Thechaoticcharacteristicsoftimeseriesoffivepartialdischarge(PD)patternsinoil-paperinsulationarestudied.TheresultsverifyobviouschaoticcharacteristicofthetimeseriesofdischargesignalsandthefactthatPDisachaoticprocess.Thesetimeserieshavedistinctivefeatures,andthechaoticattractorsobtainedfromtimeseriesdifferedgreatlyfromeachotherbyshapesinthephasespace,sotheycouldbeusedtoqualitativelyidentifythePDpatterns.Thephasespaceparametersareselected,thenthechaoticcharacteristicquantitiescanbeextracted.ThesequantitiescouldquantificationallycharacterizethePDpatterns.TheeffectsonpatternrecognitionofPRPDandCAPDarecomparedbyusingtheneuralnetworkofradialbasisfunction.Theresultsshowthatbothofthetworecognitionmethodsworkwellandhavetheirrespectiveadvantages.Then,boththestatisticaloperatorsunderPRPDmodeandthechaoticcharacteristicquantitiesunderCAPDmodeareselectedcomprehensivelyastheinputvectorsofneuralnetwork,andthePDpatternrecognitionaccuracyistherebygreatlyimproved.
