数据智能与应用 (13).pdfVIP

Received27December2022,accepted19January2023,dateofpublication27January2023,dateofcurrentversion1February2023.

DigitalObjectIdentifier10.1109/ACCESS.2023.3240306

ConditionMonitoringofWindTurbine

GeneratorsBasedonSCADADataand

FeatureTransferLearning

1,2,324

XIAOHANGJIN,(SeniorMember,IEEE),HAOWANG,ZIQIANKONG,

ZHENGGUOXU4,(Member,IEEE),ANDWEIQIAO5,(Fellow,IEEE)

1KeyLaboratoryofSpecialPurposeEquipmentandAdvancedManufacturingTechnology,MinistryofEducationandZhejiangProvince,ZhejiangUniversityof

Technology,Hangzhou310023,China

2CollegeofMechanicalEngineering,ZhejiangUniversityofTechnology,Hangzhou310023,China

3NinghaiZJUTAcademyofScienceandTechnology,Ninghai315600,China

4CollegeofControlScienceandEngineering,ZhejiangUniversity,Hangzhou310027,China

5PowerandEnergySystemsLaboratory,DepartmentofElectricalandComputerEngineering,UniversityofNebraska–Lincoln,Lincoln,NE68588,USA

Correspondingauthor:XiaohangJin(xhjin@)

TheworkofXiaohangJinandHaoWangwassupportedinpartbytheNationalKeyResearchandDevelopmentProgramofChinaunder

Grant2022YFE0198900,andinpartbytheNingboNaturalScienceFoundationunderGrant2021J038.

ABSTRACTInordertobuildaneffectiveconditionmonitoring(CM)modelforthetargetwindtur-

bines(WTs)withfewoperationaldata,anapproachbasedonthefeaturetransferlearningandamodified

generativeadversarialnetworkisproposed.First,alargeamountoflabelleddatafromWTsareanalyzed

toconstructaCMmodelwiththeaidofanautoencoder.ThisformstheknowledgeofCMforWTsinthe

sourcedomain.Second,agenerativeadversarialnetworkistrainedtobuildamappingrelati