电气专业英文文献

AnExpertSystemforTransformerFaultDiagnosisUsingDissolvedGasAnalysis

1.INTRODUCTION

Thepowertransformerisamajorapparatusinapowersystem,anditscorrectfunctioningitsvitaltominimizesystemoutages,manydeviceshaveevolvedtomonitortheserviceabilityofpowertransformers.Thesedevices,suchas,Buchholzrelaysordifferentialrelays,respondonlytoaseverepowerfailurerequiringimmediateremovalofthetransformerfromservice,inwhichcase,outagesareinevitable.Thus,preventivetechniquesforearlydetectionfaultstoavoidoutageswouldbevaluable.Inthisway,analysisofthemixtureofthefaultygasesdissolvedininsulationoilofpowertransformerhasreceivedworldwiderecognitionasaneffectivemethodforthedetectionofoncipientfaults.ManyresearchersandelectricalutilitieshavereportedontheirexperienceanddevelopedinterpretativecriteriaonthebasisofDGA.However,criteriatendtovaryfromutilitytoutility.Therefore,transformerdiagnosisisstillintheheuristicstage.Forthisreason,knowledge-basedprogrammingisasuitableapproachtoimplementinsuchadiagnosticproblem.

BasedontheinterpretationofDGA,aprototypeofanexpertsystemfordiagnosisofsuspectedtransformerfaultsandtheirmaintenanceproceduresisproposed.Thesignificantsourceinthisknowledgebaseisthegasratiomethod.Somelimitationsofthisapproachareovercomebyincorporatingthediagnosticprocedureandthesyntheticexpertisemethod.Furthermore,databasesadoptedfromTPC'Sgasrecordsoftransformersareincorporatedintotheexpertsystemtoincreasethepracticalperformance.Uncertaintyofdiagnosisismanagedbyusingfuzzysetconcepts.Thisexpertsystemisconstructedwithrulebasedknowledgerepresentation,sinceitcanbeexpressedbyexperts.Theexpertsystembuildingtool,knowledgeEngineeringSystem(KES),isusedinthedevelopmentoftheknowledgesystembecause,ithasexcellentman-machineinterfacethatprovidessuggestions.Moreover,itsinferencestrategyissimilartotheMYCIN.Afamousrule-basedexpertsystemusedformedicaldiagnosis.Theuncertaintyofhumanqualitativediagnosticexpertise,e.g.,keygasanalysis,andanotherquantitativeimprecision,suchas,normsthresholdandgasratioboundariesetc.,aresmoothedbyappropriatefuzzymodels.Withtheresultsofsuchimplementation,differentcertaintyfactorswillbeassignedtothecorrespondingexpertisevariables.Bothevent-driven(forwardchaining)andgoal-driven(backwardchaining)inferencesareusedintheinferenceenginetoimprovetheinferenceefficiency.Todemonstratethefeasibilityoftheproposedexpertsystem,aroundhundredsofTPChistoricalgasrecordshavebeentested.Itisfoundthatmoreappropriatefaultytypesandmaintenancesuggestionscansupportthemaintenancepersonalstoincreasetheperformanceoftransformerdiagnosis.

2.DEVELOPMENTOFDIAGNOSISANDINTERPRETATION

Likemanydiagnosticproblems,diagnosisofoil-immersedpowertransformerisaskilledtask.Atransformermayfunctionwellexternallywithmonitors,whilesomeincipientdeteriorationmayoccurinternallytocauseafatalprobleminthelatterdevelopment.AccordingtoaJapaneseexperience,nearly80%ofallfaultsresultfromincipientdeteriorations.Therefore,faultsshouldbeidentifiedandavoidedattheearliestpossiblestagebysomepredictivemaintenancetechnique.DGAisoneofthemostpopulartechniquesforthisproblem.Faultgasesintransformersaregenerallyproducedbyoildegradationandotherinsulatingmaterial,e.g.,celluloseandpaper.Theoretically,ifanincipientoractivefaultispresent,theindividualdissolvedgasconcentration,gassingrate,totalcombustiblegas(TCG)andcellulosedegradationareallsignificantlyincreased.Byusinggaschromatographytoanalysethegasdissolvedinatransformer'sinsulatingoil,itbecomesfeasibletojudgetheincipientfaulttypes.Thisstudyisconcernedwiththefollowingrepresentativecombustiblegases;hydrogen(H2),methane(C2H2),ethane(C2H6),ethylene(C2H2)andcarbonmonoxide(C0).

ManyinterpretativemethodsbasedonDGAtothenatureofincipientdeteriorationhavebeenreported.Evenundernormaltransformeroperationalconditions,someofthesegasesmaybeformedinside.Thus,itisnecessarytobuildconcentrationnormsfromasufficientlylargesamplingtoassessthestatistics.TPCinvestigatedgasdatafrompowertransformerstoconstructitscriteria.Thedeveloped

Theproposeddiagnosticexpertsystemiscomposedofcomponents,workingmemory,aknowledgebase,aninferenceengineandaman-machineinterface.Workingmemory(globaldatabase)containsthecurrentdatarelevanttosolvethepresentproblem.Inthisstudy,mostofthediagnosticvariablesstoredinthedatabasearecurrentgasconcentration,somearefromtheuser,othersareretrievedfromthetransformer'shistoricaldatabase.Notethatthefuzzysetconceptisincorporatedtocreatefuzzyvariablesontherequestofsystemreasoning.Aknowledgerelationship,whichusesthesefacts,asthebasisfordecisionmaking.TheproductionruleusedinthissystemisexpressedinIF-THENforms.Asuccessfulexpertsystemdependsonahighqualityknowledgebase.Forthistransformerdiagnosticsystem,theknowledgebaseincorporatessomepopularinterpretativemethodsofDGA,syntheticexpertisemethodandheuristicmaintenancerules.Section3.4willdescribethisknowledgebase.Anotherspecialconsiderationintheexpertsystemisitsinferenceengine.Theinferenceenginecontrolsthestrategiesofreasoningandsearchingforappropriateknowledge.Thereasoningstrategyemploysbothforwardchaining(data-driven)andbackwardchaining(goal-driven).Fuzzyrules,normsrules,gasratiorules,syntheticexpertiserulesandsomeofthemaintenancerulesandsomemaintenancerules,useforwardchaining.

AsforthesearchingstrategyinKES,thedepthfirstsearchingandshort-circuitevaluationareadopted.Theformercanimprovethesearchefficiencybyproperlyarrangingthelocationofsignificantrulesintheinferenceprocedures.Thelatterstrategyonlysearchesthekeyconditionalstatementsintheantecedentthatareresponsibleforestablishingwhethertheentireruleistrueorfalse.Takingtheadvantagesofthesetwoapproachesinthebuildingandstructuringofaknowledgebaseimprovesinferenceefficiencysignificantly.

Asforman-machineinterface.KEShasaneffectiveinterfacewhichisbetterthantypicalknowledgeprogramminglanguages,suchas,PROLOGorLISP.Withthehelpofthisinterface,thecapabilityoftracing,explainingandtraininginanexpertsystemisgreatlysimplified.

IMPLEMENTATIONOFTHEPROPOSEDEXPERTSYSTEM

Anexpertsystemisdevelopedbasedontheproposedinterpretativerulesanddiagnosticproceduresoftheoverallsystem.Todemonstratethefeasibilityofthisexpertsystemindiagnosis,thegasdatasupportedbyMTLofTPChavebeentested.InTaiwan,theMTLofTPCperformstheDGAandsendstheresultstoallactingdivisionsrelatingtopowertransformers.Inreturn,theseactingdivisionsarerequestedtocollectandsupplytheirtransformeroilsamplesperiodically.

Afteranalysingoilsamples,morethantenyears'worthygasrecordsarecollectedandclassifiedintothreevoltagelevel,69KV,16KVand345KV.Thus,gasrecordsforonetransformerarecomposedofseveralgroupsofdata.IntheprocessofDGAinterpretation,allofthesedatamaybeconsidered,butonlytherecentdatawhichhavesignificanteffectsondiagnosisarelistedinthelaterdemonstration.InMTL,allgasconcentrationsareexpressedbypminvolumeconcentration.100pmisequalto0.01ml(gas)/100ml(oil).

Fromtheexpertiseofdiagnosis,thenormalstatecanbeconfirmedonlybyinspectionofthetransformer'snormslevel.Inpractice,mostofthetransformeroilsamplesarenormal,andthiscanbeinferredsuccessfullyontheearlyexecutionofthisexpertsystem.However,theSuccessofanexpertsystemismainlydependentonthecapabilityofdiagnosisforthetransformersinquestion.Intheimplementation,manygasrecordswhichareinabnormalconditionarechosentotesttheJustificationofthisdiagnosticsystem.Atotalof101transformerrecordshavebeenexecutedandtheresultsaresummarizedinTable5.Amongthoseimplemented,threearelistedanddemonstrated.

ShowninTable5aretheresultsof101unitsoftransformersinthreetypesofremedy:normal,thermalfaultandarcfault.Aftercomparingthemwiththeactualstateandexpertjudgement,asummaryofresultswasobtained.Aspreviouslystated,oneunitoftransformermayincludemanygroupsofgasdata.Inevaluation,wedepictedsomekeygroupsinoneunittojustifybecausesometransformersmayhavedifferentincipientfaultsduringdifferentoperationalstages.Somemistakesimplementedfromtestingarecausedbytheremainingoilintheoilsamplingcontainer,unstablegascharacteristicsofthenewdegassingsampleandsomeobscuregastypes.Ifmoreinformationornewtechniquessupportotheruncertainmembershipfunctions,theycanbeaddedintotheknowledgehastoenlargethetheperformanceofthisprototypeexpertsystem.Furthermore,theparametersdescribedintable2,3and4aresuitableforTPCpowertransformer.Differentregionsmaybemodifiedthemaintenancepersonnelfindmoresuitablesystemparameters.

CONCLUSIONS

Aprototypeexpertsystemisdevelopedonapersonal