支持向量机在牵引变流器故障模式分类中的应用分析-application analysis of support vector machine in traction converter fault mode classification.docxVIP

AbstractAspowerconversiondevice,tractionconverterisoneofthekeypartsinlocomotive.Tractionconverteriscontinuouslyonworkduringtheprocessoftrainoperation,whichwillresultinfaultsinevitably.Also,tractionconverterisofhighenergydensity,andthesemiconductordevicesarerelatively‘weak’,makingtractionconverterbethehigh-frequency-faultedpartintractionsystem.Tractionconverter’sfailurewilldirectlydislocaterailwayservicesandevencausecasualties.Asaresult,malfunctiondiagnosisfortractionconverteriscrucialtomeetthedemandformaintenanceandguaranteethesafeoperationofthelocomotive.Thestate-of-the-artinmalfunctiondiagnosisisfirstlyreviewedinthisthesis.SupportVectorMachine(SVM),adata-basedmalfunctiondiagnosisapproach,isrecommendedfortractionconverter’smalfunctiondiagnosis.TherearetwomainfactorsaffectingtheclassificationperformanceofSVM:kernelfunctionandparametersoptimization.RadialBiasFunctionisselectedtobetheSVM’skernelfunctionaftercomparingtheimpactofdifferentkindsofkernelfunctionsusedinSVMforbinaryclassificationperformance;grid-searchbasedoncross-validationisrecommendedbeingSVMsparameteroptimizationalgorithmaftercomparingtheimpactofdifferentkindsofparameteroptimizationalgorithmsusedinSVMformulti-classificationperformance.C-SVMwithselectedkernelfunctionandparameteroptimizationalgorithmisutilizedfortractionconverter’smalfunctionclassification,andalsoitsclassificationresultiscomparedwiththatofusingleastsquareSVM(LS-SVM).Studyandsimulationresultshowsthat,it’sentirelyeffectivetoapplySVMwithRBFkernelfunctionandcross-validationbasedgrid-searchoptimizationtotractionconverter’smalfunctionclassification,anditsgeneralizationperformanceismuchbetterthanthatofLS-SVMunderrelativelygreatsampledata.Thispaperestablishesthefoundationforfurtherstudyofmalfunctiondiagnosistechnologyfortractionconverter,andalsobeinstructiveandmeaningfultofurtherresearchofSVMmethod.Keywords:SupportVectorMachine,Kernelfunction,parameteroptimization,tractionconverter,malfunctionclassification目录摘要.....................................