基于证据推理的电力变压器故障诊断策略

第26卷第1期中国电机工程学报Vol.26No.1Jan.20062006年1月ProceedingsoftheCSEE©2006Chin.Soc.forElec.Eng.文章编号：0258-8013（2006）01-0106-09中图分类号：TM42文献标识码：A学科分类号：470⋅40基于证据推理的电力变压器故障诊断策略董明1，严璋1，杨莉2，M.D.Judd3（1．西安交通大学电气工程学院，陕西省西安市710049；2．卡里多尼亚大学高电压绝缘诊断研究所，英国格拉斯哥；3．斯特拉思克莱德大学能源与环境学院，英国格拉斯哥）AnEvidentialReasoningApproachtoTransformerFaultDiagnosisDONGMing1,YANZhang1,YANGLi2,M.D.Judd3（1.SchoolofElectricalEngineering,Xi’anJiaotongUniversity,Xi’an710049,ShaanxiProvince,China;2.HighVoltageInsulationDiagnosticsGroup,GlasgowCaledonianUniversity,Glasgow,Scotland,UK;3.InstituteforEnergyandEnvironment,UniversityofStrathclyde,Glasgow,Scotland,UK）ABSTRACT:Methodsusedtoassesstheinsulationstatusofpowertransformersbeforetheydeterioratetoacriticalstateincludedissolvedgasanalysis(DGA),partialdischarge(PD)detectionandtransferfunctiontechniques,etc.Alloftheseapproachesrequireexperienceinordertocorrectlyinterprettheobservations.ArtificialIntelligence(AI)isincreasinglyusedtoimproveinterpretationoftheindividualdatasets.However,asatisfactorydiagnosismaynotbeobtainedifonlyonetechniqueisused.Forexample,theexactlocationofPDcannotbepredictedifonlyDGAisperformed.However,usingdiversemethodsmayresultindifferentdiagnosissolutions,aproblemthatisaddressedthroughtheintroductionofafuzzyinformationinfusionmodel.Aninferenceschemeisproposedthatyieldsconsistentconclusionsandmanagestheinherentuncertaintyinthevariousmethods.Withtheaidofinformationfusion,aframeworkisestablishedthatallowsdifferentdiagnostictoolstobecombinedinasystematicway.Theapplicationofinformationfusiontechniqueforinsulationdiagnosticsoftransformeriseffectivebymeansofexamples.KEYWORDS:Powertransformers;Conditionmonitoring;Dissolvedgasanalysis(DGA);Informationfusion;Insulationdiagnostics摘要：在变压器绝缘劣化之前，可以进行油中溶解气体分析、局部放电检测、传递函数测量等试验方法对其状态进行评估。所有这些试验现象需要很多实际经验才能正确解释。因此人工智能技术逐渐被应用于提高单一试验数据的分析中。但是，仅使用一种方法，可能难以得到满意的诊断结果，如基金项目：国家自然科学基金项目（59637200）。ProjectSupportedbyNationalNaturalScienceFoundationofChina(59637200).油中溶解气体分析是不能准确对局部放电进行定位。然而，应用不同的方法可能产生各异的诊断结果，因此文中引入模糊信息融合系统来解决此问题，提出了产生一致性结论和处理不同方法中不确定性的证据推理策略。并在信息融合的帮助下，建立了有机组合多种诊断方法系统框架。通过实例证明，基于信息融合的变压器绝缘故障诊断方法是有效的。关键词：电力变压器；状态监测；油中溶解气体分析；信息融合；绝缘诊断学1INTRODUCTIONPowertransformersarekeycomponentsinelectricitytransmissionanddistributionsystems.Manydifferenttechniquescanbeusedtodetectmalfunctionsofpowertransformerinsulation.Thesemethodsallhavearoletoplayinestablishingtheinsulationconditionofpowertransformers.However,fewofthesemethodscan,inisolation,providealloftheinformationthatthetransformeroperatormightrequiretodecideuponacourseofaction.Moreover,notallofthemethodscanbeapplieduniversally.Frompracticalstandpoint,afterwerecognizetheexistenceofafaultandassessitsseverity,wearelikelytoknowitslocationwithinthetransformer,whichisusefulforguidingfurthermaintenanceandrepair.Asmoremethodsareintroducedintothefaultdiagnosisprocess,moredetailandmorereliableresultsshouldbeexpected.Hence,inaccordancewiththebasicframeworkofinformationfusion,aPDF文件使用pdfFactoryPro试用版本创建www.fineprint.com.cn第1期董明等：基于证据推理的电力变压器故障诊断策略107multi-levelcomprehensivefaultdecisionmodelisproposed.TheDGAandtheresultsofotherelectricaltestsofpowertransformersarecombinedefficientlyinthemodel.Inaddition,themodeltakesintoaccounton-siteexperiencesofoperationandmaintenanceinreachingitsdiagnosis.2TRANSFORMERDIAGNOSTICMET-HODSDGAisasimple,effective,cheapandnon-intrusivemethodthathasbeenwidelyappliedtothefaultdiagnosisofoil-immersedtransformers.Tab.1showsIEC/IEEEcriteriainwhichoverheatinganddischargingfaultsarecategorized[1].Withthedevelopmentofcomputerscience,artificialintelli-gence(AI)isincreasinglyintroducedtodiagnosefaultytransformersbyanalyzingthegases.Techni-quesavailableincludeexpertsystems[2],fuzzylogic[3],evolutionaryalgorithm(EA)[4]andtheartificialneuralnetwork(ANN)[5].Furthermore,bodiessuchasutilities,transformermanufacturers,testinglaboratoriesandconsultantshavedevelopedadditional,moreadvancedtestingmethods.Forexample,furanscanbeusedasindicatorfortheagingstatusofthesolidinsulation[6];acousticandUHFsensorsareemployedtodetectandsubsequentlylocatePD[7];frequencyresponsemeasurement[8]andrecoveryvoltagemeasurement[9]areappliedtodetectdamagedwindings.Eachtechniquehasitsstrengthsandweaknesses.Ashortcomingoftheplethoraoftechniquesavailableisthatalthoughtheyweredevelopedtogivemorespecificresults,thediagnosisobtainedissometimesnotcomparable表1IEC溶解气体比值法Tab.1IEC/IEEEcodesforinterpretationofDGAmethodNoClassificationofFaultType2224(CH)(CH)jj42(CH)(H)jj2226(CH)(CH)jj0Nofault0.10.1~1.01.01Lowenergypartialdischarges0.10.11.02Highenergypartialdischarges0.1~3.00.11.03Lowenergydischarges0.1~3to30.1~1.01~3to34Highenergydischarges0.1~30.1~1.03.05150ºCthermalfault0.10.1~1.01.0~3.06150-300ºCthermalfault0.11.01.07300-700ºCthermalfault0.11.01.0~3.08700ºCthermalfault0.11.03.0acrosstechniques,e.g.,DGAisatthistimenotascience,butanartsubjecttovariability[1].Nevertheless,whenanindividualmethodindicatesaproblem,theevidencemayoftenbeconfirmedorenhancedbyothermethods.Forexample,whenDGAindicatesPD,acousticorUHFsensorscouldb