基于粒子群神经网络的桥墩冲刷深度预测(IJISA-V11-N11-4)

I.J.IntelligentSystemsandApplications,2019,11,38-47PublishedOnlineNovember2019inMECS()DOI:10.5815/ijisa.2019.11.04Copyright©2019MECSI.J.IntelligentSystemsandApplications,2019,11,38-47ApplicationofParticleSwarmbasedNeuralNetworktoPredictScourDeptharoundtheBridgePierSreedharaBMAssistantProfessor,BMSInstituteofTechnologyandManagement,Bengaluru-560064,IndiaE-mail:[e-mail]tProfessor,BMSCollegeofEngineering,Bengaluru-560019,IndiaE-mail:[e-mail]r,NationalInstituteofTechnologyKarnatakaSurathkal,Mangaluru–575025,IndiaE-mail:[e-mail],PresidencyUniversity,Bengaluru–560085,IndiaE-mail:[e-mail]:29July2018;Accepted:12August2019;Published:08November2019Abstract—Scouraroundthebridgepierisoneofthemajorfactorswhichaffectthesafetyandstabilityofthebridgestructure.Duetothepresenceofcomplexityinthescourmechanism,thereisnocommonandsimplemethodtoestimatethescourdepth.ThepresentpapergivesanideaofhybridizingtwotechniquessuchasanartificialneuralnetworkwithswarmintelligencetechniqueparticleswarmoptimizationtoestimatethescourdeptharoundthebridgepierandabbreviatedasPSO-ANN.Thepresentdiscussioncoverstheestimationofscourdepthforclearwaterandlivebedscourconditionaroundcircularandrectangularpiershapes.Theindependentvariables,Sedimentsize(d50),sedimentquantity(Sq),velocity(u)andtime(t)areusedasinputtodevelopthemodelstoestimateorquantifyadependentvariablescourdepth(ds).TheefficiencyandaccuracyofthemodelaremeasuredusingmodelperformancesindicatorssuchasCorrelationCoefficient(CC),NormalizedRootMeanSquareError(NRMSE),NashSutcliffeError(NSE),andNormalizedMeanBias(NMB).Thepredictedresultsofboththemodelsarecomparedwitheachotherandalsocomparedwithmeasuredscourdepth.ThestudyconcludesthattheproposedPSO-ANNmodelissuitabletoestimatethescourdepthinboththecasesforcircularandrectangularpiershapes.IndexTerms—Pierscour,clearwatercondition,livebedcondition,ParticleSwarmOptimization-ArtificialNeuralNetwork(PSO-ANN).I.INTRODUCTIONScouristhecomplexphenomenonoccursatthestructurebaselocatedacrosstheflowingwater.Later,loweringoftheriverbedleadstotheexposureofbridgefoundation.Scourmechanismoccurswhenthenormalflowinteractswiththeobstacleanddevelopslargescaleeddiesatthestructurebase.Thisleadstotheformationofhorseshoesandwakevortexatthebaseanddownstreamofthestructurerespectively.Thisvortexsystemleadstotheerosionandformsscourhole.ThetypicalscourmechanismisillustratedinFig.1.Fig.1.Mechanismoflocalscour[1]ApplicationofParticleSwarmbasedNeuralNetworktoPredictScourDeptharoundtheBridgePier39Copyright©2019MECSI.J.IntelligentSystemsandApplications,2019,11,38-47Basedonthecriteria,characteristicsofriverbedmaterial,scourhappensundertwostatesofclearwaterandlivebedscour.Thetypeonecaseofclearwaterscourhappensintheabsenceofbedmaterialmovementintheupstreamoftheintersection.Duetotheobstacles(piers,abutments)intheflow.Thescourholecontinuestogrowuntiltheequilibriumscour-depthisreached.Inthetypetwocaseoflive-bedscourhappensduetoaconsistentsupplyofsedimentfromtheupstreamtothescourhole.Therateofsedimentsupplytothescourholeequaltotherateofsedimenttransportedoutofthescourholeandthenequilibriumofscourdepthachieved.II.RELATEDWORKSAsidefromexperimentalstudiesandempiricalmodels,inlatedecadessoftcomputingtechniques,artificialneuralnetwork(ANN),adaptiveNeuroinferencesystem(ANFIS)[2–4],supportvectorregression(SVR)[5,6],geneticalgorithm(GA)[7],groupmethod-datahandling(GMDH)[8]etc.,arebeingusedwidelytopredictthescourdepthusingexperimentalvalues.Theindividualmodelshavebeenappliedbyvariousresearcherstoestimatescourdeptharoundpiers[9,10],riskassessmentforstructuremaintenance[11]andscourbelowspillways[12].Inrecentyears,thecombinedeffectoftheevolutionaryalgorithmisdevelopedusingoptimizationtechnique,particleswarmoptimization(PSO)[2]withANNasanemergingtoolinvariousfields.The(PSO-ANN)techniqueisusedtoidentifythesurfacesettlementduetotunneling[13],tocheckthefloatingtypebreakwaterefficiency[14],topredictwavetransmissionoftandembreakwater[15]andfaultpredictionofobject-orientedsystems[16].ThehighlightofthepresentpaperistochecktheapplicabilityofPSO-ANNmodeltoestimatethescourdepthacriticalparameter,wheretheparametersofANNmodelsareoptimizedusingPSOtoimprovethemodelefficiencyofANNmodel.Fromthepaststudy,itisnoticedthatnoneofthestudieshasbeencarriedoutusingPSO-ANNmodeltopredictscourdepth.Hence,aneffortismadetopredictthescourdeptharoundabridgepierusingPSO-ANNhybridtechnique.Theexperimentaldataonscourdeptharoundcircularandrectangularpierareobtainedundertwodifferentconditionareutilizedtodevelopthesoftcomputingmodelswithdifferentcombinations.Theestimatedscourdeptharecomparedwithexperimentalresultsintermsofstatisticalparameters.III.EXPERIMENTALDATACLASSIFICATIONGoswamiPankaj[17]conductedthetestsonscourdeptharoundthecircularandrectangularpierinclearwaterandlivebedcondition.Theflumeusedis1.0mwide,1.3mdepthand19.25minlength.Thetwotestconditions,forexample,clearwater,andlivebedco