artificial neural network modelling of photodegradation … · oxide nanoparticles under...

Research ArticleArtificial Neural Network Modelling ofPhotodegradation in Suspension of Manganese Doped ZincOxide Nanoparticles under Visible-Light Irradiation

Yadollah Abdollahi12 Azmi Zakaria1 Nor Asrina Sairi2 Khamirul Amin Matori1

Hamid Reza Fard Masoumi3 Amir Reza Sadrolhosseini4 and Hossein Jahangirian5

1 Material Synthesis and Characterization Laboratory Institute of Advanced Technology Universiti Putra Malaysia43400 Serdang Selangor Malaysia

2 Chemistry Department Faculty of Science University of Malaya 50603 Kuala Lumpur Malaysia3 Department of Chemistry Faculty of Science University Putra Malaysia 43400 Serdang Selangor Malaysia4Wireless and Photonics Networks Research Center (WiPNET) Faculty of Engineering Universiti Putra Malaysia Serdang Malaysia5 Department of Chemical and Environmental Engineering Faculty of Engineering Universiti Putra Malaysia (UPM)43400 Serdang Selangor Malaysia

Correspondence should be addressed to Yadollah Abdollahi yadollahabdollahiyahoocom

Received 17 March 2014 Revised 30 July 2014 Accepted 8 September 2014 Published 4 November 2014

Academic Editor J Paul Chen

Copyright copy 2014 Yadollah Abdollahi et al This is an open access article distributed under the Creative Commons AttributionLicense which permits unrestricted use distribution and reproduction in any medium provided the original work is properlycited

The artificial neural network (ANN) modeling of m-cresol photodegradation was carried out for determination of the optimumand importance values of the effective variables to achieve the maximum efficiency The photodegradation was carried out in thesuspension of synthesized manganese doped ZnO nanoparticles under visible-light irradiation The input considered effectivevariables of the photodegradation were irradiation time pH photocatalyst amount and concentration of m-cresol while theefficiency was the only response as output The performed experiments were designed into three data sets such as training testingand validation that were randomly splitted by the softwarersquos option To obtain the optimum topologies ANN was trained byquick propagation (QP) Incremental Back Propagation (IBP) Batch Back Propagation (BBP) and Levenberg-Marquardt (LM)algorithms for testing data set The topologies were determined by the indicator of minimized root mean squared error (RMSE)for each algorithm According to the indicator the QP-4-8-1 IBP-4-15-1 BBP-4-6-1 and LM-4-10-1 were selected as the optimizedtopologies Among the topologies QP-4-8-1 has presented theminimumRMSE and absolute average deviation as well asmaximumR-squared Therefore QP-4-8-1 was selected as final model for validation test and navigation of the process The model was usedfor determination of the optimum values of the effective variables by a few three-dimensional plots The optimum points of thevariables were confirmed by further validated experiments Moreover the model predicted the relative importance of the variableswhich showed none of them was neglectable in this work

1 Introduction

According to the last report of united nation world waterdevelopment most untreated industrial wastewater whichcontains several kinds of organic pollutants such as phenoliccompounds is flowing into the productive lands surfaceand underground water sources [1] To prevent the haz-ardous materials from entering into the environment theeffective and environmental compatibility removal methods

are attracting the attentions Therefore several chemicalphysical and biologicalmethods have been applied to removethe pollutants by using chemical coagulation oxidationflocculation precipitation froth floatation reverse osmosisand biological techniques [2 3] However the chemicalmethods are unable to mineralize all the organics and alsogenerate new environmental pollutants [4] In the sametrend the biological methods are slow selective pH andtemperature sensitive [5 6] The physical methods such as

Hindawi Publishing Corporatione Scientific World JournalVolume 2014 Article ID 726101 10 pageshttpdxdoiorg1011552014726101

2 The Scientific World Journal

adsorption techniques are unable to remove the hazardousfrom the environment On the other side advanced oxidationprocesses (AOPs) such as heterogeneous photocatalytic pro-cesses including photocatalyst Fenton photo-Fenton andelectrooxidation are powerful and nonselective methods thathave been used to convert the persistent organic pollutantsto an environmental friendly product [7ndash9] Among thevarious AOPs methods the heterogeneous photocatalyticprocess has been succeed due to its ability for destroyinga wide range of the pollutants at ambient temperature andpressure without generation of harmful intermediates [10ndash17] The processes use a catalyst that is active under UVor visible-light irradiation to generate hydroxyl radical [18]Zinc oxide (ZnO) is well-known nontoxic semiconductormaterials that has been used as heterogeneous photocatalystto investigate water purification [19 20] ZnO has facilitatedseveral degradations of thewater organic contaminants underUV irradiation [21 22] In addition the grate advantages ofZnO are absorption of a large fraction of the solar spectrum(sunlight) which is free and available around the world [23]Sunlight consists of 47 visible-light with wavelength of 400to 700 nmor energy of 177 to 276 eVThus visible-light couldbe an excellent source of energy for the photocatalytic activityHowever the photo activity of ZnO under the energy wasvery low due to its high direct band gap energy 32 eV [19 20]Therefore several methods of synthesis have been examinedto improve the band gap such as transitionmetals doped ZnO[24ndash29] As the absorption spectra red shift has showed the3D orbital of the metals goes between valence band (VB)and conduction band (CB) of the semiconductors in dopingprocess [30 31] It depends on the energy of the 3D orbitalsthey overlap with the VB or CB of the semiconductors [32]For the reason the electrons are excited from VB (the 3DoverlappedVB) to CB (3D overlapped CB) during irradiationprocess On the other hand the energy of manganese (Mn)3D orbitals is very close to the VB of ZnO which easilyoverlap to decrease the 119864119892 [10 17] In our previous workMn doped ZnO was synthesized and applied for degrada-tion of organic pollutants [29] The photodegradation wasstudied by one variable at a time technique with the effectivevariables of irradiation times pH photocatalyst amount andconcentration of the pollutants [33 34] The problem is thatthe technique varies one of the parameters while the otherterms are kept constant during themultivariate performanceTherefore it has adverse effects on the photodegradationthat should be studied by multivariate methods [17 21]Moreover the kinetic determination of the process is quitecomplicated by consideration of themass transfer the radiantenergy balance the spatial distribution of the absorbedradiation andmechanisms of the photochemical degradation[35] The photodegradation as a process consists of inputfactors such as effective variables and the efficiency as outputresponse The changing amount of the effective variablesaffects the value of the efficiency Therefore the amount ofthe variable could be optimized to achieve the maximumefficiency that is free of the mentioned complexities Theknown multivariate methods that used to optimize theproductive process included response surface methodology(RSM) and artificial neural network (ANN) [35 36]TheRSM

designs the related experiments and then fits the observedresults of performed design to appropriate polynomial andsuggests the qualified model for more validation The modelas a mathematic equation indicates the relationships betweenvariables themselves variables and response(s) Thereafterthe validated model is used to optimize the effective variablesto achieve the maximum yield of the products Howeverthe method is involved with the complicated statisticalcalculation such as analysis of variance fitting and regressionprocess for modeling process [37 38] On the other handANNs have been widely used for modeling of chemicaland biochemical reaction process [39ndash43] The ANN mod-eling has been reliable robust and salient characteristicsin capturing the nonlinear relationship between the inputand output variables which is free of complexities In thiswork the multilayer feed-forward neural network was usedto model m-cresol photodegradation in manganese dopedZnO nanoparticles (Mn doped ZnO NPs) suspension undervisible-light irradiation The cresol is widely used in severalmanufacturing products with high water solubility which hasbeen listed as priority pollutants persistent toxic chemicaland a significant threat to the environment [44 45]The inputeffective operational parameters were including irradiationtime pH photocatalyst amount and concentration of thecresol while the efficiency was the only response as output

2 Materials and Methods

The chemicals of this work were obtained from Merck andwere used without further purification The m-cresol (99)was used as organic water pollutant while H2SO4 (95ndash97)and NaOH (99) were applied to set the appropriate pHThe Mn doped ZnO NPs with average particles size 35 nmbang gap energy 22 eV and surface area 35m2gminus1 wereused as photocatalyst The photocatalyst was synthesized byprecipitation method in absolute alcohol according to ourpublished work [29] To degrade the pollutant the variousconcentrations of m-cresol were mixed with appropriateamount of the photocatalyst in 500mL deionized water Themixture (suspension) solution was irradiated by a Philipslamp (23watts) as light source in a batch homemade pho-toreactor that was used in our previous work [46] Thesuspension was magnetically stirred during the irradiationat 200 rpm Moreover air was blown into the solution byusing an air pump at a flow rate of 150mLmin to increasesolution fluidization access oxygen volatile the produced gas(CO2) and keep the temperature at around 25∘C During theperformance samples were withdrawn from the bulk solu-tion at specific time intervals and centrifuged at 14000 rpmfor 20min and then they filtered through 02 120583m PTFE filterto measure the remained concentration of the m-cresol Themeasurementwas carried out by a ShimadzuUV-1650PCanda TOC-VCSN analyzer respectively In addition the initialcatalyst absorption and photocatalyst were investigated indark and absence of the photocatalyst at normal pH (75)that were considered in the efficiencies calculation [47] Theefficiencies were used as actual responses for modeling of thephotodegradation which was carried out by Neural PowerSoftware version 25 [48 49] The total of 31 experiment

The Scientific World Journal 3

Table 1 The training testing and validation data sets of the effective variables and actual and models predication efficiency of m-cresolphotodegradation by Mn doped ZnO NPs

Run Irradiation time pH Photocatalyst m-cresol EfficiencyActual Predicted

Training data set1 360 763 15 65 389 370572 120 763 15 35 3086 304353 360 4 15 35 464 464574 360 10 15 35 487 489225 360 763 3 35 837 840016 360 763 17 35 95 958647 240 763 15 35 6508 660738 360 85 15 35 893 903239 360 6 15 35 672 6833310 360 8 15 35 934 9474111 360 763 1 35 799 813312 300 763 15 35 8142 829613 360 763 15 55 4680 4962514 360 763 35 35 724 8001615 360 763 15 15 100 10916

Testing data set17 360 9 15 35 727 7950418 360 763 05 35 502 6402619 360 763 15 41 816 7484520 360 763 13 35 942 8892621 360 763 15 30 992 9770322 360 763 15 25 100 9957223 360 763 25 35 879 9013124 60 763 15 35 1248 17993

Validation data set26 360 763 15 45 733 6278927 360 763 15 35 9898 9193928 360 763 15 76 325 2891529 180 763 15 35 4755 4674730 360 5 15 35 508 5244131 360 763 2 35 932 94923

points have been randomly splitted into training (15 points)testing (8 points) and validation (6 points) data sets (Table 1)by using the facilitated option in the software The trainingwas used to compute and ensure robustness of the networkparameters while the testing stage was used as control errorsto avoid overfitting [50] The validation data which wasexcluded from training and testing considered to assess thepredictive ability of the generated model [51]

3 Theory of the Work

31 The Theory of ANN ANNs are semiempirical multivari-ate methods that are used in mathematic free fictionalizationof the complicated productive process The networks containinput hidden and output layers which are made of severalnodes The nodes are connected by multilayer normal feed-forward or feed-back connection formula [52] The nodes

are simple artificial neurons which stimulates the behavior ofbiological neural networks The hidden layer could be morethan one parallel layer however the single hidden layer isuniversally suggested In the network the nodes of particularlayer are connected to the nodes of the next layer from leftto right by feed-forward formula The nodes of input layerare qualified by sending data via the special weights to thenodes of hidden layer and then to the output layer [52 53]The qualification is carried out by associated weights duringlearning process by well-known mathematic algorithms

32 The Learning Process The learning process determinesthe number of nodes in the hidden layer (topology) by usingtrial and error calculation The calculation is examined fromone to ldquo119899rdquo nodes to discover the architecture with minimumroot mean square error (RMSE) by using testing data set andparticular algorithm In learning process the input layer acts


as distributor for the hidden layer and the inputs and outputof the hidden layer are multiplied by weighted summation asfollows

119878 =

119899ℎ

sum

119894=1

(119887 minus 119882119894119868119894) (1)

where 119878 is summation 119887 is a bias [54] 119868119894 is the 119894th inputto hidden neuron and 119882119894 is the weight associated with 119868119894The bias shifts the space of the nonlinearity properties [55]Therefore the outputs of the hidden layer act as inputs tofinal layer (output) which are undergoing a transfer functionThe popular transfer function is the logarithmic sigmoid forboth hidden and output layers that is bounded from 0 to 1[56]The sigmoid bounded area is used to normalize the inputand output data that is provided by the software scaling Thescaled data are passed into the first layer and propagated tohidden layer and finally meet the output layer of the networkby iterative procedure The iteration is an act of repeating aprocess to approach a desire result The results of iterationare used as starting point of next iteration For examplewhen the results of last iteration become almost equal to theresults of previous iteration the process will be terminatedThe iteration process is continued by self-similarity methodas follows [57]

119878 (119861) =

119898

sum

119868=1

[119910119894 minus 119891 (119909119894120573)2] (2)

where ldquo119898rdquo is an empirical data pairs of independent anddependent variables such as (119909119894 119909119894) and 119891(119909119894 120573) is the modelcurve In self-similarity process the 120573 parameter of119891(119909119894 120573) isoptimized by minimizing the sum of the squares As a resultthe main aim of the learning process is to find the weightsfor minimizing the error of (RMSE) which is obtained fromdifference between network prediction and actual responsesas follows

RMSE = (1119899

119899

sum

119894=1

(119910119894 minus 119910119889119894)2)

12

(3)

where ldquo119899rdquo is number of the points 119910119894 is the predicted valuesand 119910119889119894 is the actual values To avoid random correlation dueto the random initialization of the weights the examinationof each node is repeated several times Among the repeatedexamination the architecture with lowest RMSE is selectedfor each node The RMSE of the architectures are comparedto find the best topology for the particular algorithm Thetopology is architecture with minimum relative RMSE Formore certainty the 119877-square (1198772) (see (4)) and absolute aver-age deviation (AAD) (see (5)) are calculated by performanceof the topology for training and testing data sets

1198772= 1 minussum119899

119894=1(119910119894 minus 119910119889119894)

2

sum119899

119894=1(119910119889119894 minus 119910119898)

2(4)

AAD = 1119899

119899

sum

119868=1

1003816100381610038161003816119910119894 minus 1199101198891198941003816100381610038161003816

119910119889119894

(5)

25

4

55

7

0 3 6 9 12 15

RMSE

Node number

QPIBP

BBPLM

Figure 1 The selected RMSE versus node number of the photodeg-radation networkrsquos hidden layer for QP IBP BBP and LM algorithmThe smallest RMSE belong to node of 8 (QP) 15 (IBP) 6 (BBP) and10 (LM)

where ldquo119899rdquo is the number of points 119910119894 is the predictedvalue 119910119889119894 is the actual value and 119910119898 is the average of theactual values The learning process is carried out for differentalgorithms to obtain the best topology Then the RMSEAAD and 1198772 of the topologies are compared to find theoptimized topology that is selected as provisional model forthe process The model is evaluated by validation data set(Table 1) Thereafter it is used for navigation of the processthat determines the optimum and importance of the inputvariables to maximize the yield of the process

4 Results and Discussion

41 The Modeling Process The network of the photodegra-dation process contains input hidden and output layerswhich they are made of one or more number of nodes Thestructure of the input and output layer was determined bynumber of the effective variables and the efficiency of thephotodegradation while the structure of the hidden layer wasdetermined by the modelingThe effective variables includedtemperature pH photocatalyst amount and concentration ofm-cresol

411 The Structure of the Hidden Layer To obtain the struc-ture of the hidden layer 15 architectures that contained 1to 15 nodes were examined for quick propagation (QP)Incremental Back Propagation (IBP) Batch Back Propaga-tion (BBP) and Levenberg-Marquardt (LM) algorithms Theexamination was repeated 10 times for each node by testingdata setThen among the 10 repetitions the architecture withthe smallest RMSE was selected for each node Therefore 15architectures were considered for each algorithm illustratedin Figure 1 to find the optimized topologies Figure 1 plotsthe value of RMSE versus the node number for QP IBPBBP and LM algorithms For each algorithm one of the15 architectures has presented minimum RMSE (topology)considered for more evaluation The hidden layerrsquos nodenumbers of selected topologies were 8 15 6 and 10 for QPIBP BBP and LM logarithms respectively The evaluation of


0

20

40

60

80

100

0 20 40 60 80 100

Pred

icte

d effi

cien

cy

Actual efficiency

QP-4-8-1

Testing

Best linear fit

R2= 09938

A = P

0

20

40

60

80

100

0 20 40 60 80 100

Pred

icte

d effi

cien

cy

Actual efficiency

BBP-4-6-1

Testing

Best linear fit

R2= 0988

A = P

0

20

40

60

80

100

0 20 40 60 80 100

Pred

icte

d effi

cien

cy

Actual efficiency

IBP-4-15-1

Testing

Best linear fit

R2= 09857

A = P

0

20

40

60

80

100

0 20 40 60 80 100

Pred

icte

d effi

cien

cy

Actual efficiency

LM-4-10-1

Testing

Best linear fit

R2= 09883

A = P

Figure 2 The scatter plots of the predicted efficiency versus actual efficiency for testing data set that shows the performed 1198772 of optimizedtopologies QP-4-8-1 BBP-4-6-1 IBP-4-15-1 and LM-4-10-1

the topologies was carried out by comparison of minimumRMSE and AAD as well as maximum 1198772 to discover theprovisional model of the photodegradation The comparisonof the RMSE proved that the QP with 4 nodes in input 8nodes in hidden and 1 node in output layer (QP-4-8-1) haspresented the minimum root mean squared error (Figure 1)

Then the performed results of the topologies were usedto calculate 1198772 (see (4)) and ADD (see (5)) To calculatethe 1198772 the prediction of the topologies and actual values ofthe efficiency were plotted for testing data set in Figure 2As the scatter plots showed the topology of QP-4-8-1 haspresented the highest 1198772 0993 that in comparison withother topologies has the best performance

Moreover the AAD of the topologies in testing data setwas calculated for QP-4-8-1 IBP-4-15-1 BBP-4-6-1 and LM-4-10-1 (Figure 3) As shown the lowest value of the AADhas also belonged to topology of QP-4-8-1 As a result thetopology of QP-4-8-1 was pioneer in minimum RMSE andAAD as well as at maximum of 1198772 among those topologiesfor testing data sets Therefore QP-4-8-1 was selected asfinal optimumprovisionalmodel of the photodegradation forvalidation test

412 Validation of the Selected Model The provisionalmodel (QP-4-8-1) was validated by 6 experimental pointsexcluded from training and testing data sets (Table 1)


7

85 8693

0

2

4

6

8

10

QP-4-8-1 LM-4-10-1 BBP-4-6-1 IBP-4-15-1

AD

D

Figure 3 The AAD of QP-4-8-1 BBP-4-6-1 IBP-4-15-1 and LM-4-10-1 topologies in testing data set

0

20

40

60

80

100

0 20 40 60 80 100

Pred

icte

d effi

cien

cy

Actual efficiency

QP-4-8-1

Validation

Best linear fit

R2= 09724

A = P

Figure 4 The validation test of the provisional model (QP-4-8-1)which consists of actual and predicted photodegradation efficiencyas well as the best linear fit and 119877-square

The validation was investigated by scatter plots of the modelprediction versus actual values of the photodegradationefficiency (Figure 4) As demonstrated the 1198772 0972 ofthe performance was quite close to 1 that confirmed themodel is significant In addition the RMSE and AAD of theperformed validation included 4574 and 5668 respectivelywhich proved the great predictive accuracy of the model

413 The Final Model of the Photodegradation Figure 5shows the structure of QP-4-8-1 topology as final model form-cresol photodegradation The model consists of three lay-ers that included input hidden and output layers The inputlayer with 4 nodes of effective variables acts as distributor forthe hidden layer with 8 nodes The inputs as well as output ofhidden nodes are multiplied by the appropriate weights [55]

Bias

Inputs Outputs

Figure 5The structure of the photodegradation final model of QP-4-8-1 that consists of 4 nodes in input layer 8 nodes in hidden layerand 1 node in output layer as response

Then the nodes outputs of hidden layer are transferred tooutput layer by using log-sigmoid transfer function (see (6))[58] The function normalizes transfers the data between thelayers [56]

119891 (119909) =1

1 + exp (minus119909) (6)

where 119891 (119909) is the hidden output neurons Therefore QP-4-8-1 was used to navigate the process for determination ofoptimum and importance values of the photodegradationinput variables

42 The Model Applications of the Photodegradation As ashort overview the modeling process optimized the topolo-gies of different learning algorithms by using testing andtraining data sets Then the best topology with optimum1198772 RMSE and AAD was selected as provisional model for

more evaluation The adequacy of the selected model (QP-4-8-1) was evaluated by validation data set The validationmodel QP-4-8-1 was used to navigate the photodegradationThe navigation has contained graphical optimization and therelative importance of the input effective

421 The Variables Graphical Optimization The validatedmodel QP-4-8-1 simulated the behavior of the photodegra-dation without further requirement of mathematic knowl-edge (Figures 6 to 8)The simulations consist of effect of non-linear relationship of two variables on photodegradation effi-ciency which is graphically presented by three dimensionalplots (3D) while the other parameters were kept constant atthe middle of their levelsrsquo values The values of irradiationtime were 210min pH was 75 photocatalyst amount was2 (gL) and concentration ofm-cresol was 455 (mgL)

Figure 6 shows the variation of the efficiency in pH 5to pH 10 during irradiation time 60 to 360min while theamount of photocatalyst and m-cresol were kept constant at


pH

5 567 6 667 7767 8

867 9 967

Effici

ency

()

8575655545352515

Irrad

iation

time (

min)

340300

260220

180140

10060

Figure 6 The variation of the photodegradation efficiency in pH5 to pH 10 and irradiation time (60 to 360min) the amount ofphotocatalyst (2 gL) and m-cresol (455mgL) were at middle oftheir levels

Catalyst (gL)

05 09 13 17 21 25 29 33

Effici

ency

()

655545352515

5

Irrad

iation

time (

min)

340300

260220

180140

10060

Figure 7 The efficiency variation of the photodegradation withvarying photocatalyst from 05 to 35 gL and irradiation time(60minus360min) the pH was 75 andm-cresol was 455mgL

2 gL and 455mgL respectively As shown the efficiencywas increased up to pH 9 and thereafter it was decreasedfor irradiation time 300 to 360minTherefore the maximumsurface of the 3D plot has been demonstrated at pH (8 to 9)and the 340 to 360min of irradiation time

Figure 7 shows the photodegradation efficiency versusthe photocatalyst amount from 05 to 35 gL and irradiationtime (60minus360min) The pH andm-cresol were kept constant75 and 455mgL respectively As shown the efficiency wasincreased up to 13 gL of the photocatalyst then it wasconstant up to 17 gL and finally it was decreasedThereforethe optimum value of the photocatalyst was in the level of 13to 17 gL

Figure 8 shows the efficiency of the photodegradation indifferent concentration of m-cresol (25 to 80mgL) in levelof irradiation time from 60 to 360min while the pH 75 andthe photocatalyst 2 gL were kept constant As the 3D plotdemonstrates the efficiency was continually decreasing with

15 231 313 394 475 557 638 719

Effici

ency

()

100

80

60

40

20

Irrad

iation

time (

min)

340300

260220

180140

10060M-cresol (mgL)

Figure 8 The variation of the efficiency in different concentrationofm-cresol from 25 to 80mgL with irradiation time (60minus360min)the pH was 75 and amount of the photocatalyst was 2 gL

increasing m-cresol concentration However the removedm-cresol was more than 80 of 455mgL at the end ofthe irradiation time (340ndash360min)Therefore themaximumamount of removedm-cresol was 42mgL

The optimum value of the irradiation time was investi-gated by Figures 6 to 8 obtained in levels 340 to 360min atthe middle values of the other variables levels such as thephotocatalyst pH andm-cresol concentration As a result ofFigures 6 to 8 the optimum levels of the effective variableswere photocatalyst (13ndash17 gL) pH (8-9) m-cresol (40ndash45mgL) and irradiation time (340ndash360min) To predict theoptimum points of these levels the desirable condition suchas maximum m-cresolrsquos concentration minimum amount ofthe photocatalyst and pH value at the end of the irradiationtime was considered as input for the model The modelprediction included 13 gL photocatalyst 42mgL m-cresolpH 8 and 360min of the irradiation time while the predictedefficiency was 100 Then this condition was evaluated byfurther experiments and observed efficiencywas 99 (almostall m-cresol was removed) that confirmed the improvementin comparison with other previous works [59]

422 Importance of the Effective Variables The relative im-portance of the effective variables in the optimum levelswas determined by the model as presented in Figure 9 Asdemonstrated the greatest importance belonged to m-cresol(3293) and pH (3202) However the effects of othervariables such as photocatalyst (2128) and irradiation time(1377) were also important for the efficiency As a resultthe selected variableswere effectiveness andnone of themwasneglectable in this photodegradation

423 The Model Multivariate Navigation To navigate thephotodegradation the model (Figure 5) was used to deter-mine the optimum levels predict optimumpoints and obtainthe importance of the effective variables The variables wereinitially used in a wide range and identical importance andwithout any considered points The obtained information ispresented by Table 2 The optimum levels were achieved by


Table 2 The results of multivariate modeling and optimization ofm-cresol photodegradation under visible-light irradiation

Variable m-cresol (mgL) pH Photocatalyst (gL) Irradiation time (min) Efficiency ()Optimum level 40 to 45 8 to 9 13 to 17 340 to 460 mdashPredicted point 42 8 13 360 100Verified point 42 8 13 360 99Importance () 3293 3202 2128 1377 mdash

M-cresol (mgL)pH

Catalyst (gL)Irradiation time (min)

2128

1377

3202

3293

Figure 9 The relative importance of the photodegradation inputvariables such as irradiation time pH amount of the photocatalystandm-cresolrsquos concentration

3D plots in graphical vision (Figures 6 to 8) The optimumpoints of the variables in the optimized narrow levels werefirstly predicted by the model and verified by further exper-iment with 1 error Moreover the model determined therelative importance of the variables which showed none ofthem were neglectable in this work

5 Conclusions

The AAN modeling of m-cresol photodegradation was car-ried out to determination of optimum and importancevalues of the effective variables to achieve maximum effi-ciencyThephotodegradationwas performed in synthesisMndoped ZnO suspension and under visible-light irradiationThe input considered effective variables of the photodegra-dation were irradiation time pH photocatalyst amountand concentration of m-cresol while the efficiency was theonly response as output The performed experiments weredesigned in three data sets such as training testing andvalidation that were randomly splitted by the softwarersquosoption To obtain the optimum topologies AAN was trainedby QP IBP BBP and LM algorithms for testing data set Thetopologies were determined by the indicator of minimizedRMSE for each algorithmAccording to the indicator theQP-4-8-1 IBP-4-15-1 BBP-4-6-1 and LM-4-10-1 were selectedas the optimized topologies Among the topologies QP-4-8-1 has presented the minimum RMSE and AAD as wellas maximum 1198772 Therefore QP-4-8-1 was selected as finalmodel for the photodegradation navigation The model wasused for determination of the optimum values of the effectivevariables to achieve the maximum efficiency by using graph-ical vision The predicted optimum points of the variables

were confirmed by further validated experiments Moreoverthe model predicted the relative importance of the variableswhich showed none of them was neglectable in this work

Conflict of Interests

The authors declare that they have no competing interests

Acknowledgments

The authors wish to thank Dr Mina Abbasiyannejad forpolishing and proofreading the paper Also they wish tothank Dr Abdul Halim Abdullah for his help in this study

References

[1] P H GleickTheWorldrsquos Water 2004-2005 The Biennial Reporton Freshwater Resources Island Press 2004

[2] T Robinson G McMullan R Marchant and P Nigam ldquoReme-diation of dyes in textile effluent a critical review on currenttreatment technologieswith a proposed alternativerdquoBioresourceTechnology vol 77 no 3 pp 247ndash255 2001

[3] M Tamez Uddin M Rukanuzzaman M Maksudur RahmanKhan and M Akhtarul Islam ldquoAdsorption of methylene bluefrom aqueous solution by jackfruit (Artocarpus heteropyllus)leaf powder a fixed-bed column studyrdquo Journal of Environmen-tal Management vol 90 no 11 pp 3443ndash3450 2009

[4] N Takeda and K Teranishi ldquoGeneration of superoxide anionradical from atmospheric organic matterrdquo Bulletin of Environ-mental Contamination and Toxicology vol 40 no 5 pp 678ndash682 1988

[5] O Aviam G Bar-Nes Y Zeiri and A Sivan ldquoAccelerated bio-degradation of cement by sulfur-oxidizing bacteria as a bioassayfor evaluating immobilization of low-level radioactive wasterdquoApplied and Environmental Microbiology vol 70 no 10 pp6031ndash6036 2004

[6] M L Krumme and S A Boyd ldquoReductive dechlorination ofchlorinated phenols in anaerobic upflow bioreactorsrdquo WaterResearch vol 22 no 2 pp 171ndash177 1988

[7] O Carp C L Huisman and A Reller ldquoPhotoinduced reactivityof titanium dioxiderdquo Progress in Solid State Chemistry vol 32no 1-2 pp 33ndash177 2004

[8] A Fujishima T N Rao and D A Tryk ldquoTitanium dioxidephotocatalysisrdquo Journal of Photochemistry and Photobiology CPhotochemistry Reviews vol 1 no 1 pp 1ndash21 2000

[9] A R Khataee V Vatanpour andA R A Ghadim ldquoDecoloriza-tion of CI Acid Blue 9 solution by UVNano-TiO

2 Fenton

Fenton-like electro-Fenton and electrocoagulation processes acomparative studyrdquo Journal of Hazardous Materials vol 161 no2-3 pp 1225ndash1233 2009


[10] S K Pardeshi and A B Patil ldquoA simple route for photocatalyticdegradation of phenol in aqueous zinc oxide suspension usingsolar energyrdquo Solar Energy vol 82 no 8 pp 700ndash705 2008

[11] A Sharma P Rao R P Mathur and S C Ameta ldquoPhotocat-alytic reactions of xylidine ponceau on semiconducting zincoxide powderrdquo Journal of Photochemistry and Photobiology AChemistry vol 86 no 1ndash3 pp 197ndash200 1995

[12] N Daneshvar S Aber M S Seyed Dorraji A R Khataeeand M H Rasoulifard ldquoPhotocatalytic degradation of theinsecticide diazinon in the presence of prepared nanocrystallineZnO powders under irradiation of UV-C lightrdquo Separation andPurification Technology vol 58 no 1 pp 91ndash98 2007

[13] S K Kansal M Singh and D Sud ldquoStudies on TiO2ZnO

photocatalysed degradation of ligninrdquo Journal of HazardousMaterials vol 153 no 1-2 pp 412ndash417 2008

[14] A Akyol H C Yatmaz and M Bayramoglu ldquoPhotocatalyticdecolorization of Remazol Red RR in aqueous ZnO suspen-sionsrdquo Applied Catalysis B Environmental vol 54 no 1 pp 19ndash24 2004

[15] Y Abdollahi A H Abdullah and Z Zainal ldquoPhotodegradationof m-cresol by zinc oxide under visible-light irradiationrdquoInternational Journal of Chemistry vol 3 no 3 pp 31ndash43 2011

[16] Y Abdollahi1 Z Zainal and N A Yusof ldquoPhotodegradationof o-cresol by ZnO under UV irradiationrdquo Journal of AmericanScience vol 7 no 8 pp 165ndash170 2011

[17] Y Abdollahi Z Zainal and N A Yusof ldquoPhotodegradation ofp-cresol by zinc oxide under visible lightrdquo International Journalof Applied Science and Technology vol 1 no 5 pp 99ndash105 2011

[18] U I Gaya and A H Abdullah ldquoHeterogeneous photocatalyticdegradation of organic contaminants over titanium dioxidea review of fundamentals progress and problemsrdquo Journal ofPhotochemistry and Photobiology C Photochemistry Reviewsvol 9 no 1 pp 1ndash12 2008

[19] N Vitchuli Q Shi J Nowak et al ldquoMultifunctional ZnONylon6 nanofiber mats by an electrospinning-electrospraying hybridprocess for use in protective applicationsrdquo Science and Technol-ogy of AdvancedMaterials vol 12 no 5 Article ID 055004 2011

[20] F Kayaci C Ozgit-Akgun N Biyikli and T Uyar ldquoSurface-decorated ZnO nanoparticles and ZnO nanocoating on elec-trospun polymeric nanofibers by atomic layer deposition forflexible photocatalytic nanofibrousmembranesrdquoRSCAdvancesvol 3 no 19 pp 6817ndash6820 2013

[21] Y Abdollahi A H Abdullah Z Zainal and N A YusofldquoPhotocatalytic degradation of p-cresol by zinc oxide under UVirradiationrdquo International Journal of Molecular Sciences vol 13no 1 pp 302ndash315 2012

[22] Y Abdollahi A Zakaria and N A Sairi ldquoDegradation of highlevel m-Cresol by zinc oxide as photocatalystrdquo CLEANmdashSoilAir Water vol 42 no 9 pp 1292ndash1297 2013

[23] B Dindar and S Icli ldquoUnusual photoreactivity of zinc oxideirradiated by concentrated sunlightrdquo Journal of Photochemistryand Photobiology A Chemistry vol 140 no 3 pp 263ndash2682001

[24] F Kayaci C Ozgit-Akgun I Donmez N Biyikli and T UyarldquoPolymer-inorganic core-shell nanofibers by electrospinningand atomic layer deposition flexible nylon-ZnO core-shellnanofiber mats and their photocatalytic activityrdquo ACS AppliedMaterials and Interfaces vol 4 no 11 pp 6185ndash6194 2012

[25] C-C Chen ldquoDegradation pathways of ethyl violet by photo-catalytic reaction with ZnO dispersionsrdquo Journal of MolecularCatalysis A Chemical vol 264 no 1-2 pp 82ndash92 2007

[26] B Pandey S Ghosh P Srivastava D K Avasthi D Kabiraj andJ C Pivin ldquoSynthesis and characterization of Ni-doped ZnOa transparent magnetic semiconductorrdquo Journal of Magnetismand Magnetic Materials vol 320 no 24 pp 3347ndash3351 2008

[27] D Chu Y-P Zeng and D Jiang ldquoSynthesis and growthmechanism of Cr-doped ZnO single-crystalline nanowiresrdquoSolid State Communications vol 143 no 6-7 pp 308ndash312 2007

[28] Y R Park and K J Kim ldquoOptical and electrical propertiesof Ti-doped ZnO films observation of semiconductor-metaltransitionrdquo Solid State Communications vol 123 no 3-4 pp147ndash150 2002

[29] Y Abdollahi A H Abdullah Z Zainal and N A YusofldquoSynthesis and characterization of manganese doped ZnOnanoparticlesrdquo International Journal of BasicampApplied Sciencesvol 11 no 4 pp 62ndash69 2011

[30] J Moser M Gratzel and R Gallay ldquoInhibition of electron-holerecombination in substitutionally doped colloidal semiconduc-tor crystallitesrdquoHelvetica Chimica Acta vol 70 no 6 pp 1596ndash1604 1987

[31] E Borgarello J Kiwi M Gratzel E Pelizzetti and M ViscaldquoVisible light induced water cleavage in colloidal solutions ofchromium-doped titanium dioxide particlesrdquo Journal of theAmerican Chemical Society vol 104 no 11 pp 2996ndash3002 1982

[32] W Choi A Termin and M R Hoffmann ldquoThe role ofmetal ion dopants in quantum-sized TiO

2 correlation between

photoreactivity and charge carrier recombination dynamicsrdquoThe Journal of Physical Chemistry vol 98 no 51 pp 13669ndash13679 1994

[33] Y Abdollahi A H Abdullah U I Gaya Z Zainal and NA Yusof ldquoEnhanced photodegradation of o-cresol in aqueousMn(1)-doped ZnO suspensionsrdquo Environmental Technologyvol 33 no 10 pp 1183ndash1189 2012

[34] Y Abdollahi A H Abdullah U I Gaya et al ldquoPhotocatalyticdegradation of 14-Benzoquinone in aqueous ZnO dispersionsrdquoJournal of the Brazilian Chemical Society vol 23 no 2 pp 236ndash240 2012

[35] A R Khataee and M B Kasiri ldquoArtificial neural networksmodeling of contaminated water treatment processes by homo-geneous and heterogeneous nanocatalysisrdquo Journal ofMolecularCatalysis A Chemical vol 331 no 1-2 pp 86ndash100 2010

[36] Y Abdollahi A Zakaria A H Abdullah et al ldquoSemi-empiricalstudy of ortho-cresol photo degradation in manganese-dopedzinc oxide nanoparticles suspensionsrdquo Chemistry Central Jour-nal vol 6 no 1 article 88 2012

[37] Y Abdollahi A Zakaria K A Matori et al ldquoInteractionsbetween photodegradation componentsrdquo Chemistry CentralJournal vol 6 no 1 article 100 2012

[38] S Weisberg Applied Linear Regression John Wiley amp Sons2005

[39] D Salari N Daneshvar F Aghazadeh and A R KhataeeldquoApplication of artificial neural networks for modeling of thetreatment of wastewater contaminated with methyl tert-butylether (MTBE) by UVH

2O2processrdquo Journal of Hazardous

Materials vol 125 no 1ndash3 pp 205ndash210 2005[40] S Aber A R Amani-Ghadim and V Mirzajani ldquoRemoval of

Cr(VI) frompolluted solutions by electrocoagulationmodelingof experimental results using artificial neural networkrdquo Journalof Hazardous Materials vol 171 no 1ndash3 pp 484ndash490 2009

[41] S Gob E Oliveros S H Bossmann A M Braun R Guardaniand C A O Nascimento ldquoModeling the kinetics of a pho-tochemical water treatment process by means of artificial


neural networksrdquo Chemical Engineering and Processing ProcessIntensification vol 38 no 4ndash6 pp 373ndash382 1999

[42] P Calza V A Sakkas A Villioti et al ldquoMultivariate experimen-tal design for the photocatalytic degradation of imipraminedetermination of the reaction pathway and identification ofintermediate productsrdquoApplied Catalysis B Environmental vol84 no 3 pp 379ndash388 2008

[43] V K Pareek M P Brungs A A Adesina and R Sharma ldquoArti-ficial neural network modeling of a multiphase photodegra-dation systemrdquo Journal of Photochemistry and Photobiology AChemistry vol 149 no 1ndash3 pp 139ndash146 2002

[44] N Shivaraman and R Pandey ldquoCharacterization and biodegra-dation of phenolic wastewaterrdquo Journal of Indian Association forEnvironmental Management vol 27 pp 12ndash15 2000

[45] M Callahan M Slimak N Gbel et al ldquoWater related environ-mental fate of 120 priority pollutantsrdquo Tech Rep EPA-44014-79-029a b United States Environmental Protection AgencyNTIS Washington DC USA 1979

[46] Y Abdollahi A H Abdullah A Zakaria Z Zainal H R FMasoumi and N A Yusof ldquoPhotodegradation of p-cresol inaqueousMn(1)-dopedZnO suspensionsrdquo Journal of AdvancedOxidation Technologies vol 15 no 1 pp 146ndash152 2012

[47] M A Fox and M T Dulay ldquoHeterogeneous photocatalysisrdquoChemical Reviews vol 93 no 1 pp 341ndash357 1993

[48] M Sanchooli and M G Mighaaddam ldquoEvaluation of acid-ity constants of anthraquinone derivatives in methanolwatermixtures using real quantum descriptorsrdquo Journal of ChemicalEngineering of Japan vol 45 no 6 pp 373ndash379 2012

[49] M G Moghaddam and M Khajeh ldquoComparison of responsesurfacemethodology and artificial neural network in predictingthe microwave-assisted extraction procedure to determine zincin fish musclesrdquo Food and Nutrition vol 2 pp 803ndash808 2011

[50] X Song A Mitnitski C MacKnight and K RockwoodldquoAssessment of individual risk of death using self-report data anartificial neural network compared with a frailty indexrdquo Journalof the American Geriatrics Society vol 52 no 7 pp 1180ndash11842004

[51] A Amani P York H Chrystyn B J Clark and D QDo ldquoDetermination of factors controlling the particle sizein nanoemulsions using artificial neural networksrdquo EuropeanJournal of Pharmaceutical Sciences vol 35 no 1-2 pp 42ndash512008

[52] A Ghaffari H Abdollahi M R Khoshayand I S BozchalooiA Dadgar and M Rafiee-Tehrani ldquoPerformance comparisonof neural network training algorithms in modeling of bimodaldrug deliveryrdquo International Journal of Pharmaceutics vol 327no 1-2 pp 126ndash138 2006

[53] G D Garson ldquoInterpreting neural-network connectionweightsrdquo AI Expert vol 6 no 4 pp 46ndash51 1991

[54] FDespagne andD LMassart ldquoNeural networks inmultivariatecalibrationrdquo Analyst vol 123 no 11 pp 157Rndash178R 1998

[55] J Zhang A J Morris E B Martin and C KiparissidesldquoPrediction of polymer quality in batch polymerisation reactorsusing robust neural networksrdquo Chemical Engineering Journalvol 69 no 2 pp 135ndash143 1998

[56] Y Abdollahi A Zakaria M Abbasiyannejad et al ldquoArtificialneural networkmodeling of p-cresol photodegradationrdquoChem-istry Central Journal vol 7 no 1 article 96 2013

[57] B B Mandelbrot ldquoHow long is the coast of Britain Statisticalself-similarity and fractional dimensionrdquo Science vol 156 no3775 pp 636ndash638 1967

[58] A Li H Li K Li and Z Gu ldquoApplications of neural networksand genetic algorithms to CVI processes in carboncarboncompositesrdquo Acta Materialia vol 52 no 2 pp 299ndash305 2004

[59] Y Abdollahi A H Abdullah Z Zainal and N A YusofldquoDegradation of m-cresol with Mn doped ZnO nanoparticlesunder visible light irradiationrdquo Fresenius Environmental Bul-letin vol 21 no 2 pp 256ndash262 2012

Submit your manuscripts athttpwwwhindawicom

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Inorganic ChemistryInternational Journal of

Hindawi Publishing Corporation httpwwwhindawicom Volume 2014

International Journal ofPhotoenergy


Carbohydrate Chemistry

International Journal of


Journal of

Chemistry


Advances in

Physical Chemistry

Hindawi Publishing Corporationhttpwwwhindawicom

Analytical Methods in Chemistry

Journal of

Volume 2014

Bioinorganic Chemistry and ApplicationsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

SpectroscopyInternational Journal of


The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

Medicinal ChemistryInternational Journal of


Chromatography Research International


Applied ChemistryJournal of



Theoretical ChemistryJournal of


Journal of

Spectroscopy

Analytical ChemistryInternational Journal of


Journal of


Quantum Chemistry


Organic Chemistry International

ElectrochemistryInternational Journal of



CatalystsJournal of


adsorption techniques are unable to remove the hazardousfrom the environment On the other side advanced oxidationprocesses (AOPs) such as heterogeneous photocatalytic pro-cesses including photocatalyst Fenton photo-Fenton andelectrooxidation are powerful and nonselective methods thathave been used to convert the persistent organic pollutantsto an environmental friendly product [7ndash9] Among thevarious AOPs methods the heterogeneous photocatalyticprocess has been succeed due to its ability for destroyinga wide range of the pollutants at ambient temperature andpressure without generation of harmful intermediates [10ndash17] The processes use a catalyst that is active under UVor visible-light irradiation to generate hydroxyl radical [18]Zinc oxide (ZnO) is well-known nontoxic semiconductormaterials that has been used as heterogeneous photocatalystto investigate water purification [19 20] ZnO has facilitatedseveral degradations of thewater organic contaminants underUV irradiation [21 22] In addition the grate advantages ofZnO are absorption of a large fraction of the solar spectrum(sunlight) which is free and available around the world [23]Sunlight consists of 47 visible-light with wavelength of 400to 700 nmor energy of 177 to 276 eVThus visible-light couldbe an excellent source of energy for the photocatalytic activityHowever the photo activity of ZnO under the energy wasvery low due to its high direct band gap energy 32 eV [19 20]Therefore several methods of synthesis have been examinedto improve the band gap such as transitionmetals doped ZnO[24ndash29] As the absorption spectra red shift has showed the3D orbital of the metals goes between valence band (VB)and conduction band (CB) of the semiconductors in dopingprocess [30 31] It depends on the energy of the 3D orbitalsthey overlap with the VB or CB of the semiconductors [32]For the reason the electrons are excited from VB (the 3DoverlappedVB) to CB (3D overlapped CB) during irradiationprocess On the other hand the energy of manganese (Mn)3D orbitals is very close to the VB of ZnO which easilyoverlap to decrease the 119864119892 [10 17] In our previous workMn doped ZnO was synthesized and applied for degrada-tion of organic pollutants [29] The photodegradation wasstudied by one variable at a time technique with the effectivevariables of irradiation times pH photocatalyst amount andconcentration of the pollutants [33 34] The problem is thatthe technique varies one of the parameters while the otherterms are kept constant during themultivariate performanceTherefore it has adverse effects on the photodegradationthat should be studied by multivariate methods [17 21]Moreover the kinetic determination of the process is quitecomplicated by consideration of themass transfer the radiantenergy balance the spatial distribution of the absorbedradiation andmechanisms of the photochemical degradation[35] The photodegradation as a process consists of inputfactors such as effective variables and the efficiency as outputresponse The changing amount of the effective variablesaffects the value of the efficiency Therefore the amount ofthe variable could be optimized to achieve the maximumefficiency that is free of the mentioned complexities Theknown multivariate methods that used to optimize theproductive process included response surface methodology(RSM) and artificial neural network (ANN) [35 36]TheRSM

designs the related experiments and then fits the observedresults of performed design to appropriate polynomial andsuggests the qualified model for more validation The modelas a mathematic equation indicates the relationships betweenvariables themselves variables and response(s) Thereafterthe validated model is used to optimize the effective variablesto achieve the maximum yield of the products Howeverthe method is involved with the complicated statisticalcalculation such as analysis of variance fitting and regressionprocess for modeling process [37 38] On the other handANNs have been widely used for modeling of chemicaland biochemical reaction process [39ndash43] The ANN mod-eling has been reliable robust and salient characteristicsin capturing the nonlinear relationship between the inputand output variables which is free of complexities In thiswork the multilayer feed-forward neural network was usedto model m-cresol photodegradation in manganese dopedZnO nanoparticles (Mn doped ZnO NPs) suspension undervisible-light irradiation The cresol is widely used in severalmanufacturing products with high water solubility which hasbeen listed as priority pollutants persistent toxic chemicaland a significant threat to the environment [44 45]The inputeffective operational parameters were including irradiationtime pH photocatalyst amount and concentration of thecresol while the efficiency was the only response as output

2 Materials and Methods

The chemicals of this work were obtained from Merck andwere used without further purification The m-cresol (99)was used as organic water pollutant while H2SO4 (95ndash97)and NaOH (99) were applied to set the appropriate pHThe Mn doped ZnO NPs with average particles size 35 nmbang gap energy 22 eV and surface area 35m2gminus1 wereused as photocatalyst The photocatalyst was synthesized byprecipitation method in absolute alcohol according to ourpublished work [29] To degrade the pollutant the variousconcentrations of m-cresol were mixed with appropriateamount of the photocatalyst in 500mL deionized water Themixture (suspension) solution was irradiated by a Philipslamp (23watts) as light source in a batch homemade pho-toreactor that was used in our previous work [46] Thesuspension was magnetically stirred during the irradiationat 200 rpm Moreover air was blown into the solution byusing an air pump at a flow rate of 150mLmin to increasesolution fluidization access oxygen volatile the produced gas(CO2) and keep the temperature at around 25∘C During theperformance samples were withdrawn from the bulk solu-tion at specific time intervals and centrifuged at 14000 rpmfor 20min and then they filtered through 02 120583m PTFE filterto measure the remained concentration of the m-cresol Themeasurementwas carried out by a ShimadzuUV-1650PCanda TOC-VCSN analyzer respectively In addition the initialcatalyst absorption and photocatalyst were investigated indark and absence of the photocatalyst at normal pH (75)that were considered in the efficiencies calculation [47] Theefficiencies were used as actual responses for modeling of thephotodegradation which was carried out by Neural PowerSoftware version 25 [48 49] The total of 31 experiment





Testing data set17 360 9 15 35 727 7950418 360 763 05 35 502 6402619 360 763 15 41 816 7484520 360 763 13 35 942 8892621 360 763 15 30 992 9770322 360 763 15 25 100 9957223 360 763 25 35 879 9013124 60 763 15 35 1248 17993









119878 =

119899ℎ

sum

119894=1

(119887 minus 119882119894119868119894) (1)


119878 (119861) =

119898

sum

119868=1

[119910119894 minus 119891 (119909119894120573)2] (2)


RMSE = (1119899

119899

sum

119894=1

(119910119894 minus 119910119889119894)2)

12

(3)


1198772= 1 minussum119899

119894=1(119910119894 minus 119910119889119894)

2

sum119899

119894=1(119910119889119894 minus 119910119898)

2(4)

AAD = 1119899

119899

sum

119868=1

1003816100381610038161003816119910119894 minus 1199101198891198941003816100381610038161003816

119910119889119894

(5)

25

4

55

7

0 3 6 9 12 15

RMSE

Node number

QPIBP

BBPLM







0

20

40

60

80

100

0 20 40 60 80 100

Pred

icte

d effi

cien

cy

Actual efficiency

QP-4-8-1

Testing

Best linear fit

R2= 09938

A = P

0

20

40

60

80

100

0 20 40 60 80 100

Pred

icte

d effi

cien

cy

Actual efficiency

BBP-4-6-1

Testing

Best linear fit

R2= 0988

A = P

0

20

40

60

80

100

0 20 40 60 80 100

Pred

icte

d effi

cien

cy

Actual efficiency

IBP-4-15-1

Testing

Best linear fit

R2= 09857

A = P

0

20

40

60

80

100

0 20 40 60 80 100

Pred

icte

d effi

cien

cy

Actual efficiency

LM-4-10-1

Testing

Best linear fit

R2= 09883

A = P







7

85 8693

0

2

4

6

8

10

QP-4-8-1 LM-4-10-1 BBP-4-6-1 IBP-4-15-1

AD

D


0

20

40

60

80

100

0 20 40 60 80 100

Pred

icte

d effi

cien

cy

Actual efficiency

QP-4-8-1

Validation

Best linear fit

R2= 09724

A = P




Bias

Inputs Outputs



119891 (119909) =1

1 + exp (minus119909) (6)







pH

5 567 6 667 7767 8

867 9 967

Effici

ency

()

8575655545352515

Irrad

iation

time (

min)

340300

260220

180140

10060


Catalyst (gL)

05 09 13 17 21 25 29 33

Effici

ency

()

655545352515

5

Irrad

iation

time (

min)

340300

260220

180140

10060





15 231 313 394 475 557 638 719

Effici

ency

()

100

80

60

40

20

Irrad

iation

time (

min)

340300

260220

180140

10060M-cresol (mgL)









M-cresol (mgL)pH


2128

1377

3202

3293



5 Conclusions





Acknowledgments


References










2 Fenton





































































Journal of

Chemistry


Advances in

Physical Chemistry



Journal of

Volume 2014














Journal of

Spectroscopy



Journal of


Quantum Chemistry






CatalystsJournal of





Testing data set17 360 9 15 35 727 7950418 360 763 05 35 502 6402619 360 763 15 41 816 7484520 360 763 13 35 942 8892621 360 763 15 30 992 9770322 360 763 15 25 100 9957223 360 763 25 35 879 9013124 60 763 15 35 1248 17993









119878 =

119899ℎ

sum

119894=1

(119887 minus 119882119894119868119894) (1)


119878 (119861) =

119898

sum

119868=1

[119910119894 minus 119891 (119909119894120573)2] (2)


RMSE = (1119899

119899

sum

119894=1

(119910119894 minus 119910119889119894)2)

12

(3)


1198772= 1 minussum119899

119894=1(119910119894 minus 119910119889119894)

2

sum119899

119894=1(119910119889119894 minus 119910119898)

2(4)

AAD = 1119899

119899

sum

119868=1

1003816100381610038161003816119910119894 minus 1199101198891198941003816100381610038161003816

119910119889119894

(5)

25

4

55

7

0 3 6 9 12 15

RMSE

Node number

QPIBP

BBPLM







0

20

40

60

80

100

0 20 40 60 80 100

Pred

icte

d effi

cien

cy

Actual efficiency

QP-4-8-1

Testing

Best linear fit

R2= 09938

A = P

0

20

40

60

80

100

0 20 40 60 80 100

Pred

icte

d effi

cien

cy

Actual efficiency

BBP-4-6-1

Testing

Best linear fit

R2= 0988

A = P

0

20

40

60

80

100

0 20 40 60 80 100

Pred

icte

d effi

cien

cy

Actual efficiency

IBP-4-15-1

Testing

Best linear fit

R2= 09857

A = P

0

20

40

60

80

100

0 20 40 60 80 100

Pred

icte

d effi

cien

cy

Actual efficiency

LM-4-10-1

Testing

Best linear fit

R2= 09883

A = P







7

85 8693

0

2

4

6

8

10

QP-4-8-1 LM-4-10-1 BBP-4-6-1 IBP-4-15-1

AD

D


0

20

40

60

80

100

0 20 40 60 80 100

Pred

icte

d effi

cien

cy

Actual efficiency

QP-4-8-1

Validation

Best linear fit

R2= 09724

A = P




Bias

Inputs Outputs



119891 (119909) =1

1 + exp (minus119909) (6)







pH

5 567 6 667 7767 8

867 9 967

Effici

ency

()

8575655545352515

Irrad

iation

time (

min)

340300

260220

180140

10060


Catalyst (gL)

05 09 13 17 21 25 29 33

Effici

ency

()

655545352515

5

Irrad

iation

time (

min)

340300

260220

180140

10060





15 231 313 394 475 557 638 719

Effici

ency

()

100

80

60

40

20

Irrad

iation

time (

min)

340300

260220

180140

10060M-cresol (mgL)









M-cresol (mgL)pH


2128

1377

3202

3293



5 Conclusions





Acknowledgments


References










2 Fenton





































































Journal of

Chemistry


Advances in

Physical Chemistry



Journal of

Volume 2014














Journal of

Spectroscopy



Journal of


Quantum Chemistry






CatalystsJournal of



119878 =

119899ℎ

sum

119894=1

(119887 minus 119882119894119868119894) (1)


119878 (119861) =

119898

sum

119868=1

[119910119894 minus 119891 (119909119894120573)2] (2)


RMSE = (1119899

119899

sum

119894=1

(119910119894 minus 119910119889119894)2)

12

(3)


1198772= 1 minussum119899

119894=1(119910119894 minus 119910119889119894)

2

sum119899

119894=1(119910119889119894 minus 119910119898)

2(4)

AAD = 1119899

119899

sum

119868=1

1003816100381610038161003816119910119894 minus 1199101198891198941003816100381610038161003816

119910119889119894

(5)

25

4

55

7

0 3 6 9 12 15

RMSE

Node number

QPIBP

BBPLM







0

20

40

60

80

100

0 20 40 60 80 100

Pred

icte

d effi

cien

cy

Actual efficiency

QP-4-8-1

Testing

Best linear fit

R2= 09938

A = P

0

20

40

60

80

100

0 20 40 60 80 100

Pred

icte

d effi

cien

cy

Actual efficiency

BBP-4-6-1

Testing

Best linear fit

R2= 0988

A = P

0

20

40

60

80

100

0 20 40 60 80 100

Pred

icte

d effi

cien

cy

Actual efficiency

IBP-4-15-1

Testing

Best linear fit

R2= 09857

A = P

0

20

40

60

80

100

0 20 40 60 80 100

Pred

icte

d effi

cien

cy

Actual efficiency

LM-4-10-1

Testing

Best linear fit

R2= 09883

A = P







7

85 8693

0

2

4

6

8

10

QP-4-8-1 LM-4-10-1 BBP-4-6-1 IBP-4-15-1

AD

D


0

20

40

60

80

100

0 20 40 60 80 100

Pred

icte

d effi

cien

cy

Actual efficiency

QP-4-8-1

Validation

Best linear fit

R2= 09724

A = P




Bias

Inputs Outputs



119891 (119909) =1

1 + exp (minus119909) (6)







pH

5 567 6 667 7767 8

867 9 967

Effici

ency

()

8575655545352515

Irrad

iation

time (

min)

340300

260220

180140

10060


Catalyst (gL)

05 09 13 17 21 25 29 33

Effici

ency

()

655545352515

5

Irrad

iation

time (

min)

340300

260220

180140

10060





15 231 313 394 475 557 638 719

Effici

ency

()

100

80

60

40

20

Irrad

iation

time (

min)

340300

260220

180140

10060M-cresol (mgL)









M-cresol (mgL)pH


2128

1377

3202

3293



5 Conclusions





Acknowledgments


References










2 Fenton





































































Journal of

Chemistry


Advances in

Physical Chemistry



Journal of

Volume 2014














Journal of

Spectroscopy



Journal of


Quantum Chemistry






CatalystsJournal of


0

20

40

60

80

100

0 20 40 60 80 100

Pred

icte

d effi

cien

cy

Actual efficiency

QP-4-8-1

Testing

Best linear fit

R2= 09938

A = P

0

20

40

60

80

100

0 20 40 60 80 100

Pred

icte

d effi

cien

cy

Actual efficiency

BBP-4-6-1

Testing

Best linear fit

R2= 0988

A = P

0

20

40

60

80

100

0 20 40 60 80 100

Pred

icte

d effi

cien

cy

Actual efficiency

IBP-4-15-1

Testing

Best linear fit

R2= 09857

A = P

0

20

40

60

80

100

0 20 40 60 80 100

Pred

icte

d effi

cien

cy

Actual efficiency

LM-4-10-1

Testing

Best linear fit

R2= 09883

A = P







7

85 8693

0

2

4

6

8

10

QP-4-8-1 LM-4-10-1 BBP-4-6-1 IBP-4-15-1

AD

D


0

20

40

60

80

100

0 20 40 60 80 100

Pred

icte

d effi

cien

cy

Actual efficiency

QP-4-8-1

Validation

Best linear fit

R2= 09724

A = P




Bias

Inputs Outputs



119891 (119909) =1

1 + exp (minus119909) (6)







pH

5 567 6 667 7767 8

867 9 967

Effici

ency

()

8575655545352515

Irrad

iation

time (

min)

340300

260220

180140

10060


Catalyst (gL)

05 09 13 17 21 25 29 33

Effici

ency

()

655545352515

5

Irrad

iation

time (

min)

340300

260220

180140

10060





15 231 313 394 475 557 638 719

Effici

ency

()

100

80

60

40

20

Irrad

iation

time (

min)

340300

260220

180140

10060M-cresol (mgL)









M-cresol (mgL)pH


2128

1377

3202

3293



5 Conclusions





Acknowledgments


References










2 Fenton





































































Journal of

Chemistry


Advances in

Physical Chemistry



Journal of

Volume 2014














Journal of

Spectroscopy



Journal of


Quantum Chemistry






CatalystsJournal of


7

85 8693

0

2

4

6

8

10

QP-4-8-1 LM-4-10-1 BBP-4-6-1 IBP-4-15-1

AD

D


0

20

40

60

80

100

0 20 40 60 80 100

Pred

icte

d effi

cien

cy

Actual efficiency

QP-4-8-1

Validation

Best linear fit

R2= 09724

A = P




Bias

Inputs Outputs



119891 (119909) =1

1 + exp (minus119909) (6)







pH

5 567 6 667 7767 8

867 9 967

Effici

ency

()

8575655545352515

Irrad

iation

time (

min)

340300

260220

180140

10060


Catalyst (gL)

05 09 13 17 21 25 29 33

Effici

ency

()

655545352515

5

Irrad

iation

time (

min)

340300

260220

180140

10060





15 231 313 394 475 557 638 719

Effici

ency

()

100

80

60

40

20

Irrad

iation

time (

min)

340300

260220

180140

10060M-cresol (mgL)









M-cresol (mgL)pH


2128

1377

3202

3293



5 Conclusions





Acknowledgments


References










2 Fenton





































































Journal of

Chemistry


Advances in

Physical Chemistry



Journal of

Volume 2014














Journal of

Spectroscopy



Journal of


Quantum Chemistry






CatalystsJournal of


pH

5 567 6 667 7767 8

867 9 967

Effici

ency

()

8575655545352515

Irrad

iation

time (

min)

340300

260220

180140

10060


Catalyst (gL)

05 09 13 17 21 25 29 33

Effici

ency

()

655545352515

5

Irrad

iation

time (

min)

340300

260220

180140

10060





15 231 313 394 475 557 638 719

Effici

ency

()

100

80

60

40

20

Irrad

iation

time (

min)

340300

260220

180140

10060M-cresol (mgL)









M-cresol (mgL)pH


2128

1377

3202

3293



5 Conclusions





Acknowledgments


References










2 Fenton





































































Journal of

Chemistry


Advances in

Physical Chemistry



Journal of

Volume 2014














Journal of

Spectroscopy



Journal of


Quantum Chemistry






CatalystsJournal of




M-cresol (mgL)pH


2128

1377

3202

3293



5 Conclusions





Acknowledgments


References










2 Fenton





































































Journal of

Chemistry


Advances in

Physical Chemistry



Journal of

Volume 2014














Journal of

Spectroscopy



Journal of


Quantum Chemistry






CatalystsJournal of




































































Journal of

Chemistry


Advances in

Physical Chemistry



Journal of

Volume 2014














Journal of

Spectroscopy



Journal of


Quantum Chemistry






CatalystsJournal of

artificial neural network modelling of photodegradation … · oxide nanoparticles under...

Documents