预览加载中，请您耐心等待几秒...

基于BDKPCA-OCSVM的间歇过程故障检测.docx

预览
1/2
2/2

在线预览结束，喜欢就下载吧，查找使用更方便

下载提示 文本预览

如果您无法下载资料，请参考说明：

1、部分资料下载需要金币，请确保您的账户上有足够的金币

2、已购买过的文档，再次下载不重复扣费

3、资料包下载后请先用软件解压，在使用对应软件打开

基于BDKPCA-OCSVM的间歇过程故障检测 Introduction Intermittentprocessesarecharacterizedbyrandomfluctuationsintheirbehaviorandmayexhibitintermittentfaults.Thesefaultsmayoccurperiodicallyorrandomlyandcanleadtosystemfailures.Hence,itisessentialtodevelopareliableandeffectivemethodtodetectthesefaults.Inrecentyears,machinelearningalgorithmshavebeenusedtoenhancethefaultdetectionprocess.Inthispaper,weproposeanintermittentprocessfaultdetectionmethodbasedonBDKPCA-OCSVM. BDKPCA-OCSVM BDKPCA-OCSVMisacombinationoftwomachinelearningalgorithms:BalancedDissimilarityKPCA(BDKPCA)andOne-ClassSupportVectorMachine(OCSVM).BDKPCAisakernelprincipalcomponentanalysisalgorithm,whileOCSVMisabinaryclassificationalgorithm.Thecombinationofthesetwoalgorithmscanimprovefaultdetectioninintermittentprocesses.TheBDKPCAalgorithmreduceshigh-dimensionalinputdatatoalowerdimension,whiletheOCSVMalgorithmdetectsoutliersinthereduceddata. One-ClassSupportVectorMachine(OCSVM) OCSVMisamachinelearningalgorithmsuitablefordetectingoutliersinhigh-dimensionaldatasets.Itfindsahyperplanethatseparatesthedatafromtheorigin,andtheoutliersfalloutsidethehyperplane.OCSVMtrainsmodelsondatawithoutlabels,soitisanunsupervisedlearningalgorithmusedmainlyforanomalydetection.Thealgorithm'sprimaryadvantagesareitsabilitytohandlehigh-dimensionaldataanditsabilitytodetectoutliersaccurately. BalancedDissimilarityKPCA(BDKPCA) BDKPCAisavariationofKernelPrincipalComponentAnalysis(KPCA),whichhastheabilitytohandledatasetswithimbalancedclasses.KPCAisadimensionalityreductiontechniqueusedfornon-lineardata.BDKPCAusesadissimilaritymeasureinsteadofthestandardEuclideandistancetoaccountfortheimbalancedclassdistribution.ThedissimilaritymeasureallowsBDKPCAtoreducetheimportanceofthemajorityclassinstancesinadataset. IntermittentFaultDetection Intermittentfaultscanbedefinedastemporarychangesintheprocessbehaviorthatdonotleadtoanimmediatefailure.Thesefaultscanberandomorperiodicandcanresultinsystemfailuresifnotdetectedandcorrectedearly.Intermittentfaultsarecommoninmanyproce