


SignalProcessing88(2008)468–491
doc.guandang.net
Review
Propagator-basedmethodsforrecursivesubspace
modelidenti cation
rea,Ã,LaurentBakob,c,StephaneLec ucheb,cGuillaumeMerce
´dePoitiers,Laboratoired’Automatiqueetd’InformatiqueIndustrielle(EA1219),Universite
40avenuedurecteurPineau,86022Poitiers,France
b´coledesMinesdeDouai,941,rueCharlesBourseulB.P.838,59508DouaiCedex,France´partementInformatiqueetAutomatique,EDe
c
´nieInformatiqueetSignal(UMRCNRS8146),Cite´Scienti queB.P.48,Laboratoired’Automatique,Ge
59651Villeneuved’AscqCedex,France
Received1February2007;receivedinrevisedform10September2007;accepted14September2007
Availableonline25September2007
a
Abstract
Theproblemoftheonlineidenti cationofmulti-inputmulti-output(MIMO)state-spacemodelsintheframeworkofdiscrete-timesubspacemethodsisconsideredinthispaper.Severalalgorithms,basedonarecursiveformulationoftheMIMOOutputErrorState-Space(MOESP)identi cationclass,aredeveloped.Themaingoalsoftheproposedmethodsaretocircumventthehugecomplexityofeigenvaluesorsingularvaluesdecompositiontechniquesusedbytheof inealgorithmandtoprovideconsistentstate-spacematricesestimatesinanoisyframework.Theunderlyingprincipleconsistsinusingtherelationshipbetweenarraysignalprocessingandsubspaceidenti cationtoadjustthepropagatormethod(originallydevelopedinarraysignalprocessing)totrackthesubspacespannedbytheobservabilitymatrix.Theproblemofthe(coloured)disturbancesactingonthesystemissolvedbyintroducinganinstrumentalvariableintheminimizedcostfunctions.Aparticularattentionispaidtothealgorithmicdevelopmentandtothecomputationalcost.Thebene tsofthesealgorithmsincomparisonwithexistingmethodsareemphasizedwithasimulationstudyintime-invariantandtime-varyingscenarios.
r2007ElsevierB.V.Allrightsreserved.
Keywords:Systemidenti cation;Recursivealgorithm;Subspacemethod;Instrumentalvariable;Multivariablesystem
Contents1.2.3.
Introduction.....................................................................Notationsandproblemformulation....................................................Recursiveestimationoftheobservationvector:anoverview...................................3.1.UpdatingoftheordinaryMOESPRQfactorization....................................3.2.UpdatingofthePI/POMOESPRQfactorization......................................Recursiveestimationofanobservabilitymatrixbasis........................................
469470472472473474
4.
Correspondingauthor.Tel.:+33549453509;fax:+33549454034.
re),bako@ensm-douai.fr(L.Bako),lecoeuche@ensm-douai.frE-mailaddresses:guillaume.mercere@univ-poitiers.fr(G.Merce
(S.Lec uche).
0165-1684/$-seefrontmatterr2007ElsevierB.V.Allrightsreserved.doi:10.1016/j.sigpro.2007.09.012
热门文档