SystemReliabilityEvaluationUsingMonteCarlo&SupportVectorMachine
Reliabilityevaluationofrealengineeringsystemsisofteperformedusingsimulationtools.Ingeneral,todeterminethstateofthesystem(operatingorfailed)asafunctionofit
components,itisnecessarytoevaluateafunctionthatiscalleSystemFunctionorStructureFunction(SF).InaMonteCarlsimulationsystemreliabilityisevaluatedbygeneratinga
randomseveralsystemsstatesandevaluatingtheSF.SincelargenumberofSFevaluationsarerequired,itcouldbenecessarytosubstitutethoseevaluationswithafast
approximatedalgorithm.Severalapproacheshavebeenusedtoaddressthedefinitiono
theseapproximatedalgorithms.ThispaperdealswiththfeasibilityofusingSupportVectorMachine(SVM)tobuilempiricalmodelsforuseinreliabilityevaluation.TheapproachtakesadvantageofthespeedofSVMinthnumerousmodelcalculationstypicallyrequiredtoperform
MonteCarloreliabilityevaluation.Themainideaistdevelopanestimationalgorithm,bytrainingamodelonrestricteddataset,andreplaceSFevaluationbyasimple
calculation,whichprovidesreasonablyaccuratemodeoutputs.Intheexamplepresented,relatedtoanelectricpowesystem,theSVMbuiltfromaverysmallfractionofthetota
statespace,producesveryclosereliabilityestimation. |