资源描述:
《Instrumental variable methods for system identification 2002》由会员上传分享,免费在线阅读,更多相关内容在行业资料-天天文库。
1、CIRCUITSSYSTEMSSIGNALPROCESSINGVOL.21,NO.1,2002,PP.1-9INSTRUMENTALVARIABLEMETHODSFORSYSTEMIDENTIFICATION*TorstenSOderstrOmtandPetreStoicaIAbstract.Thispapergivesatutorialoverviewofinstrumentalvariablemethods.Com-parisonsaremadetotheleast-squaresmethod.Ananalysisin
2、cludingconsistencyandasymptoticdistributionoftheparameterestimatesisincluded.Keywords:Least-squaresestimation,consistencyanalysis,asymptoticdistribution,recur-sivealgorithm.1.IntroductionOneusualwayofdealingwiththemanysystems(orsignals,iftherearenoobservableinputs
3、)weencounterinourendeavorsistomodelthemandthenusethemodeltoachieveourgoals,asifitwereanexactdescriptionoftherealsystem.Thegoalsinclude:
4、Acquiringinformationandknowledgeaboutthesystemunderstudy;
5、Simulatingthebehaviorofthesystemunderconditionsthatmightbedetri-mental
6、totherealobject;
7、Predictingthebehaviorofthesystemoveracertaintimehorizoninthefuture;
8、Designingaregulatorthatcontrolsthebehaviorofthesystem;9Usingthemodelparametersas"datacompressors",e.g.,forfaultdetectionorrecognition.Themodelsusedtoachievesuchgoalscanbeeitherno
9、nmathematical(e.g.,mental)ormathematical(suchasdifferenceordifferentialequations,state-spaceequations,transferfunctions,andsoon).Systemidentificationisoneofthetwo*ReceivedNovember16,2000.WorksupportedinpartbySwedishResearchCouncilforEngineeringSciencesundercontrac
10、t95-800,andtheFoundationforStrategicResearch.DepartmentofSystemsandControl,InformationTechnology,UppsalaUniversity,Uppsala,Sweden.E-mailforSSderstrOm:ts@SysCon.uu.seE-mailforStoica:ps@SysCon.uu.se2SODERSTROMANDSTO~CApossiblewaysofobtainingmathematicalmodelsforacer
11、tainobject(eitherasignalorasystem)byprocessingexperimentaldatacollectedinanoperationregimethatisthoughttobeclosetothatinwhichthemodelisgoingtobeused.Theotherpossibleway,calledmathematicalmodeling,consistsofmakinguseofthelawsthatareknowntogovernthebehaviorofthesyst
12、emtoobtainamathematicalmodelforit.Themaindrawbackofmathematicalmodelingisthatitcannotbeappliedtocomplicatedsystemseitherbecausetheresultingmodelwouldbet