modelling and analysis of future data rich and uncertain
TRANSCRIPT
Modelling and analysis of future data rich and uncertain power
systems Prof Jovica V. Milanović
Dipl.Ing., MSc, PhD, DSc, CEng, FIET, FIEEE, Foreign member of SAES
Deputy Head of School & Head of Electrical Energy and Power Systems Group
Manchester, M13 9PL, United Kingdom
18st December 2015, Sveučilište Josipa Jurja Strossmayera, Osijek, & Faculty of Electrical Engineering and Computing, University of Zagreb, Zagreb, Croatia
School of Electrical & Electronic Engineering
1/35
Overview of the presentation
• Structure, operation and key attributes of Future netwroks
• Modelling and control challenges • Examples of current thinking and potential
solutions • Summary
2/35
Structure, operation and key attributes
of Future networks
3/35
Evolving Power Network - Future
• Liberalised market • Increased cross-boarder
bulk power transfers to facilitate effectiveness of market mechanisms
B5
B6
B1B2B3B4
B8 B9
B7
B10
B11
B12B13
B14B19B18
B17
B16
B15
B20
B21
B22
B23
B24
B25
B26
B29
B27
B28
B30
B31 B32
B33
B34
B35
B36 B37
B38B39
B40
B49B48
B47
B46
B45
B44
B43B42
B41B56
B57
B55
B54
B53
B52
B51
B50
P
P
P
P
P
P
P P
P
P
P
• Increased presence of static and active shunt and series compensation • Increased deployment of FACTS devices in general
• Increased use of HVDC lines of both, LCC and predominantly VSC MMC technology (in meshed networks and as a super grid)
4/35
Evolving Power Network - Future B5
B6
B1B2B3B4
B8 B9
B7
B10
B11
B12B13
B14B19B18
B17
B16
B15
B20
B21
B22
B23
B24
B25
B26
B29
B27
B28
B30
B31 B32
B33
B34
B35
B36 B37
B38B39
B40
B49B48
B47
B46
B45
B44
B43B42
B41B56
B57
B55
B54
B53
B52
B51
B50
P
P
P
P
P
P
P P
P
P
P
• Proliferation of non- conventional renewable generation – largely stochastic and intermittent (wind, PV, marine) at all levels and of various sizes
0 100 200 300 4000
0.002
0.004
0.006
0.008
0.01
0.012
0.014
0.016
Wind Direction (degrees)
Prob
abilit
y
Year 2000
0 5 10 15 20 25 300
1
2
3
4
5
6
7
8x 10-3
Wind speed (m/s)
Pro
babi
lity
Year 2000
• Large on-shore and off- shore wind farms
5/35
Evolving Power Network - Future B5
B6
B1B2B3B4
B8 B9
B7
B10
B11
B12B13
B14B19B18
B17
B16
B15
B20
B21
B22
B23
B24
B25
B26
B29
B27
B28
B30
B31 B32
B33
B34
B35
B36 B37
B38B39
B40
B49B48
B47
B46
B45
B44
B43B42
B41B56
B57
B55
B54
B53
B52
B51
B50
P
P
P
P
P
P
P P
P
P
P
T&
D
G
G
HV
Cell
G
MV
Cell
MV
Cell
- -
- M
V C
ell
G
G
G
G
P
P
P
P
P
P
P
P P
P
P
P
• Bi-directional energy flow
• Different energy carriers
• Small scale (widely dispersed) technologies in DN • Active distribution networks • New types of loads within customer premises (PE, LED)
• Electric vehicles (spatial and temporal uncertainty) • Integrated “intelligent” PE devices • Integrated ICT & storage
6/35
Communications in power systems
CC – Control Centre SS – Substation VPP – Virtual Power Plant Reg - Regional
Adopted from presentations at • Bi-directional info flow
Power & Information Flow
7/35
Uncertainties • There are generally two forms of uncertainty
– Aleatory uncertainty (irreducable uncertainty and variability) represents the inherent random behaviour of a system Commonly modelled by probabilistic distribution functions and
propagated by probability based approaches (sampling, analytical methods, probabilistic chaos expansion)
– Epistemic uncertainty (reducible uncertainty and state of knowledge uncertainty) models the uncertainty in parameter estimation due to data shortages or model simplification To avoid the subjective assignment of uncertainty due to data
deficiency different representation and quantification methods can be used, e.g., Fuzzy sets, Possibility theory, Evidence theory, etc.
Representation depends on amount and accuracy of available data. With sufficient data it can be modelled by specific probability density functions, otherwise uniform or normal distribution is used within specific range
8/35
Sources of uncertainties • Network
– topology, parameters & settings (e.g., tap settings, temperature dependent line ratings) – observability & controlability
• Generation – pattern (size, output of generators, types and location of generators, i.e., conventional,
renewable, storage) – parameters (conventional and renewable generation and storage)
• Load (time and spatial variation in load, load composition (mix), models and parameters)
• Controls – parameters of generator controllers (AVRs, Governors, PSSs, PE interface), network
controllers (secondary voltage controller), FACTS devices and HVDC line controllers
• Contractual power flow (consequence of different market mechanisms and price)
• Faults (type, location, duration, frequency, distribution, impedance)
• ICT related uncertainties (noise, measurement errors, time delays, loss of signals, bandwidth)
• Weather/climate related uncertainties (wind speed, wind direction, temperature, solar irradiation, tidal/wave conditions)
9/35
Sources of “Big Data” in Power System • SCADA (Supervisory Control And Data Acquisition) systems (1-5min updates)
• WAMS (Wide Area Monitoring Systems) (50-60 updates per second) (by the end of 2013 China State Grid Corporation installed 2027 PMUs at V>110kV)
• Advanced metering devices, IEDs (“Intelligent/Smart” meters with some degree of monitoring capabilities - even in low cost smart meters for hundreds of thousands household customers)
• Bi-directional communication enabled mobile (e.g. EVs) and stationary devices (e.g. domestic appliances)
• Conventional PQ monitoring and energy/power metering at load and generation (including RES) buses (3-5sec updates)
• Historical monitoring and incident/control reporting data • Customer surveys • Internet resources (related to network and generation performance,
customer behaviour and preferences, user and generation groups /associations)
Integrated monitoring system reporting to some other hierarchical systems such as distribution or transmission management system
10/35
Data Analytics
• Analytics - discovery and communication of meaningful patterns in data. – Analysing structured information unstructured big data (e.g.,, video, images and content from sensors, social
and mobile devices) – to support continuous data load, run multiple concurrent queries, – deliver analytics in real-time supported by massive I/O bandwidth.
11/35
Depending on the application the data may have different characteristics: • Multidimensional • Multiscale • Spatially distributed • Time series • Event triggered • Faulty (sensing and/or communication problems) • Incomplete
Data...Questions that need to be asked? • What do we want to learn from collected data and why (for what purpose)?
• How will this facilitate “better” network control and operation? • Do we know what data we already have and what to do with them? • Do we already have all the data that we need? • What “additional data” do we need?
• Where and how should we collect the data? Do we have a choice? Can we collect (privacy issues) all the data that we need?
• What should be the “sampling” rate and the “quality “ of additional data? • Do we know what techniques to use to efficiently and effectively process new
(and old) data? Have we answered any of these questions yet,
or at least have we asked them?
If not, isn’t it the time to start doing so?
12/35
• Trans national/ continental • Tailored for individuals • Tolerant to behavioural patterns • Trade-off analysis
• Signal & data processing • Secure (Cyber &
energy) • State estimation • Stochastic generation & load • Self*
• Monitoring (WAMS, SCADA, IED) • Management of uncertainties • Multi energy carriers • Multivariable optimisation/control • Mixed generation • Market enabling
• Adaptive, hierarchical, distributed control • Augmented power transfers (FACTS/HVDC) • Affordable solutions • Aggregators (for service provision) • Ancillary services devolved
• Robust • Responsive & Involved demand • Reliable • RES friendly • Real time control • Reduced uncertainty • Risk based designed & managed
So, Future Power Network will be
B5
B6
B1B2B3B4
B8 B9
B7
B10
B11
B12B13
B14B19B18
B17
B16
B15
B20
B21
B22
B23
B24
B25
B26
B29
B27
B28
B30
B31 B32
B33
B34
B35
B36 B37
B38B39
B40
B49B48
B47
B46
B45
B44
B43B42
B41B56
B57
B55
B54
B53
B52
B51
B50
P
P
P
P
P
P
P P
P
P
P
P
P
P
P
P
P
P P
P
P
P
13/35
The question therefore is
Considering evolving power system with increased uncertainties and abundance of measurement data, are the tools currently in use for system modelling and analysis adequate, and if not, how should we modify them, or what other tools should we be using?
14/35
What Modelling and Control Challenges this new “environment” will result in?
15/35
• Efficient use and reliance on existing and new global monitoring data (WAMS) for state estimation, static and dynamic equivalents and control (including real time control) – Efficient data management (signal processing, aggregation, transmission) and ICT
network reliability are essential for both static and dynamic observability – Optimal placement of monitoring devices (PMU) may not be an issue due
to perceived high deployment of those
Challenges - 1
• Modelling requirements for steady state & dynamic studies – Large interconnected networks with mixed generation, FACTS and
short/long distance bulk power transfers using HVDC lines – Clusters of RES (generation and storage) of the same or different type – LV and MV distribution network cell (DNC) with thousands of RES – Demand, including new types of energy efficient and PE controlled loads,
customer participation and behavioural patterns, EV, etc.
16/35
• Design of supplementary controllers based on WAMS to control and stabilise large system (including real-time) or parts of it (which may vary) with uncertain power transfers and load models and stochastically varying and intermittent PE connected generation, demand and storage – stochastic/probabilistic control of systems with reduced inertia
• Design of new control systems/structure (distributed or hierarchical, adaptive, close to real time) for power networks with fully integrated sensing, ICT technologies and protection systems – risk limiting control
• Modelling/analysis of efficient and effective integration of different energy carriers into self sufficient energy module/cell
Challenges - 2
Risk is an uncertain event or condition that, if it occurs, has an effect on at least one objective, it is the probability of something happening multiplied by the resulting cost or benefit if it does.
17/35
Examples of current thinking and potential solutions
18/35
1. Dynamic equivalent models of Wind farm using probabilistic clustering
2. Data mining for on-line transient stability assessment (TSA) 3. Modal estimation with Probabilistic Collocation Method
(PCM) 4. Reduction of number of uncertain parameters 5. Probabilistic demand response and composition forecasting 6. Probabilistic estimation of PQ performance of active
distribution network
6 Examples
19/35
Acknowledgement The 18 people strong team supported by £2.8M (€3.5M) in
research grants from UK Research Council, EU and UK Industry who is currently working with me on trying to solve some of the problems discussed in this presentation includes:
- 6 Postdoctoral Research Associates (Dr H. Liao, Dr P. Papadopoulos, Dr A. Adrees, Dr K. Hasan, Dr X. Tang, Dr T. Guo)
- 5 PhD students ( Mr S. Abdelrahman, Miss J.Ponocko , Mr Y.Zhou, Mr W. Zhou, Mr B. Qi) )
- 2 Advanced MSc students - 5 Final year project students
Some of them and many (100+) other researchers who worked with
me in the past contributed to different extent and at different stages of research to the results presented on the following slides
20/35
Current Research Team
21/35
7 PDRA
5 PhD 2 AMSc
Dynamic equivalent models of Wind farm using probabilistic clustering
PCCString 1String 2String 3String 4String 5String 6String 7
30kV 132 kV
3-phase faultGrid
=
1 2 3 4 5 6 7 8 9 100
0.05
0.1
0.15
0.2
0.25
0.3
0.35
No. of equivalent turbinesP
roba
bilit
y
Probability of equivalent turbines
7
6
5
4
3
2
1
14
13
12
11
10
9
8
21
20
19
18
17
16
15
28
27
26
25
24
23
22
35
34
33
32
31
30
29
42
41
40
39
38
37
36
49
48
47
46
45
44
43
15 m/s, 322o
Wind speed variation inside a wind farm at 15 m/s, 322o
Support Vector Machine
based clustering
22/35
Dynamic equivalent models of Wind farm using probabilistic clustering
0 1 2 3 4 5 6 7 8 9 100
0.2
0.4
0.6
0.8
1
Time (sec)
Act
ive
Pow
er (p
.u)
3 4
0.3
0.35
0.4
Detail modelAggregate model
0 1 2 3 4 5 6 7 8 9 10-5
0
5
10
15
20
25
30
35
40
Time (sec)
Rea
ctiv
e Po
wer
(p.u
)
2 3 40
2
4
6
8
Detail modelAggregate model
P and Q response for Detailed and Probabilistic model at wind speed = 10 m/s, wind direction = 100° In the case studied, simulation time was reduced by up to 96%.
23/35
Dynamic equivalent models of Wind farm using probabilistic clustering
24/35
M.Ali, Irinel-Sorrin Ilie, J.V.Milanović and Gianfranco Chicco,“Wind farm model aggregation using probabilistic clustering”, IEEE Transactions on Power Systems, Vol. 28, No 1, 2013, pp. 309-316
Data mining for on-line transient stability assessment (TSA)
For any type and duration of fault at any possible location and any system load level, the Decision Tree predicts the system stability status correctly using 12 input signals from PMUs (rotor angles and speeds) with over 98.5% accuracy within 10 cycles (0.2 s) after the clearance of the fault, and it is close to 100% accurate after about 1 s.
G7
24. 19.
37.
29.
15. 14.
11. 6.
5.
43.
22.
8.
9.
17.
34.
35.
32. 30.
38.
51.
33.
42.
27.
G6 G4
G5 G3 G2
G2 G1
G9
G16
G15
G14
G12G13
G11
G10
NEW ENGLAND TEST SYSTEM NEW YORK POWER SYSTEM
40.
41.
18. 50.
49.
1.
20.
2.
59.
58.
57.
56. 23.
21.
61. 28.
26.
16.
52.
3.
25.
4.
12.
13. 10.
55. 54. 7.
47. 48.
62.
31. 63.
46.
36.
64. 65.
44.
39.
45.
66.
67.
68.
60 53
95
96
97
98
99
100
10 20 30 40 50 60
Acc
urac
y of
Pre
dict
ion
(%)
Length of Rotor Angles (cycle)
Stable Unstable Overall
At 60 cycles (1 s) after the fault, only 3 out of 1883 stable cases and 2 out of 167 unstable cases are misclassified.
25/35
Data mining for on line TSA – Hierarchical Clustering & Multiclass Classification
Accuracy of prediction of both generator grouping and unsynchronized groups
Accuracy of prediction of generator grouping in case of unstable dynamic behavior.
83
85
87
89
91
93
10 20 30 40 50 60
Acc
urac
y of
Pre
dict
ion
(%)
Length of Rotor Angles (cycle)
CART C5.0
Boosting C5.0 RF
MSVM
85
87
89
91
93
10 20 30 40 50 60
Acc
urac
y of
Pre
dict
ion
(%)
Length of Rotor Angles (cycle)
Target I Target II
The aim is identify characteristic patterns of unstable dynamic behavior of a power system (and generators that lose synchronism) when the number of groups is not specified in advance
26/35
Data mining for on line TSA – Hierarchical Clustering & Multiclass Classification
27/35
Tingyan Guo and J.V.Milanović, “Probabilistic framework for assessing the accuracy of data mining tool for on-line prediction of transient stability”, IEEE Transactions on Power Systems, Vol. 29, No 1, 2014, pp. 377 – 385 Tingyan Guo and J.V.Milanović, “On-line identification of power system dynamic signature using PMU measurements and data mining”, accepted for publication in the IEEE Transactions on Power Systems, TPWRS-00940-2014 (2/06/15)
PCM can be used to quickly assess the Risk of Instability (RoI). True value = 1.91%; PCM-based value = 1.89%.
Modal estimation with Probabilistic Collocation Method (PCM)
G1
G2 G4
G3L2L1 C1 C2
1
2 4
311
6
5
7 8 9
10
25 km 10 km110 km 110 km
10 km 25 km
VSC-HVDC
True value
PCM value
28/35
Modal estimation with Probabilistic Collocation Method (PCM)
29/35
R.Preece, J.V.Milanović A.M.Almutairi and O.Marjanovic, "Probabilistic evaluation of damping controller in networks with multiple VSC-HVDC lines", IEEE Transactions on Power Systems, Vol. 28, No 1, 2013, pp. 367-376 R.Preece, N.C.Woolley and J.V.Milanović, "The probabilistic collocation method for power system damping and voltage collapse studies in the presence of uncertainties", IEEE Transactions on Power Systems, Vol. 28, No 3, 2013, pp. 2253-2262 R.Preece and J.V.Milanović, "Tuning of a damping controller for multi-terminal VSC-HVDC grids using the probabilistic collocation method", Special Issue of IEEE Transactions on Power Delivery: HVDC Systems and Technologies, Vol. 29, No 1, 2014, pp. 318-326
• Large Test System with 2 VSC HVDC lines • Fifty-one uncertain parameters reduced to eight using ranking
Method based on eigenvalue sensitivites • Completed in just 2.44% of the time required for standard
(full-simulation) technique – over forty times faster.
Modal estimation with PCM and reduction of uncertain parameters – large network
30/35
Modal estimation with PCM and reduction of uncertain parameters – large network
31/35
R.Preece, Kaijia Huang and J.V.Milanović, "Probabilistic small-disturbance stability assessment of uncertain power systems using efficient estimation methods", IEEE Transactions on Power Systems, Vol. 29, No 5, 2014, pp. 2509 - 2517 R.Preece and J.V.Milanović, “Risk-based small-disturbance security assessment of power systems", IEEE Transactions on Power Delivery, Vol. 30, No 2, 2015, pp. 590 – 598 R.Preece and J.V.Milanović, "Probabilistic risk assessment of rotor angle instability using fuzzy inference systems", accepted for publication in the IEEE Transactions on Power Systems, TPWRS-01355-2013 (14/8/14) R.Preece and J.V.Milanović, "Efficient estimation of the probability of small-disturbance instability of large uncertain power systems", accepted for publication in the IEEE Transactions on Power Systems, TPWRS-00963-2014 (23/3/15) R.Preece and J.V.Milanović, "Assessing the applicability of uncertainty importance measures for power system studies", accepted for publication in the IEEE Transactions on Power Systems, TPWRS-01564-2014 (22/06/15)
Probabilistic forecasting of demand controllability and response
Metering Data Customer Surveys General Knowledge of Demand Composition
DDLC Based on Load Classes
Participation of Load Categories in Different
Load Classes
DDLC Based on Load Categories
Data Processing with Parameter Uncertainty
Estimation of Dynamic Response of Aggregate Demand at Given Time
Dynamic Signatures/ Responses of Different Load
Categories
Field Measurement Lab Measurement Computer Simulations
Data Processing with Parameter Uncertainty
Probabilistic Dynamic Signatures of Different Load
Categories
Time Dependant Dynamic Response of Aggregate
Demand
Probabilistic DDLC
32/35
Demand composition forecasting
33/35
Forecasted demand response
Comparison of different a) P and b) Q responses at different times of day (solid line: 03:00; dashed line:
04:00; dash-dot line: 12:00; dotted line: 18:00)
0 0.5 1 1.5 2 2.5 395.5
9696.5
9797.5
9898.5
9999.5100
100.5
t(sec)
P(%
of P
at t
his
hour
)
3:00
Before ShiftingIM+ResistiveResistiveIM
35/35
Forecasted demand response
36/35
Yizheng Xu and J.V. Milanović, “Artificial intelligence based methodology for load disaggregation at bulk supply point”, IEEE Transactions on Power Systems, Vol. 30, No 2, 2015, pp. 795 – 803 J.V. Milanović and Yizheng Xu, “Methodology for estimation of dynamic response of demand using limited data”, IEEE Transactions on Power Systems, Vol. 30, No 3, 2015, pp. 1288 – 1297 Yizheng Xu and J.V. Milanović, “Day-ahead Prediction and Shaping of Dynamic Response of Demand at Bulk Supply Points ”, accepted for publication in the IEEE Transactions on Power Systems, TPWRS-00371-2015, (5/9/15)
Probabilistic estimation of PQ performance of active distribution network
• Load profiles (industrial, commercial and domestic loads): – Load profiles taken from a day in August 2012 in UK. – Different load types/classes follow different loading curves.
• Variation of DG outputs: – Wind turbines: based on practical outputs of wind turbines in UK. – Photovoltaic: based on true data available in the network. – Fuel cell: assumed variable daily output
• Power quality phenomena: – Unbalance: 9 fixed unbalance loads whose power factors follow pre-set normal
distribution; twenty-one single phase connected DGs. – Harmonic current injection: 10 fixed non-linear loads; 20 randomly selected
non-linear loads; wind turbines; photovoltaic; fuel cell – as above. The ratio of magnitude of the injected harmonic current to that of the fundamental component follow pre-set normal distributions.
– Voltage Sag:s based on statistic fault rates of components in the network.
37/35
Probabilistic estimation of PQ performance of active distribution network
0 50 100 150 200 2500
0.2
0.4
0.6
0.8
1
1.2
Bus No
UBPI
(Har
monic
, Sag
, Unb
alanc
e)
Without mitigationWith mitigationErrorZone threshold
Overall Power Quality performance
Provision of “contracted PQ” for individual zones using FACTS & PHF only 38/35
Probabilistic estimation of PQ performance of active distribution network
39/35
Huilian Liao, Sami Abdelrahman, Yue Guo and J.V. Milanović, “Identification of weak areas of network based on exposure to voltage sags—Part I: Development of sag severity index for single-event characterization”, accepted for publication in the IEEE Transactions on Power Delivery, TPWRD-01430-2013 (5/10/14) Huilian Liao, Sami Abdelrahman, Yue Guo and J.V. Milanović, “Identification of weak areas of network based on exposure to voltage sags — Part II: Assessment of network performance using sag severity index”, accepted for publication in the IEEE Transactions on Power Delivery, TPWRD-01434-2013 (5/10/14) Huilian Liao, Zhixuan Liu, J. V. Milanović, and Nick C. Woolley, “Optimisation Framework for Development of Cost-effective Monitoring in Distribution Networks“, accepted for publication in the IET Generation, Transmission and Distribution, GTD-2015-0757, (16/09/15)
Prob
abili
stic
ann
ual
estim
atio
n an
d op
timis
atio
n of
PQ
per
form
ance
of
dist
ribu
tion
netw
ork
Annual performance
Optimised annual technical performance
Optimised annual techno – economic performance
Summary
41/35
Future power networks need to be (will likely be) modelled and operated by exploiting possibilities offered by state-of-the-art WAMS, integrated ICT systems and “intelligent” PE devices and using – Non-deterministic & close to real time approaches for (energy)
system control and operation – Stochastic, probabilistic and computer intelligence based
models, data handling and methodologies to minimise the effect of uncertainties and maximise the use of information contained in available data
Summary
42/35
Key areas to watch
43/35