL Kang SS Jang D SH Wong and DS Chen
Development of a variable selection method for soft sensor using artificial
neural network and nonnegative garrote Journal of Process Control vol 24 pp
1068-1075 2014
29
439
440
441
442
443
444
445
446
447
448
449
450
451
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
operation the temperature fluctuates around the target value because of the imperfect
control performance However in order to simplify the analysis and compute the
maximum possible energy saving perfect temperature control is assumed when
adopting the model-based optimization strategy In other words it is assumed that the
actual temperature in the lsquouprsquo area of the soaking zone can be adjusted to the expected
temperature ie 1200
0 02 04 06 08 1 12 14 16 18 2
x 104
1150
1200
1250
1300
Sampling intervals
T S-u (
)
OriginalTarget
Fig 11 Original and target temperature in the lsquouprsquo area of the soaking zone
The adjustment scheme takes the maximum value and minimum value of the original
operation data as the upper and lower bounds for the adjustment Moreover in order to
assure the validity of the linear model the increment or decrement of AG and G should
not beyond of the original value (considered as 10 for the purposes of this paper)
The adjustment strategy can be formulated as
19
307
308
309
310
311
312
313
314
315
316
317
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
(3)
where NNG() denotes the NNG regression model of the TG ratio and are the
NNG regression coefficients of the AG and G in the lsquouprsquo area of the soaking zone for
modeling TG in same area
With the model-based strategy and adaptive model of the TG ratio given in Section
42 the optimization adjustment results can be obtained as follows
The adjustment of gas flow in the lsquouprsquo area of the soaking zone shown in Fig 12
illustrates that the implementation of the model-based optimization operation can reduce
the consumption of the gas flow Compared with the original case 938 of the gas can
be saved on average by utilizing the model-based adjustment
0 02 04 06 08 1 12 14 16 18 2
x 104
-800
-700
-600
-500
-400
-300
-200
-100
0
Sampling intervals
Adj
ustm
ent o
f gas
flow
m(
3 h)
20
318
319
320
321
322
323
324
325
326
327
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
(a) Adjustment amount of gas flow
0 02 04 06 08 1 12 14 16 18 2
x 104
500
1000
1500
2000
2500
3000
3500
4000
4500
5000G
as fl
owm
(3 h
)
Sampling intervals
OriginalAdjusted
(b) Comparison of original and adjusted gas flow
095 096 097 098 099 1 101 102 103 104 105
x 104
500
1000
1500
2000
2500
3000
3500
4000
4500
5000
Gas
flow
m(
3 h)
Sampling intervals
OriginalAdjusted
(c) Comparison of original and adjusted gas flow in interval [9501-10500]
Fig 12 Adjustment of gas flow in the lsquouprsquo area of the soaking zone
The adjustment of the air flow in the lsquouprsquo area of the soaking zone given in Fig 13
shows that the air flow is reduced in most cases and only increased for a few cases In
21
328
329
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
general 681 of the air flow is reduced after the implementation of the model-based
optimization
0 02 04 06 08 1 12 14 16 18 2
x 104
-2000
-1500
-1000
-500
0
500
1000
Sampling intervals
Adj
ustm
ent o
f air
flow
(m3 h
)
(a) Adjustment amount of air flow
0 02 04 06 08 1 12 14 16 18 2
x 104
4000
5000
6000
7000
8000
9000
10000
Air
flow
m(
3 h)
Sampling intervals
OriginalAdjusted
(b) Comparison of original and adjusted air flow
22
330
331
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
095 096 097 098 099 1 101 102 103 104 105
x 104
4000
5000
6000
7000
8000
9000
10000
Air
flow
m(
3 h)
Sampling intervals
OriginalAdjusted
(c) Comparison of original and adjusted air flow in interval [9501-10500]
Fig 13 Adjustment of air flow in the lsquouprsquo area of the soaking zone
52 Discussions
In this section detailed analysis for the optimization operation results is provided
The statistics of optimization at the boundary conditions shown in Table 3 illustrates
that the adjusted value for G reaches its lower limit in most cases (8045 for
and 687 for ) which is the most energy-efficient point In a number of cases
(1079 for and 001 for ) the adjusted AG reaches its upper
limit These results indicate that the optimization operations maximize the combustion
efficiency by decreasing G and increasing the AG ratio up to the boundary conditions
However the lower limit is also unexpectedly attained in a small number of cases
(110 for and 078 for ) for which a more detailed analysis is
given below
23
332
333
334
335
336
337
338
339
340
341
342
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
Table 3 Statistics of optimization at boundary conditions
Total
Amount 0 1374 0 16090 17464
Percentag
e 0 687 0 8045
8732
Total
Amount 2 220 2158 156 2536
Percentag
e 001 110 1079 078
1268
Table 4 Statistics for different optimization operations
Adjustment
Gdarr
Guarr TotalAGuarr AGdarr
Tdarr Tuarr Tdarr Tuarr
Amount 9548 3744 6328 380 0 20000
Percentage 4774 1872 3164 190 0 100
The statistics for different optimization operations is performed and shown in Table
4 It can be seen that all the adjusted operations result in the reduction of gas
consumption In over half the cases (4774 with decreased temperature and 1872
with increased temperature) AG is adjusted to higher levels Meanwhile a number of
the adjustments (3164) lower the temperature by reducing the G and AG at the same
time This result indicates that the original temperature cannot be adjusted to the target
by only reducing G in the constraint conditions The remaining cases (19) are
24
343
344
345
346
347
348
349
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
relatively special in which an excess of air is supplied in the original operation and the
adjusted operation thereby increasing the temperature by a smaller AG with less gas
consumption
6 Conclusion
Based on the actual operation data this paper aims to explore the improvement of the
combustion efficiency and the room for energy conservation Correlation analysis and
LDA show that it is of great significance to introduce two derived ratio variables which
are the AG ratio and the TG ratio A type of combustion efficiency utilizing an on-line
soft sensor is put forward by employing a NNG variable selection algorithm which
provides a good solution to the combustion efficiency real-time measurement problem
of a reheating furnace The implementation of the model-based optimization is studied
based on the actual operational data Detailed analysis for the optimization results is
given for the different cases The results show that significant energy conservation can
be achieved when the furnace operation is optimized by using the developed soft sensor
model
Acknowledgement
The authors would like to thank the financial support provided by the National Nature
Science Foundation of China under Grant 61171145 Y Yao was supported by Ministry
of Science amp Technology ROC under Grant number MOST 104-2221-E-007-129
References
25
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[1] Z J Wang Q D Wu and T Y Chai Optimal-setting control for complicated
industrial processes and its application studyControl Engineering Practice vol
12 pp 65-74 2004
[2] A Steinboeck K Graichen and A Kugi Dynamic Optimization of a Slab
Reheating Furnace With Consistent Approximation of Control VariablesIEEE
Transactions on Control Systems Technology vol 19 pp 1444-1456 2011
[3] B T Zhang C Y Wang Q Qin and L Li Influence of Boiler Combustion
Adjustment on NOxEmission and Boiler EfficiencyAdvanced Materials
Research vol 732-733 pp 234-237 2013
[4] C K Yoo and IB Lee Soft Sensor and Adaptive Model-Based Dissolved
Oxygen Control for Biological Wastewater Treatment ProcessesEnvironmental
Engineering Science vol 21 pp 331-340 2004
[5] S A Bhat D N Saraf S Gupta and S K Gupta Use of Agitator Power as a
Soft Sensor for Bulk Free-Radical Polymerization of Methyl Methacrylate in
Batch ReactorsIndustrial amp Engineering Chemistry Research vol 45 pp 4243-
4255 2006
[6] K Desai Y Badhe S S Tambe and B D Kulkarni Soft-sensor development
for fed-batch bioreactors using support vector regressionBiochemical
Engineering Journal vol 27 pp 225-239 2006
[7] Y P Badhe Lonari J Tambe S S amp Kulkarni B D Improve polyethylene
process control and product qualityHydrocarbon Processing vol 86 pp 53-60
2007
26
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[8] N K Nath K Mandal A K Singh B Basu C Bhanu S Kumar et al Ladle
furnace on-line reckoner for prediction and control of steel temperature and
compositionIronmaking amp Steelmaking vol 33 pp 140-150 2006
[9] A J Yan T Y Chai F H Wu and P Wang Hybrid intelligent control of
combustion process for ore-roasting furnaceJournal of Control Theory and
Applications vol 6 pp 80-85 2008
[10] J Li W M Zhong H Cheng X D Kong and F Qian A data-driven soft
sensor modeling for furnace temperature of Opposed Multi-Burner gasifier pp
705-710 2011
[11] Y H Yang Y H Liu X Z Liu and S K Qin Billet temperature soft sensor
model of reheating furnace based on RVM method pp 4003-4006 2011
[12] J H Wang C Wang X F Zhu and X K Fang Application of soft sensor in
welding seam tracking prediction based on LSSVM and PSO with compression
factor pp 2441-2446 2013
[13] L Balbis J Balderud and M J Grimble Nonlinear predictive control of steel
slab reheating furnace pp 1679-1684 2008
[14] A Steinboeck D Wild T Kiefer and A Kugi A mathematical model of a slab
reheating furnace with radiative heat transfer and non-participating gaseous
media International Journal Of Heat And Mass Transfer vol 53 pp 5933-
5946 Dec 2010
[15] A Steinboeck D Wild and A Kugi Nonlinear model predictive control of a
continuous slab reheating furnace Control Engineering Practice vol 21 pp
495-508 2013
27
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[16] C Zhang T Ishii and S Sugiyama Numerical Modeling Of the Thermal
Performance Of Regenerative Slab Reheat Furnaces Numerical Heat Transfer
Part A Applications vol 32 pp 613-631 1997
[17] J G Kim and K Y Huh Prediction of Transient Slab Temperature Distribution
in the Re-heating Furnace of a Walking-beam Type for Rolling of Steel Slabs
ISIJ International vol 40 pp 1115-1123 2000
[18] J G Kim K Y and H I T K Three-Dimensional Analysis Of the Walking-
Beam-Type Slab Reheating Furnace In Hot Strip Mills Numerical Heat
Transfer Part A Applications vol 38 pp 589-609 2000
[19] CT Hsieh MJ Huang ST Lee and CH Wang Numerical Modeling of a
Walking-Beam-Type Slab Reheating Furnace Numerical Heat Transfer Part A
Applications vol 53 pp 966-981 2008
[20] MJ Huang CT Hsieh ST Lee and CH Wang A Coupled Numerical Study
of Slab Temperature and Gas Temperature in the Walking-Beam-Type Slab
Reheating Furnace Numerical Heat Transfer Part A Applications vol 54 pp
625-646 2008
[21] Z Song and A Kusiak Constraint-Based Control of Boiler Efficiency A Data-
Mining Approach IEEE Transactions on Industrial Informatics vol 3 pp 73-
83 2007
[22] J Q Li J J Gu and C L Niu The Operation Optimization based on
Correlation Analysis of Operation Parameters in Power Plant pp 138-141
2008
28
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[23] J G Wang S S Shieh S S Jang D S H Wong and C W Wu A two-tier
approach to the data-driven modeling on thermal efficiency of a BFGcoal co-
firing boiler Fuel vol 111 pp 528-534 Sep 2013
[24] L Breiman Better Subset Regression Using the Nonnegative Garrote
Technometrics vol 37 pp 373-384 1995
[25] R A Fisher The use of multiple measurements in taxonomic problems
AnnHum Genet vol 7 pp 179-188 1936
[26] J Liu Developing a soft sensor based on sparse partial least squares with
variable selection Journal of Process Control vol 24 pp 1046-1056 2014
[27] K Sun J Liu JL Kang SS Jang D SH Wong and DS Chen
Development of a variable selection method for soft sensor using artificial
neural network and nonnegative garrote Journal of Process Control vol 24 pp
1068-1075 2014
29
439
440
441
442
443
444
445
446
447
448
449
450
451
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
(3)
where NNG() denotes the NNG regression model of the TG ratio and are the
NNG regression coefficients of the AG and G in the lsquouprsquo area of the soaking zone for
modeling TG in same area
With the model-based strategy and adaptive model of the TG ratio given in Section
42 the optimization adjustment results can be obtained as follows
The adjustment of gas flow in the lsquouprsquo area of the soaking zone shown in Fig 12
illustrates that the implementation of the model-based optimization operation can reduce
the consumption of the gas flow Compared with the original case 938 of the gas can
be saved on average by utilizing the model-based adjustment
0 02 04 06 08 1 12 14 16 18 2
x 104
-800
-700
-600
-500
-400
-300
-200
-100
0
Sampling intervals
Adj
ustm
ent o
f gas
flow
m(
3 h)
20
318
319
320
321
322
323
324
325
326
327
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
(a) Adjustment amount of gas flow
0 02 04 06 08 1 12 14 16 18 2
x 104
500
1000
1500
2000
2500
3000
3500
4000
4500
5000G
as fl
owm
(3 h
)
Sampling intervals
OriginalAdjusted
(b) Comparison of original and adjusted gas flow
095 096 097 098 099 1 101 102 103 104 105
x 104
500
1000
1500
2000
2500
3000
3500
4000
4500
5000
Gas
flow
m(
3 h)
Sampling intervals
OriginalAdjusted
(c) Comparison of original and adjusted gas flow in interval [9501-10500]
Fig 12 Adjustment of gas flow in the lsquouprsquo area of the soaking zone
The adjustment of the air flow in the lsquouprsquo area of the soaking zone given in Fig 13
shows that the air flow is reduced in most cases and only increased for a few cases In
21
328
329
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
general 681 of the air flow is reduced after the implementation of the model-based
optimization
0 02 04 06 08 1 12 14 16 18 2
x 104
-2000
-1500
-1000
-500
0
500
1000
Sampling intervals
Adj
ustm
ent o
f air
flow
(m3 h
)
(a) Adjustment amount of air flow
0 02 04 06 08 1 12 14 16 18 2
x 104
4000
5000
6000
7000
8000
9000
10000
Air
flow
m(
3 h)
Sampling intervals
OriginalAdjusted
(b) Comparison of original and adjusted air flow
22
330
331
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
095 096 097 098 099 1 101 102 103 104 105
x 104
4000
5000
6000
7000
8000
9000
10000
Air
flow
m(
3 h)
Sampling intervals
OriginalAdjusted
(c) Comparison of original and adjusted air flow in interval [9501-10500]
Fig 13 Adjustment of air flow in the lsquouprsquo area of the soaking zone
52 Discussions
In this section detailed analysis for the optimization operation results is provided
The statistics of optimization at the boundary conditions shown in Table 3 illustrates
that the adjusted value for G reaches its lower limit in most cases (8045 for
and 687 for ) which is the most energy-efficient point In a number of cases
(1079 for and 001 for ) the adjusted AG reaches its upper
limit These results indicate that the optimization operations maximize the combustion
efficiency by decreasing G and increasing the AG ratio up to the boundary conditions
However the lower limit is also unexpectedly attained in a small number of cases
(110 for and 078 for ) for which a more detailed analysis is
given below
23
332
333
334
335
336
337
338
339
340
341
342
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
Table 3 Statistics of optimization at boundary conditions
Total
Amount 0 1374 0 16090 17464
Percentag
e 0 687 0 8045
8732
Total
Amount 2 220 2158 156 2536
Percentag
e 001 110 1079 078
1268
Table 4 Statistics for different optimization operations
Adjustment
Gdarr
Guarr TotalAGuarr AGdarr
Tdarr Tuarr Tdarr Tuarr
Amount 9548 3744 6328 380 0 20000
Percentage 4774 1872 3164 190 0 100
The statistics for different optimization operations is performed and shown in Table
4 It can be seen that all the adjusted operations result in the reduction of gas
consumption In over half the cases (4774 with decreased temperature and 1872
with increased temperature) AG is adjusted to higher levels Meanwhile a number of
the adjustments (3164) lower the temperature by reducing the G and AG at the same
time This result indicates that the original temperature cannot be adjusted to the target
by only reducing G in the constraint conditions The remaining cases (19) are
24
343
344
345
346
347
348
349
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
relatively special in which an excess of air is supplied in the original operation and the
adjusted operation thereby increasing the temperature by a smaller AG with less gas
consumption
6 Conclusion
Based on the actual operation data this paper aims to explore the improvement of the
combustion efficiency and the room for energy conservation Correlation analysis and
LDA show that it is of great significance to introduce two derived ratio variables which
are the AG ratio and the TG ratio A type of combustion efficiency utilizing an on-line
soft sensor is put forward by employing a NNG variable selection algorithm which
provides a good solution to the combustion efficiency real-time measurement problem
of a reheating furnace The implementation of the model-based optimization is studied
based on the actual operational data Detailed analysis for the optimization results is
given for the different cases The results show that significant energy conservation can
be achieved when the furnace operation is optimized by using the developed soft sensor
model
Acknowledgement
The authors would like to thank the financial support provided by the National Nature
Science Foundation of China under Grant 61171145 Y Yao was supported by Ministry
of Science amp Technology ROC under Grant number MOST 104-2221-E-007-129
References
25
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[1] Z J Wang Q D Wu and T Y Chai Optimal-setting control for complicated
industrial processes and its application studyControl Engineering Practice vol
12 pp 65-74 2004
[2] A Steinboeck K Graichen and A Kugi Dynamic Optimization of a Slab
Reheating Furnace With Consistent Approximation of Control VariablesIEEE
Transactions on Control Systems Technology vol 19 pp 1444-1456 2011
[3] B T Zhang C Y Wang Q Qin and L Li Influence of Boiler Combustion
Adjustment on NOxEmission and Boiler EfficiencyAdvanced Materials
Research vol 732-733 pp 234-237 2013
[4] C K Yoo and IB Lee Soft Sensor and Adaptive Model-Based Dissolved
Oxygen Control for Biological Wastewater Treatment ProcessesEnvironmental
Engineering Science vol 21 pp 331-340 2004
[5] S A Bhat D N Saraf S Gupta and S K Gupta Use of Agitator Power as a
Soft Sensor for Bulk Free-Radical Polymerization of Methyl Methacrylate in
Batch ReactorsIndustrial amp Engineering Chemistry Research vol 45 pp 4243-
4255 2006
[6] K Desai Y Badhe S S Tambe and B D Kulkarni Soft-sensor development
for fed-batch bioreactors using support vector regressionBiochemical
Engineering Journal vol 27 pp 225-239 2006
[7] Y P Badhe Lonari J Tambe S S amp Kulkarni B D Improve polyethylene
process control and product qualityHydrocarbon Processing vol 86 pp 53-60
2007
26
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[8] N K Nath K Mandal A K Singh B Basu C Bhanu S Kumar et al Ladle
furnace on-line reckoner for prediction and control of steel temperature and
compositionIronmaking amp Steelmaking vol 33 pp 140-150 2006
[9] A J Yan T Y Chai F H Wu and P Wang Hybrid intelligent control of
combustion process for ore-roasting furnaceJournal of Control Theory and
Applications vol 6 pp 80-85 2008
[10] J Li W M Zhong H Cheng X D Kong and F Qian A data-driven soft
sensor modeling for furnace temperature of Opposed Multi-Burner gasifier pp
705-710 2011
[11] Y H Yang Y H Liu X Z Liu and S K Qin Billet temperature soft sensor
model of reheating furnace based on RVM method pp 4003-4006 2011
[12] J H Wang C Wang X F Zhu and X K Fang Application of soft sensor in
welding seam tracking prediction based on LSSVM and PSO with compression
factor pp 2441-2446 2013
[13] L Balbis J Balderud and M J Grimble Nonlinear predictive control of steel
slab reheating furnace pp 1679-1684 2008
[14] A Steinboeck D Wild T Kiefer and A Kugi A mathematical model of a slab
reheating furnace with radiative heat transfer and non-participating gaseous
media International Journal Of Heat And Mass Transfer vol 53 pp 5933-
5946 Dec 2010
[15] A Steinboeck D Wild and A Kugi Nonlinear model predictive control of a
continuous slab reheating furnace Control Engineering Practice vol 21 pp
495-508 2013
27
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[16] C Zhang T Ishii and S Sugiyama Numerical Modeling Of the Thermal
Performance Of Regenerative Slab Reheat Furnaces Numerical Heat Transfer
Part A Applications vol 32 pp 613-631 1997
[17] J G Kim and K Y Huh Prediction of Transient Slab Temperature Distribution
in the Re-heating Furnace of a Walking-beam Type for Rolling of Steel Slabs
ISIJ International vol 40 pp 1115-1123 2000
[18] J G Kim K Y and H I T K Three-Dimensional Analysis Of the Walking-
Beam-Type Slab Reheating Furnace In Hot Strip Mills Numerical Heat
Transfer Part A Applications vol 38 pp 589-609 2000
[19] CT Hsieh MJ Huang ST Lee and CH Wang Numerical Modeling of a
Walking-Beam-Type Slab Reheating Furnace Numerical Heat Transfer Part A
Applications vol 53 pp 966-981 2008
[20] MJ Huang CT Hsieh ST Lee and CH Wang A Coupled Numerical Study
of Slab Temperature and Gas Temperature in the Walking-Beam-Type Slab
Reheating Furnace Numerical Heat Transfer Part A Applications vol 54 pp
625-646 2008
[21] Z Song and A Kusiak Constraint-Based Control of Boiler Efficiency A Data-
Mining Approach IEEE Transactions on Industrial Informatics vol 3 pp 73-
83 2007
[22] J Q Li J J Gu and C L Niu The Operation Optimization based on
Correlation Analysis of Operation Parameters in Power Plant pp 138-141
2008
28
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[23] J G Wang S S Shieh S S Jang D S H Wong and C W Wu A two-tier
approach to the data-driven modeling on thermal efficiency of a BFGcoal co-
firing boiler Fuel vol 111 pp 528-534 Sep 2013
[24] L Breiman Better Subset Regression Using the Nonnegative Garrote
Technometrics vol 37 pp 373-384 1995
[25] R A Fisher The use of multiple measurements in taxonomic problems
AnnHum Genet vol 7 pp 179-188 1936
[26] J Liu Developing a soft sensor based on sparse partial least squares with
variable selection Journal of Process Control vol 24 pp 1046-1056 2014
[27] K Sun J Liu JL Kang SS Jang D SH Wong and DS Chen
Development of a variable selection method for soft sensor using artificial
neural network and nonnegative garrote Journal of Process Control vol 24 pp
1068-1075 2014
29
439
440
441
442
443
444
445
446
447
448
449
450
451
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
(a) Adjustment amount of gas flow
0 02 04 06 08 1 12 14 16 18 2
x 104
500
1000
1500
2000
2500
3000
3500
4000
4500
5000G
as fl
owm
(3 h
)
Sampling intervals
OriginalAdjusted
(b) Comparison of original and adjusted gas flow
095 096 097 098 099 1 101 102 103 104 105
x 104
500
1000
1500
2000
2500
3000
3500
4000
4500
5000
Gas
flow
m(
3 h)
Sampling intervals
OriginalAdjusted
(c) Comparison of original and adjusted gas flow in interval [9501-10500]
Fig 12 Adjustment of gas flow in the lsquouprsquo area of the soaking zone
The adjustment of the air flow in the lsquouprsquo area of the soaking zone given in Fig 13
shows that the air flow is reduced in most cases and only increased for a few cases In
21
328
329
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
general 681 of the air flow is reduced after the implementation of the model-based
optimization
0 02 04 06 08 1 12 14 16 18 2
x 104
-2000
-1500
-1000
-500
0
500
1000
Sampling intervals
Adj
ustm
ent o
f air
flow
(m3 h
)
(a) Adjustment amount of air flow
0 02 04 06 08 1 12 14 16 18 2
x 104
4000
5000
6000
7000
8000
9000
10000
Air
flow
m(
3 h)
Sampling intervals
OriginalAdjusted
(b) Comparison of original and adjusted air flow
22
330
331
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
095 096 097 098 099 1 101 102 103 104 105
x 104
4000
5000
6000
7000
8000
9000
10000
Air
flow
m(
3 h)
Sampling intervals
OriginalAdjusted
(c) Comparison of original and adjusted air flow in interval [9501-10500]
Fig 13 Adjustment of air flow in the lsquouprsquo area of the soaking zone
52 Discussions
In this section detailed analysis for the optimization operation results is provided
The statistics of optimization at the boundary conditions shown in Table 3 illustrates
that the adjusted value for G reaches its lower limit in most cases (8045 for
and 687 for ) which is the most energy-efficient point In a number of cases
(1079 for and 001 for ) the adjusted AG reaches its upper
limit These results indicate that the optimization operations maximize the combustion
efficiency by decreasing G and increasing the AG ratio up to the boundary conditions
However the lower limit is also unexpectedly attained in a small number of cases
(110 for and 078 for ) for which a more detailed analysis is
given below
23
332
333
334
335
336
337
338
339
340
341
342
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
Table 3 Statistics of optimization at boundary conditions
Total
Amount 0 1374 0 16090 17464
Percentag
e 0 687 0 8045
8732
Total
Amount 2 220 2158 156 2536
Percentag
e 001 110 1079 078
1268
Table 4 Statistics for different optimization operations
Adjustment
Gdarr
Guarr TotalAGuarr AGdarr
Tdarr Tuarr Tdarr Tuarr
Amount 9548 3744 6328 380 0 20000
Percentage 4774 1872 3164 190 0 100
The statistics for different optimization operations is performed and shown in Table
4 It can be seen that all the adjusted operations result in the reduction of gas
consumption In over half the cases (4774 with decreased temperature and 1872
with increased temperature) AG is adjusted to higher levels Meanwhile a number of
the adjustments (3164) lower the temperature by reducing the G and AG at the same
time This result indicates that the original temperature cannot be adjusted to the target
by only reducing G in the constraint conditions The remaining cases (19) are
24
343
344
345
346
347
348
349
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
relatively special in which an excess of air is supplied in the original operation and the
adjusted operation thereby increasing the temperature by a smaller AG with less gas
consumption
6 Conclusion
Based on the actual operation data this paper aims to explore the improvement of the
combustion efficiency and the room for energy conservation Correlation analysis and
LDA show that it is of great significance to introduce two derived ratio variables which
are the AG ratio and the TG ratio A type of combustion efficiency utilizing an on-line
soft sensor is put forward by employing a NNG variable selection algorithm which
provides a good solution to the combustion efficiency real-time measurement problem
of a reheating furnace The implementation of the model-based optimization is studied
based on the actual operational data Detailed analysis for the optimization results is
given for the different cases The results show that significant energy conservation can
be achieved when the furnace operation is optimized by using the developed soft sensor
model
Acknowledgement
The authors would like to thank the financial support provided by the National Nature
Science Foundation of China under Grant 61171145 Y Yao was supported by Ministry
of Science amp Technology ROC under Grant number MOST 104-2221-E-007-129
References
25
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[1] Z J Wang Q D Wu and T Y Chai Optimal-setting control for complicated
industrial processes and its application studyControl Engineering Practice vol
12 pp 65-74 2004
[2] A Steinboeck K Graichen and A Kugi Dynamic Optimization of a Slab
Reheating Furnace With Consistent Approximation of Control VariablesIEEE
Transactions on Control Systems Technology vol 19 pp 1444-1456 2011
[3] B T Zhang C Y Wang Q Qin and L Li Influence of Boiler Combustion
Adjustment on NOxEmission and Boiler EfficiencyAdvanced Materials
Research vol 732-733 pp 234-237 2013
[4] C K Yoo and IB Lee Soft Sensor and Adaptive Model-Based Dissolved
Oxygen Control for Biological Wastewater Treatment ProcessesEnvironmental
Engineering Science vol 21 pp 331-340 2004
[5] S A Bhat D N Saraf S Gupta and S K Gupta Use of Agitator Power as a
Soft Sensor for Bulk Free-Radical Polymerization of Methyl Methacrylate in
Batch ReactorsIndustrial amp Engineering Chemistry Research vol 45 pp 4243-
4255 2006
[6] K Desai Y Badhe S S Tambe and B D Kulkarni Soft-sensor development
for fed-batch bioreactors using support vector regressionBiochemical
Engineering Journal vol 27 pp 225-239 2006
[7] Y P Badhe Lonari J Tambe S S amp Kulkarni B D Improve polyethylene
process control and product qualityHydrocarbon Processing vol 86 pp 53-60
2007
26
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[8] N K Nath K Mandal A K Singh B Basu C Bhanu S Kumar et al Ladle
furnace on-line reckoner for prediction and control of steel temperature and
compositionIronmaking amp Steelmaking vol 33 pp 140-150 2006
[9] A J Yan T Y Chai F H Wu and P Wang Hybrid intelligent control of
combustion process for ore-roasting furnaceJournal of Control Theory and
Applications vol 6 pp 80-85 2008
[10] J Li W M Zhong H Cheng X D Kong and F Qian A data-driven soft
sensor modeling for furnace temperature of Opposed Multi-Burner gasifier pp
705-710 2011
[11] Y H Yang Y H Liu X Z Liu and S K Qin Billet temperature soft sensor
model of reheating furnace based on RVM method pp 4003-4006 2011
[12] J H Wang C Wang X F Zhu and X K Fang Application of soft sensor in
welding seam tracking prediction based on LSSVM and PSO with compression
factor pp 2441-2446 2013
[13] L Balbis J Balderud and M J Grimble Nonlinear predictive control of steel
slab reheating furnace pp 1679-1684 2008
[14] A Steinboeck D Wild T Kiefer and A Kugi A mathematical model of a slab
reheating furnace with radiative heat transfer and non-participating gaseous
media International Journal Of Heat And Mass Transfer vol 53 pp 5933-
5946 Dec 2010
[15] A Steinboeck D Wild and A Kugi Nonlinear model predictive control of a
continuous slab reheating furnace Control Engineering Practice vol 21 pp
495-508 2013
27
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[16] C Zhang T Ishii and S Sugiyama Numerical Modeling Of the Thermal
Performance Of Regenerative Slab Reheat Furnaces Numerical Heat Transfer
Part A Applications vol 32 pp 613-631 1997
[17] J G Kim and K Y Huh Prediction of Transient Slab Temperature Distribution
in the Re-heating Furnace of a Walking-beam Type for Rolling of Steel Slabs
ISIJ International vol 40 pp 1115-1123 2000
[18] J G Kim K Y and H I T K Three-Dimensional Analysis Of the Walking-
Beam-Type Slab Reheating Furnace In Hot Strip Mills Numerical Heat
Transfer Part A Applications vol 38 pp 589-609 2000
[19] CT Hsieh MJ Huang ST Lee and CH Wang Numerical Modeling of a
Walking-Beam-Type Slab Reheating Furnace Numerical Heat Transfer Part A
Applications vol 53 pp 966-981 2008
[20] MJ Huang CT Hsieh ST Lee and CH Wang A Coupled Numerical Study
of Slab Temperature and Gas Temperature in the Walking-Beam-Type Slab
Reheating Furnace Numerical Heat Transfer Part A Applications vol 54 pp
625-646 2008
[21] Z Song and A Kusiak Constraint-Based Control of Boiler Efficiency A Data-
Mining Approach IEEE Transactions on Industrial Informatics vol 3 pp 73-
83 2007
[22] J Q Li J J Gu and C L Niu The Operation Optimization based on
Correlation Analysis of Operation Parameters in Power Plant pp 138-141
2008
28
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[23] J G Wang S S Shieh S S Jang D S H Wong and C W Wu A two-tier
approach to the data-driven modeling on thermal efficiency of a BFGcoal co-
firing boiler Fuel vol 111 pp 528-534 Sep 2013
[24] L Breiman Better Subset Regression Using the Nonnegative Garrote
Technometrics vol 37 pp 373-384 1995
[25] R A Fisher The use of multiple measurements in taxonomic problems
AnnHum Genet vol 7 pp 179-188 1936
[26] J Liu Developing a soft sensor based on sparse partial least squares with
variable selection Journal of Process Control vol 24 pp 1046-1056 2014
[27] K Sun J Liu JL Kang SS Jang D SH Wong and DS Chen
Development of a variable selection method for soft sensor using artificial
neural network and nonnegative garrote Journal of Process Control vol 24 pp
1068-1075 2014
29
439
440
441
442
443
444
445
446
447
448
449
450
451
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
general 681 of the air flow is reduced after the implementation of the model-based
optimization
0 02 04 06 08 1 12 14 16 18 2
x 104
-2000
-1500
-1000
-500
0
500
1000
Sampling intervals
Adj
ustm
ent o
f air
flow
(m3 h
)
(a) Adjustment amount of air flow
0 02 04 06 08 1 12 14 16 18 2
x 104
4000
5000
6000
7000
8000
9000
10000
Air
flow
m(
3 h)
Sampling intervals
OriginalAdjusted
(b) Comparison of original and adjusted air flow
22
330
331
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
095 096 097 098 099 1 101 102 103 104 105
x 104
4000
5000
6000
7000
8000
9000
10000
Air
flow
m(
3 h)
Sampling intervals
OriginalAdjusted
(c) Comparison of original and adjusted air flow in interval [9501-10500]
Fig 13 Adjustment of air flow in the lsquouprsquo area of the soaking zone
52 Discussions
In this section detailed analysis for the optimization operation results is provided
The statistics of optimization at the boundary conditions shown in Table 3 illustrates
that the adjusted value for G reaches its lower limit in most cases (8045 for
and 687 for ) which is the most energy-efficient point In a number of cases
(1079 for and 001 for ) the adjusted AG reaches its upper
limit These results indicate that the optimization operations maximize the combustion
efficiency by decreasing G and increasing the AG ratio up to the boundary conditions
However the lower limit is also unexpectedly attained in a small number of cases
(110 for and 078 for ) for which a more detailed analysis is
given below
23
332
333
334
335
336
337
338
339
340
341
342
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
Table 3 Statistics of optimization at boundary conditions
Total
Amount 0 1374 0 16090 17464
Percentag
e 0 687 0 8045
8732
Total
Amount 2 220 2158 156 2536
Percentag
e 001 110 1079 078
1268
Table 4 Statistics for different optimization operations
Adjustment
Gdarr
Guarr TotalAGuarr AGdarr
Tdarr Tuarr Tdarr Tuarr
Amount 9548 3744 6328 380 0 20000
Percentage 4774 1872 3164 190 0 100
The statistics for different optimization operations is performed and shown in Table
4 It can be seen that all the adjusted operations result in the reduction of gas
consumption In over half the cases (4774 with decreased temperature and 1872
with increased temperature) AG is adjusted to higher levels Meanwhile a number of
the adjustments (3164) lower the temperature by reducing the G and AG at the same
time This result indicates that the original temperature cannot be adjusted to the target
by only reducing G in the constraint conditions The remaining cases (19) are
24
343
344
345
346
347
348
349
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
relatively special in which an excess of air is supplied in the original operation and the
adjusted operation thereby increasing the temperature by a smaller AG with less gas
consumption
6 Conclusion
Based on the actual operation data this paper aims to explore the improvement of the
combustion efficiency and the room for energy conservation Correlation analysis and
LDA show that it is of great significance to introduce two derived ratio variables which
are the AG ratio and the TG ratio A type of combustion efficiency utilizing an on-line
soft sensor is put forward by employing a NNG variable selection algorithm which
provides a good solution to the combustion efficiency real-time measurement problem
of a reheating furnace The implementation of the model-based optimization is studied
based on the actual operational data Detailed analysis for the optimization results is
given for the different cases The results show that significant energy conservation can
be achieved when the furnace operation is optimized by using the developed soft sensor
model
Acknowledgement
The authors would like to thank the financial support provided by the National Nature
Science Foundation of China under Grant 61171145 Y Yao was supported by Ministry
of Science amp Technology ROC under Grant number MOST 104-2221-E-007-129
References
25
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[1] Z J Wang Q D Wu and T Y Chai Optimal-setting control for complicated
industrial processes and its application studyControl Engineering Practice vol
12 pp 65-74 2004
[2] A Steinboeck K Graichen and A Kugi Dynamic Optimization of a Slab
Reheating Furnace With Consistent Approximation of Control VariablesIEEE
Transactions on Control Systems Technology vol 19 pp 1444-1456 2011
[3] B T Zhang C Y Wang Q Qin and L Li Influence of Boiler Combustion
Adjustment on NOxEmission and Boiler EfficiencyAdvanced Materials
Research vol 732-733 pp 234-237 2013
[4] C K Yoo and IB Lee Soft Sensor and Adaptive Model-Based Dissolved
Oxygen Control for Biological Wastewater Treatment ProcessesEnvironmental
Engineering Science vol 21 pp 331-340 2004
[5] S A Bhat D N Saraf S Gupta and S K Gupta Use of Agitator Power as a
Soft Sensor for Bulk Free-Radical Polymerization of Methyl Methacrylate in
Batch ReactorsIndustrial amp Engineering Chemistry Research vol 45 pp 4243-
4255 2006
[6] K Desai Y Badhe S S Tambe and B D Kulkarni Soft-sensor development
for fed-batch bioreactors using support vector regressionBiochemical
Engineering Journal vol 27 pp 225-239 2006
[7] Y P Badhe Lonari J Tambe S S amp Kulkarni B D Improve polyethylene
process control and product qualityHydrocarbon Processing vol 86 pp 53-60
2007
26
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[8] N K Nath K Mandal A K Singh B Basu C Bhanu S Kumar et al Ladle
furnace on-line reckoner for prediction and control of steel temperature and
compositionIronmaking amp Steelmaking vol 33 pp 140-150 2006
[9] A J Yan T Y Chai F H Wu and P Wang Hybrid intelligent control of
combustion process for ore-roasting furnaceJournal of Control Theory and
Applications vol 6 pp 80-85 2008
[10] J Li W M Zhong H Cheng X D Kong and F Qian A data-driven soft
sensor modeling for furnace temperature of Opposed Multi-Burner gasifier pp
705-710 2011
[11] Y H Yang Y H Liu X Z Liu and S K Qin Billet temperature soft sensor
model of reheating furnace based on RVM method pp 4003-4006 2011
[12] J H Wang C Wang X F Zhu and X K Fang Application of soft sensor in
welding seam tracking prediction based on LSSVM and PSO with compression
factor pp 2441-2446 2013
[13] L Balbis J Balderud and M J Grimble Nonlinear predictive control of steel
slab reheating furnace pp 1679-1684 2008
[14] A Steinboeck D Wild T Kiefer and A Kugi A mathematical model of a slab
reheating furnace with radiative heat transfer and non-participating gaseous
media International Journal Of Heat And Mass Transfer vol 53 pp 5933-
5946 Dec 2010
[15] A Steinboeck D Wild and A Kugi Nonlinear model predictive control of a
continuous slab reheating furnace Control Engineering Practice vol 21 pp
495-508 2013
27
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[16] C Zhang T Ishii and S Sugiyama Numerical Modeling Of the Thermal
Performance Of Regenerative Slab Reheat Furnaces Numerical Heat Transfer
Part A Applications vol 32 pp 613-631 1997
[17] J G Kim and K Y Huh Prediction of Transient Slab Temperature Distribution
in the Re-heating Furnace of a Walking-beam Type for Rolling of Steel Slabs
ISIJ International vol 40 pp 1115-1123 2000
[18] J G Kim K Y and H I T K Three-Dimensional Analysis Of the Walking-
Beam-Type Slab Reheating Furnace In Hot Strip Mills Numerical Heat
Transfer Part A Applications vol 38 pp 589-609 2000
[19] CT Hsieh MJ Huang ST Lee and CH Wang Numerical Modeling of a
Walking-Beam-Type Slab Reheating Furnace Numerical Heat Transfer Part A
Applications vol 53 pp 966-981 2008
[20] MJ Huang CT Hsieh ST Lee and CH Wang A Coupled Numerical Study
of Slab Temperature and Gas Temperature in the Walking-Beam-Type Slab
Reheating Furnace Numerical Heat Transfer Part A Applications vol 54 pp
625-646 2008
[21] Z Song and A Kusiak Constraint-Based Control of Boiler Efficiency A Data-
Mining Approach IEEE Transactions on Industrial Informatics vol 3 pp 73-
83 2007
[22] J Q Li J J Gu and C L Niu The Operation Optimization based on
Correlation Analysis of Operation Parameters in Power Plant pp 138-141
2008
28
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[23] J G Wang S S Shieh S S Jang D S H Wong and C W Wu A two-tier
approach to the data-driven modeling on thermal efficiency of a BFGcoal co-
firing boiler Fuel vol 111 pp 528-534 Sep 2013
[24] L Breiman Better Subset Regression Using the Nonnegative Garrote
Technometrics vol 37 pp 373-384 1995
[25] R A Fisher The use of multiple measurements in taxonomic problems
AnnHum Genet vol 7 pp 179-188 1936
[26] J Liu Developing a soft sensor based on sparse partial least squares with
variable selection Journal of Process Control vol 24 pp 1046-1056 2014
[27] K Sun J Liu JL Kang SS Jang D SH Wong and DS Chen
Development of a variable selection method for soft sensor using artificial
neural network and nonnegative garrote Journal of Process Control vol 24 pp
1068-1075 2014
29
439
440
441
442
443
444
445
446
447
448
449
450
451
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
095 096 097 098 099 1 101 102 103 104 105
x 104
4000
5000
6000
7000
8000
9000
10000
Air
flow
m(
3 h)
Sampling intervals
OriginalAdjusted
(c) Comparison of original and adjusted air flow in interval [9501-10500]
Fig 13 Adjustment of air flow in the lsquouprsquo area of the soaking zone
52 Discussions
In this section detailed analysis for the optimization operation results is provided
The statistics of optimization at the boundary conditions shown in Table 3 illustrates
that the adjusted value for G reaches its lower limit in most cases (8045 for
and 687 for ) which is the most energy-efficient point In a number of cases
(1079 for and 001 for ) the adjusted AG reaches its upper
limit These results indicate that the optimization operations maximize the combustion
efficiency by decreasing G and increasing the AG ratio up to the boundary conditions
However the lower limit is also unexpectedly attained in a small number of cases
(110 for and 078 for ) for which a more detailed analysis is
given below
23
332
333
334
335
336
337
338
339
340
341
342
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
Table 3 Statistics of optimization at boundary conditions
Total
Amount 0 1374 0 16090 17464
Percentag
e 0 687 0 8045
8732
Total
Amount 2 220 2158 156 2536
Percentag
e 001 110 1079 078
1268
Table 4 Statistics for different optimization operations
Adjustment
Gdarr
Guarr TotalAGuarr AGdarr
Tdarr Tuarr Tdarr Tuarr
Amount 9548 3744 6328 380 0 20000
Percentage 4774 1872 3164 190 0 100
The statistics for different optimization operations is performed and shown in Table
4 It can be seen that all the adjusted operations result in the reduction of gas
consumption In over half the cases (4774 with decreased temperature and 1872
with increased temperature) AG is adjusted to higher levels Meanwhile a number of
the adjustments (3164) lower the temperature by reducing the G and AG at the same
time This result indicates that the original temperature cannot be adjusted to the target
by only reducing G in the constraint conditions The remaining cases (19) are
24
343
344
345
346
347
348
349
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
relatively special in which an excess of air is supplied in the original operation and the
adjusted operation thereby increasing the temperature by a smaller AG with less gas
consumption
6 Conclusion
Based on the actual operation data this paper aims to explore the improvement of the
combustion efficiency and the room for energy conservation Correlation analysis and
LDA show that it is of great significance to introduce two derived ratio variables which
are the AG ratio and the TG ratio A type of combustion efficiency utilizing an on-line
soft sensor is put forward by employing a NNG variable selection algorithm which
provides a good solution to the combustion efficiency real-time measurement problem
of a reheating furnace The implementation of the model-based optimization is studied
based on the actual operational data Detailed analysis for the optimization results is
given for the different cases The results show that significant energy conservation can
be achieved when the furnace operation is optimized by using the developed soft sensor
model
Acknowledgement
The authors would like to thank the financial support provided by the National Nature
Science Foundation of China under Grant 61171145 Y Yao was supported by Ministry
of Science amp Technology ROC under Grant number MOST 104-2221-E-007-129
References
25
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[1] Z J Wang Q D Wu and T Y Chai Optimal-setting control for complicated
industrial processes and its application studyControl Engineering Practice vol
12 pp 65-74 2004
[2] A Steinboeck K Graichen and A Kugi Dynamic Optimization of a Slab
Reheating Furnace With Consistent Approximation of Control VariablesIEEE
Transactions on Control Systems Technology vol 19 pp 1444-1456 2011
[3] B T Zhang C Y Wang Q Qin and L Li Influence of Boiler Combustion
Adjustment on NOxEmission and Boiler EfficiencyAdvanced Materials
Research vol 732-733 pp 234-237 2013
[4] C K Yoo and IB Lee Soft Sensor and Adaptive Model-Based Dissolved
Oxygen Control for Biological Wastewater Treatment ProcessesEnvironmental
Engineering Science vol 21 pp 331-340 2004
[5] S A Bhat D N Saraf S Gupta and S K Gupta Use of Agitator Power as a
Soft Sensor for Bulk Free-Radical Polymerization of Methyl Methacrylate in
Batch ReactorsIndustrial amp Engineering Chemistry Research vol 45 pp 4243-
4255 2006
[6] K Desai Y Badhe S S Tambe and B D Kulkarni Soft-sensor development
for fed-batch bioreactors using support vector regressionBiochemical
Engineering Journal vol 27 pp 225-239 2006
[7] Y P Badhe Lonari J Tambe S S amp Kulkarni B D Improve polyethylene
process control and product qualityHydrocarbon Processing vol 86 pp 53-60
2007
26
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[8] N K Nath K Mandal A K Singh B Basu C Bhanu S Kumar et al Ladle
furnace on-line reckoner for prediction and control of steel temperature and
compositionIronmaking amp Steelmaking vol 33 pp 140-150 2006
[9] A J Yan T Y Chai F H Wu and P Wang Hybrid intelligent control of
combustion process for ore-roasting furnaceJournal of Control Theory and
Applications vol 6 pp 80-85 2008
[10] J Li W M Zhong H Cheng X D Kong and F Qian A data-driven soft
sensor modeling for furnace temperature of Opposed Multi-Burner gasifier pp
705-710 2011
[11] Y H Yang Y H Liu X Z Liu and S K Qin Billet temperature soft sensor
model of reheating furnace based on RVM method pp 4003-4006 2011
[12] J H Wang C Wang X F Zhu and X K Fang Application of soft sensor in
welding seam tracking prediction based on LSSVM and PSO with compression
factor pp 2441-2446 2013
[13] L Balbis J Balderud and M J Grimble Nonlinear predictive control of steel
slab reheating furnace pp 1679-1684 2008
[14] A Steinboeck D Wild T Kiefer and A Kugi A mathematical model of a slab
reheating furnace with radiative heat transfer and non-participating gaseous
media International Journal Of Heat And Mass Transfer vol 53 pp 5933-
5946 Dec 2010
[15] A Steinboeck D Wild and A Kugi Nonlinear model predictive control of a
continuous slab reheating furnace Control Engineering Practice vol 21 pp
495-508 2013
27
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[16] C Zhang T Ishii and S Sugiyama Numerical Modeling Of the Thermal
Performance Of Regenerative Slab Reheat Furnaces Numerical Heat Transfer
Part A Applications vol 32 pp 613-631 1997
[17] J G Kim and K Y Huh Prediction of Transient Slab Temperature Distribution
in the Re-heating Furnace of a Walking-beam Type for Rolling of Steel Slabs
ISIJ International vol 40 pp 1115-1123 2000
[18] J G Kim K Y and H I T K Three-Dimensional Analysis Of the Walking-
Beam-Type Slab Reheating Furnace In Hot Strip Mills Numerical Heat
Transfer Part A Applications vol 38 pp 589-609 2000
[19] CT Hsieh MJ Huang ST Lee and CH Wang Numerical Modeling of a
Walking-Beam-Type Slab Reheating Furnace Numerical Heat Transfer Part A
Applications vol 53 pp 966-981 2008
[20] MJ Huang CT Hsieh ST Lee and CH Wang A Coupled Numerical Study
of Slab Temperature and Gas Temperature in the Walking-Beam-Type Slab
Reheating Furnace Numerical Heat Transfer Part A Applications vol 54 pp
625-646 2008
[21] Z Song and A Kusiak Constraint-Based Control of Boiler Efficiency A Data-
Mining Approach IEEE Transactions on Industrial Informatics vol 3 pp 73-
83 2007
[22] J Q Li J J Gu and C L Niu The Operation Optimization based on
Correlation Analysis of Operation Parameters in Power Plant pp 138-141
2008
28
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[23] J G Wang S S Shieh S S Jang D S H Wong and C W Wu A two-tier
approach to the data-driven modeling on thermal efficiency of a BFGcoal co-
firing boiler Fuel vol 111 pp 528-534 Sep 2013
[24] L Breiman Better Subset Regression Using the Nonnegative Garrote
Technometrics vol 37 pp 373-384 1995
[25] R A Fisher The use of multiple measurements in taxonomic problems
AnnHum Genet vol 7 pp 179-188 1936
[26] J Liu Developing a soft sensor based on sparse partial least squares with
variable selection Journal of Process Control vol 24 pp 1046-1056 2014
[27] K Sun J Liu JL Kang SS Jang D SH Wong and DS Chen
Development of a variable selection method for soft sensor using artificial
neural network and nonnegative garrote Journal of Process Control vol 24 pp
1068-1075 2014
29
439
440
441
442
443
444
445
446
447
448
449
450
451
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
Table 3 Statistics of optimization at boundary conditions
Total
Amount 0 1374 0 16090 17464
Percentag
e 0 687 0 8045
8732
Total
Amount 2 220 2158 156 2536
Percentag
e 001 110 1079 078
1268
Table 4 Statistics for different optimization operations
Adjustment
Gdarr
Guarr TotalAGuarr AGdarr
Tdarr Tuarr Tdarr Tuarr
Amount 9548 3744 6328 380 0 20000
Percentage 4774 1872 3164 190 0 100
The statistics for different optimization operations is performed and shown in Table
4 It can be seen that all the adjusted operations result in the reduction of gas
consumption In over half the cases (4774 with decreased temperature and 1872
with increased temperature) AG is adjusted to higher levels Meanwhile a number of
the adjustments (3164) lower the temperature by reducing the G and AG at the same
time This result indicates that the original temperature cannot be adjusted to the target
by only reducing G in the constraint conditions The remaining cases (19) are
24
343
344
345
346
347
348
349
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
relatively special in which an excess of air is supplied in the original operation and the
adjusted operation thereby increasing the temperature by a smaller AG with less gas
consumption
6 Conclusion
Based on the actual operation data this paper aims to explore the improvement of the
combustion efficiency and the room for energy conservation Correlation analysis and
LDA show that it is of great significance to introduce two derived ratio variables which
are the AG ratio and the TG ratio A type of combustion efficiency utilizing an on-line
soft sensor is put forward by employing a NNG variable selection algorithm which
provides a good solution to the combustion efficiency real-time measurement problem
of a reheating furnace The implementation of the model-based optimization is studied
based on the actual operational data Detailed analysis for the optimization results is
given for the different cases The results show that significant energy conservation can
be achieved when the furnace operation is optimized by using the developed soft sensor
model
Acknowledgement
The authors would like to thank the financial support provided by the National Nature
Science Foundation of China under Grant 61171145 Y Yao was supported by Ministry
of Science amp Technology ROC under Grant number MOST 104-2221-E-007-129
References
25
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[1] Z J Wang Q D Wu and T Y Chai Optimal-setting control for complicated
industrial processes and its application studyControl Engineering Practice vol
12 pp 65-74 2004
[2] A Steinboeck K Graichen and A Kugi Dynamic Optimization of a Slab
Reheating Furnace With Consistent Approximation of Control VariablesIEEE
Transactions on Control Systems Technology vol 19 pp 1444-1456 2011
[3] B T Zhang C Y Wang Q Qin and L Li Influence of Boiler Combustion
Adjustment on NOxEmission and Boiler EfficiencyAdvanced Materials
Research vol 732-733 pp 234-237 2013
[4] C K Yoo and IB Lee Soft Sensor and Adaptive Model-Based Dissolved
Oxygen Control for Biological Wastewater Treatment ProcessesEnvironmental
Engineering Science vol 21 pp 331-340 2004
[5] S A Bhat D N Saraf S Gupta and S K Gupta Use of Agitator Power as a
Soft Sensor for Bulk Free-Radical Polymerization of Methyl Methacrylate in
Batch ReactorsIndustrial amp Engineering Chemistry Research vol 45 pp 4243-
4255 2006
[6] K Desai Y Badhe S S Tambe and B D Kulkarni Soft-sensor development
for fed-batch bioreactors using support vector regressionBiochemical
Engineering Journal vol 27 pp 225-239 2006
[7] Y P Badhe Lonari J Tambe S S amp Kulkarni B D Improve polyethylene
process control and product qualityHydrocarbon Processing vol 86 pp 53-60
2007
26
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[8] N K Nath K Mandal A K Singh B Basu C Bhanu S Kumar et al Ladle
furnace on-line reckoner for prediction and control of steel temperature and
compositionIronmaking amp Steelmaking vol 33 pp 140-150 2006
[9] A J Yan T Y Chai F H Wu and P Wang Hybrid intelligent control of
combustion process for ore-roasting furnaceJournal of Control Theory and
Applications vol 6 pp 80-85 2008
[10] J Li W M Zhong H Cheng X D Kong and F Qian A data-driven soft
sensor modeling for furnace temperature of Opposed Multi-Burner gasifier pp
705-710 2011
[11] Y H Yang Y H Liu X Z Liu and S K Qin Billet temperature soft sensor
model of reheating furnace based on RVM method pp 4003-4006 2011
[12] J H Wang C Wang X F Zhu and X K Fang Application of soft sensor in
welding seam tracking prediction based on LSSVM and PSO with compression
factor pp 2441-2446 2013
[13] L Balbis J Balderud and M J Grimble Nonlinear predictive control of steel
slab reheating furnace pp 1679-1684 2008
[14] A Steinboeck D Wild T Kiefer and A Kugi A mathematical model of a slab
reheating furnace with radiative heat transfer and non-participating gaseous
media International Journal Of Heat And Mass Transfer vol 53 pp 5933-
5946 Dec 2010
[15] A Steinboeck D Wild and A Kugi Nonlinear model predictive control of a
continuous slab reheating furnace Control Engineering Practice vol 21 pp
495-508 2013
27
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[16] C Zhang T Ishii and S Sugiyama Numerical Modeling Of the Thermal
Performance Of Regenerative Slab Reheat Furnaces Numerical Heat Transfer
Part A Applications vol 32 pp 613-631 1997
[17] J G Kim and K Y Huh Prediction of Transient Slab Temperature Distribution
in the Re-heating Furnace of a Walking-beam Type for Rolling of Steel Slabs
ISIJ International vol 40 pp 1115-1123 2000
[18] J G Kim K Y and H I T K Three-Dimensional Analysis Of the Walking-
Beam-Type Slab Reheating Furnace In Hot Strip Mills Numerical Heat
Transfer Part A Applications vol 38 pp 589-609 2000
[19] CT Hsieh MJ Huang ST Lee and CH Wang Numerical Modeling of a
Walking-Beam-Type Slab Reheating Furnace Numerical Heat Transfer Part A
Applications vol 53 pp 966-981 2008
[20] MJ Huang CT Hsieh ST Lee and CH Wang A Coupled Numerical Study
of Slab Temperature and Gas Temperature in the Walking-Beam-Type Slab
Reheating Furnace Numerical Heat Transfer Part A Applications vol 54 pp
625-646 2008
[21] Z Song and A Kusiak Constraint-Based Control of Boiler Efficiency A Data-
Mining Approach IEEE Transactions on Industrial Informatics vol 3 pp 73-
83 2007
[22] J Q Li J J Gu and C L Niu The Operation Optimization based on
Correlation Analysis of Operation Parameters in Power Plant pp 138-141
2008
28
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[23] J G Wang S S Shieh S S Jang D S H Wong and C W Wu A two-tier
approach to the data-driven modeling on thermal efficiency of a BFGcoal co-
firing boiler Fuel vol 111 pp 528-534 Sep 2013
[24] L Breiman Better Subset Regression Using the Nonnegative Garrote
Technometrics vol 37 pp 373-384 1995
[25] R A Fisher The use of multiple measurements in taxonomic problems
AnnHum Genet vol 7 pp 179-188 1936
[26] J Liu Developing a soft sensor based on sparse partial least squares with
variable selection Journal of Process Control vol 24 pp 1046-1056 2014
[27] K Sun J Liu JL Kang SS Jang D SH Wong and DS Chen
Development of a variable selection method for soft sensor using artificial
neural network and nonnegative garrote Journal of Process Control vol 24 pp
1068-1075 2014
29
439
440
441
442
443
444
445
446
447
448
449
450
451
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
relatively special in which an excess of air is supplied in the original operation and the
adjusted operation thereby increasing the temperature by a smaller AG with less gas
consumption
6 Conclusion
Based on the actual operation data this paper aims to explore the improvement of the
combustion efficiency and the room for energy conservation Correlation analysis and
LDA show that it is of great significance to introduce two derived ratio variables which
are the AG ratio and the TG ratio A type of combustion efficiency utilizing an on-line
soft sensor is put forward by employing a NNG variable selection algorithm which
provides a good solution to the combustion efficiency real-time measurement problem
of a reheating furnace The implementation of the model-based optimization is studied
based on the actual operational data Detailed analysis for the optimization results is
given for the different cases The results show that significant energy conservation can
be achieved when the furnace operation is optimized by using the developed soft sensor
model
Acknowledgement
The authors would like to thank the financial support provided by the National Nature
Science Foundation of China under Grant 61171145 Y Yao was supported by Ministry
of Science amp Technology ROC under Grant number MOST 104-2221-E-007-129
References
25
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[1] Z J Wang Q D Wu and T Y Chai Optimal-setting control for complicated
industrial processes and its application studyControl Engineering Practice vol
12 pp 65-74 2004
[2] A Steinboeck K Graichen and A Kugi Dynamic Optimization of a Slab
Reheating Furnace With Consistent Approximation of Control VariablesIEEE
Transactions on Control Systems Technology vol 19 pp 1444-1456 2011
[3] B T Zhang C Y Wang Q Qin and L Li Influence of Boiler Combustion
Adjustment on NOxEmission and Boiler EfficiencyAdvanced Materials
Research vol 732-733 pp 234-237 2013
[4] C K Yoo and IB Lee Soft Sensor and Adaptive Model-Based Dissolved
Oxygen Control for Biological Wastewater Treatment ProcessesEnvironmental
Engineering Science vol 21 pp 331-340 2004
[5] S A Bhat D N Saraf S Gupta and S K Gupta Use of Agitator Power as a
Soft Sensor for Bulk Free-Radical Polymerization of Methyl Methacrylate in
Batch ReactorsIndustrial amp Engineering Chemistry Research vol 45 pp 4243-
4255 2006
[6] K Desai Y Badhe S S Tambe and B D Kulkarni Soft-sensor development
for fed-batch bioreactors using support vector regressionBiochemical
Engineering Journal vol 27 pp 225-239 2006
[7] Y P Badhe Lonari J Tambe S S amp Kulkarni B D Improve polyethylene
process control and product qualityHydrocarbon Processing vol 86 pp 53-60
2007
26
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[8] N K Nath K Mandal A K Singh B Basu C Bhanu S Kumar et al Ladle
furnace on-line reckoner for prediction and control of steel temperature and
compositionIronmaking amp Steelmaking vol 33 pp 140-150 2006
[9] A J Yan T Y Chai F H Wu and P Wang Hybrid intelligent control of
combustion process for ore-roasting furnaceJournal of Control Theory and
Applications vol 6 pp 80-85 2008
[10] J Li W M Zhong H Cheng X D Kong and F Qian A data-driven soft
sensor modeling for furnace temperature of Opposed Multi-Burner gasifier pp
705-710 2011
[11] Y H Yang Y H Liu X Z Liu and S K Qin Billet temperature soft sensor
model of reheating furnace based on RVM method pp 4003-4006 2011
[12] J H Wang C Wang X F Zhu and X K Fang Application of soft sensor in
welding seam tracking prediction based on LSSVM and PSO with compression
factor pp 2441-2446 2013
[13] L Balbis J Balderud and M J Grimble Nonlinear predictive control of steel
slab reheating furnace pp 1679-1684 2008
[14] A Steinboeck D Wild T Kiefer and A Kugi A mathematical model of a slab
reheating furnace with radiative heat transfer and non-participating gaseous
media International Journal Of Heat And Mass Transfer vol 53 pp 5933-
5946 Dec 2010
[15] A Steinboeck D Wild and A Kugi Nonlinear model predictive control of a
continuous slab reheating furnace Control Engineering Practice vol 21 pp
495-508 2013
27
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[16] C Zhang T Ishii and S Sugiyama Numerical Modeling Of the Thermal
Performance Of Regenerative Slab Reheat Furnaces Numerical Heat Transfer
Part A Applications vol 32 pp 613-631 1997
[17] J G Kim and K Y Huh Prediction of Transient Slab Temperature Distribution
in the Re-heating Furnace of a Walking-beam Type for Rolling of Steel Slabs
ISIJ International vol 40 pp 1115-1123 2000
[18] J G Kim K Y and H I T K Three-Dimensional Analysis Of the Walking-
Beam-Type Slab Reheating Furnace In Hot Strip Mills Numerical Heat
Transfer Part A Applications vol 38 pp 589-609 2000
[19] CT Hsieh MJ Huang ST Lee and CH Wang Numerical Modeling of a
Walking-Beam-Type Slab Reheating Furnace Numerical Heat Transfer Part A
Applications vol 53 pp 966-981 2008
[20] MJ Huang CT Hsieh ST Lee and CH Wang A Coupled Numerical Study
of Slab Temperature and Gas Temperature in the Walking-Beam-Type Slab
Reheating Furnace Numerical Heat Transfer Part A Applications vol 54 pp
625-646 2008
[21] Z Song and A Kusiak Constraint-Based Control of Boiler Efficiency A Data-
Mining Approach IEEE Transactions on Industrial Informatics vol 3 pp 73-
83 2007
[22] J Q Li J J Gu and C L Niu The Operation Optimization based on
Correlation Analysis of Operation Parameters in Power Plant pp 138-141
2008
28
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[23] J G Wang S S Shieh S S Jang D S H Wong and C W Wu A two-tier
approach to the data-driven modeling on thermal efficiency of a BFGcoal co-
firing boiler Fuel vol 111 pp 528-534 Sep 2013
[24] L Breiman Better Subset Regression Using the Nonnegative Garrote
Technometrics vol 37 pp 373-384 1995
[25] R A Fisher The use of multiple measurements in taxonomic problems
AnnHum Genet vol 7 pp 179-188 1936
[26] J Liu Developing a soft sensor based on sparse partial least squares with
variable selection Journal of Process Control vol 24 pp 1046-1056 2014
[27] K Sun J Liu JL Kang SS Jang D SH Wong and DS Chen
Development of a variable selection method for soft sensor using artificial
neural network and nonnegative garrote Journal of Process Control vol 24 pp
1068-1075 2014
29
439
440
441
442
443
444
445
446
447
448
449
450
451
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[1] Z J Wang Q D Wu and T Y Chai Optimal-setting control for complicated
industrial processes and its application studyControl Engineering Practice vol
12 pp 65-74 2004
[2] A Steinboeck K Graichen and A Kugi Dynamic Optimization of a Slab
Reheating Furnace With Consistent Approximation of Control VariablesIEEE
Transactions on Control Systems Technology vol 19 pp 1444-1456 2011
[3] B T Zhang C Y Wang Q Qin and L Li Influence of Boiler Combustion
Adjustment on NOxEmission and Boiler EfficiencyAdvanced Materials
Research vol 732-733 pp 234-237 2013
[4] C K Yoo and IB Lee Soft Sensor and Adaptive Model-Based Dissolved
Oxygen Control for Biological Wastewater Treatment ProcessesEnvironmental
Engineering Science vol 21 pp 331-340 2004
[5] S A Bhat D N Saraf S Gupta and S K Gupta Use of Agitator Power as a
Soft Sensor for Bulk Free-Radical Polymerization of Methyl Methacrylate in
Batch ReactorsIndustrial amp Engineering Chemistry Research vol 45 pp 4243-
4255 2006
[6] K Desai Y Badhe S S Tambe and B D Kulkarni Soft-sensor development
for fed-batch bioreactors using support vector regressionBiochemical
Engineering Journal vol 27 pp 225-239 2006
[7] Y P Badhe Lonari J Tambe S S amp Kulkarni B D Improve polyethylene
process control and product qualityHydrocarbon Processing vol 86 pp 53-60
2007
26
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[8] N K Nath K Mandal A K Singh B Basu C Bhanu S Kumar et al Ladle
furnace on-line reckoner for prediction and control of steel temperature and
compositionIronmaking amp Steelmaking vol 33 pp 140-150 2006
[9] A J Yan T Y Chai F H Wu and P Wang Hybrid intelligent control of
combustion process for ore-roasting furnaceJournal of Control Theory and
Applications vol 6 pp 80-85 2008
[10] J Li W M Zhong H Cheng X D Kong and F Qian A data-driven soft
sensor modeling for furnace temperature of Opposed Multi-Burner gasifier pp
705-710 2011
[11] Y H Yang Y H Liu X Z Liu and S K Qin Billet temperature soft sensor
model of reheating furnace based on RVM method pp 4003-4006 2011
[12] J H Wang C Wang X F Zhu and X K Fang Application of soft sensor in
welding seam tracking prediction based on LSSVM and PSO with compression
factor pp 2441-2446 2013
[13] L Balbis J Balderud and M J Grimble Nonlinear predictive control of steel
slab reheating furnace pp 1679-1684 2008
[14] A Steinboeck D Wild T Kiefer and A Kugi A mathematical model of a slab
reheating furnace with radiative heat transfer and non-participating gaseous
media International Journal Of Heat And Mass Transfer vol 53 pp 5933-
5946 Dec 2010
[15] A Steinboeck D Wild and A Kugi Nonlinear model predictive control of a
continuous slab reheating furnace Control Engineering Practice vol 21 pp
495-508 2013
27
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[16] C Zhang T Ishii and S Sugiyama Numerical Modeling Of the Thermal
Performance Of Regenerative Slab Reheat Furnaces Numerical Heat Transfer
Part A Applications vol 32 pp 613-631 1997
[17] J G Kim and K Y Huh Prediction of Transient Slab Temperature Distribution
in the Re-heating Furnace of a Walking-beam Type for Rolling of Steel Slabs
ISIJ International vol 40 pp 1115-1123 2000
[18] J G Kim K Y and H I T K Three-Dimensional Analysis Of the Walking-
Beam-Type Slab Reheating Furnace In Hot Strip Mills Numerical Heat
Transfer Part A Applications vol 38 pp 589-609 2000
[19] CT Hsieh MJ Huang ST Lee and CH Wang Numerical Modeling of a
Walking-Beam-Type Slab Reheating Furnace Numerical Heat Transfer Part A
Applications vol 53 pp 966-981 2008
[20] MJ Huang CT Hsieh ST Lee and CH Wang A Coupled Numerical Study
of Slab Temperature and Gas Temperature in the Walking-Beam-Type Slab
Reheating Furnace Numerical Heat Transfer Part A Applications vol 54 pp
625-646 2008
[21] Z Song and A Kusiak Constraint-Based Control of Boiler Efficiency A Data-
Mining Approach IEEE Transactions on Industrial Informatics vol 3 pp 73-
83 2007
[22] J Q Li J J Gu and C L Niu The Operation Optimization based on
Correlation Analysis of Operation Parameters in Power Plant pp 138-141
2008
28
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[23] J G Wang S S Shieh S S Jang D S H Wong and C W Wu A two-tier
approach to the data-driven modeling on thermal efficiency of a BFGcoal co-
firing boiler Fuel vol 111 pp 528-534 Sep 2013
[24] L Breiman Better Subset Regression Using the Nonnegative Garrote
Technometrics vol 37 pp 373-384 1995
[25] R A Fisher The use of multiple measurements in taxonomic problems
AnnHum Genet vol 7 pp 179-188 1936
[26] J Liu Developing a soft sensor based on sparse partial least squares with
variable selection Journal of Process Control vol 24 pp 1046-1056 2014
[27] K Sun J Liu JL Kang SS Jang D SH Wong and DS Chen
Development of a variable selection method for soft sensor using artificial
neural network and nonnegative garrote Journal of Process Control vol 24 pp
1068-1075 2014
29
439
440
441
442
443
444
445
446
447
448
449
450
451
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[8] N K Nath K Mandal A K Singh B Basu C Bhanu S Kumar et al Ladle
furnace on-line reckoner for prediction and control of steel temperature and
compositionIronmaking amp Steelmaking vol 33 pp 140-150 2006
[9] A J Yan T Y Chai F H Wu and P Wang Hybrid intelligent control of
combustion process for ore-roasting furnaceJournal of Control Theory and
Applications vol 6 pp 80-85 2008
[10] J Li W M Zhong H Cheng X D Kong and F Qian A data-driven soft
sensor modeling for furnace temperature of Opposed Multi-Burner gasifier pp
705-710 2011
[11] Y H Yang Y H Liu X Z Liu and S K Qin Billet temperature soft sensor
model of reheating furnace based on RVM method pp 4003-4006 2011
[12] J H Wang C Wang X F Zhu and X K Fang Application of soft sensor in
welding seam tracking prediction based on LSSVM and PSO with compression
factor pp 2441-2446 2013
[13] L Balbis J Balderud and M J Grimble Nonlinear predictive control of steel
slab reheating furnace pp 1679-1684 2008
[14] A Steinboeck D Wild T Kiefer and A Kugi A mathematical model of a slab
reheating furnace with radiative heat transfer and non-participating gaseous
media International Journal Of Heat And Mass Transfer vol 53 pp 5933-
5946 Dec 2010
[15] A Steinboeck D Wild and A Kugi Nonlinear model predictive control of a
continuous slab reheating furnace Control Engineering Practice vol 21 pp
495-508 2013
27
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[16] C Zhang T Ishii and S Sugiyama Numerical Modeling Of the Thermal
Performance Of Regenerative Slab Reheat Furnaces Numerical Heat Transfer
Part A Applications vol 32 pp 613-631 1997
[17] J G Kim and K Y Huh Prediction of Transient Slab Temperature Distribution
in the Re-heating Furnace of a Walking-beam Type for Rolling of Steel Slabs
ISIJ International vol 40 pp 1115-1123 2000
[18] J G Kim K Y and H I T K Three-Dimensional Analysis Of the Walking-
Beam-Type Slab Reheating Furnace In Hot Strip Mills Numerical Heat
Transfer Part A Applications vol 38 pp 589-609 2000
[19] CT Hsieh MJ Huang ST Lee and CH Wang Numerical Modeling of a
Walking-Beam-Type Slab Reheating Furnace Numerical Heat Transfer Part A
Applications vol 53 pp 966-981 2008
[20] MJ Huang CT Hsieh ST Lee and CH Wang A Coupled Numerical Study
of Slab Temperature and Gas Temperature in the Walking-Beam-Type Slab
Reheating Furnace Numerical Heat Transfer Part A Applications vol 54 pp
625-646 2008
[21] Z Song and A Kusiak Constraint-Based Control of Boiler Efficiency A Data-
Mining Approach IEEE Transactions on Industrial Informatics vol 3 pp 73-
83 2007
[22] J Q Li J J Gu and C L Niu The Operation Optimization based on
Correlation Analysis of Operation Parameters in Power Plant pp 138-141
2008
28
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[23] J G Wang S S Shieh S S Jang D S H Wong and C W Wu A two-tier
approach to the data-driven modeling on thermal efficiency of a BFGcoal co-
firing boiler Fuel vol 111 pp 528-534 Sep 2013
[24] L Breiman Better Subset Regression Using the Nonnegative Garrote
Technometrics vol 37 pp 373-384 1995
[25] R A Fisher The use of multiple measurements in taxonomic problems
AnnHum Genet vol 7 pp 179-188 1936
[26] J Liu Developing a soft sensor based on sparse partial least squares with
variable selection Journal of Process Control vol 24 pp 1046-1056 2014
[27] K Sun J Liu JL Kang SS Jang D SH Wong and DS Chen
Development of a variable selection method for soft sensor using artificial
neural network and nonnegative garrote Journal of Process Control vol 24 pp
1068-1075 2014
29
439
440
441
442
443
444
445
446
447
448
449
450
451
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[16] C Zhang T Ishii and S Sugiyama Numerical Modeling Of the Thermal
Performance Of Regenerative Slab Reheat Furnaces Numerical Heat Transfer
Part A Applications vol 32 pp 613-631 1997
[17] J G Kim and K Y Huh Prediction of Transient Slab Temperature Distribution
in the Re-heating Furnace of a Walking-beam Type for Rolling of Steel Slabs
ISIJ International vol 40 pp 1115-1123 2000
[18] J G Kim K Y and H I T K Three-Dimensional Analysis Of the Walking-
Beam-Type Slab Reheating Furnace In Hot Strip Mills Numerical Heat
Transfer Part A Applications vol 38 pp 589-609 2000
[19] CT Hsieh MJ Huang ST Lee and CH Wang Numerical Modeling of a
Walking-Beam-Type Slab Reheating Furnace Numerical Heat Transfer Part A
Applications vol 53 pp 966-981 2008
[20] MJ Huang CT Hsieh ST Lee and CH Wang A Coupled Numerical Study
of Slab Temperature and Gas Temperature in the Walking-Beam-Type Slab
Reheating Furnace Numerical Heat Transfer Part A Applications vol 54 pp
625-646 2008
[21] Z Song and A Kusiak Constraint-Based Control of Boiler Efficiency A Data-
Mining Approach IEEE Transactions on Industrial Informatics vol 3 pp 73-
83 2007
[22] J Q Li J J Gu and C L Niu The Operation Optimization based on
Correlation Analysis of Operation Parameters in Power Plant pp 138-141
2008
28
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[23] J G Wang S S Shieh S S Jang D S H Wong and C W Wu A two-tier
approach to the data-driven modeling on thermal efficiency of a BFGcoal co-
firing boiler Fuel vol 111 pp 528-534 Sep 2013
[24] L Breiman Better Subset Regression Using the Nonnegative Garrote
Technometrics vol 37 pp 373-384 1995
[25] R A Fisher The use of multiple measurements in taxonomic problems
AnnHum Genet vol 7 pp 179-188 1936
[26] J Liu Developing a soft sensor based on sparse partial least squares with
variable selection Journal of Process Control vol 24 pp 1046-1056 2014
[27] K Sun J Liu JL Kang SS Jang D SH Wong and DS Chen
Development of a variable selection method for soft sensor using artificial
neural network and nonnegative garrote Journal of Process Control vol 24 pp
1068-1075 2014
29
439
440
441
442
443
444
445
446
447
448
449
450
451
A Soft-Sensing Method for Optimization of Combustion Efficiency of Reheating Furnaces
[23] J G Wang S S Shieh S S Jang D S H Wong and C W Wu A two-tier
approach to the data-driven modeling on thermal efficiency of a BFGcoal co-
firing boiler Fuel vol 111 pp 528-534 Sep 2013
[24] L Breiman Better Subset Regression Using the Nonnegative Garrote
Technometrics vol 37 pp 373-384 1995
[25] R A Fisher The use of multiple measurements in taxonomic problems
AnnHum Genet vol 7 pp 179-188 1936
[26] J Liu Developing a soft sensor based on sparse partial least squares with
variable selection Journal of Process Control vol 24 pp 1046-1056 2014
[27] K Sun J Liu JL Kang SS Jang D SH Wong and DS Chen
Development of a variable selection method for soft sensor using artificial
neural network and nonnegative garrote Journal of Process Control vol 24 pp
1068-1075 2014
29
439
440
441
442
443
444
445
446
447
448
449
450
451