资源描述：
《A Particle Algorithm for Sequential Bayesian .pdf》由会员上传分享，免费在线阅读，更多相关内容在学术论文-天天文库。
1、326IEEETRANSACTIONSONSIGNALPROCESSING,VOL.50,NO.2,FEBRUARY2002AParticleAlgorithmforSequentialBayesianParameterEstimationandModelSelectionDominicS.LeeandNicholasK.K.ChiaAbstract—Wedescribeaparticlealgorithmforthesequentialperformtheestimationsequentially,updatingtheestimateofBayesianestim
2、ationofunknownstaticparameters.Thealgorithmaseachmeasurementbecomesavailable.combinessequentialimportancesampling(SIS)andMarkovchainBydescribingimperfectionsoruncertaintiesinthephysicalMonteCarlo(MCMC)toachievecomputationalefficiencyandsta-processesusingprobabilitymodels,thecompletedescr
3、iptionofbility.Initsmostgeneralform,thealgorithmhasthreecompo-nents:i)SIS;ii)arejuvenationtest;andiii)MCMC.Measure-probabilisticobjectsisprovidedbydistributions.Henceforth,mentsareprocessedsequentially(withanartificial“time-line”ifweassumethatalldistributionsarecontinuoussothattheirthere
4、isnonaturaloneassociatedwiththemeasurements)bySIS,associateddensitiesexist.Inthediscretecase,massfunctionswhichiscomputationallyinexpensive.Aftereachmeasurementisshouldreplacedensities,andsummationsshouldreplaceinte-processed,therejuvenationtestcheckswhethertheresultingSISgralswhereverap
5、propriate.Forgenericrandomvectorsandparticleshavetoberejuvenated.Whenindicatedbythetest,theparticlesarecompletelyrejuvenatedbyMCMC,whichremoves,weusetodenotethedensityof,andletde-errorsthataccumulatefromSISduetothefinitenumberofpar-notetheconditionaldensityofgiven.Withthesenotationsticle
6、s,thusensuringstability.Wheneverpossible,theSISparticlesandletting,thesolutionsthatweseekarecanbeusedtoadvantageintheMCMC.Thereisflexibilityinthe,.choiceoftherejuvenationtestaswellastheMCMCmethod,withWehavethreemotivationsforconsideringsequentialparam-potentialtoincreasetheusefulnessofth
7、ealgorithm.Inparticular,byusingreversible-jumpMCMCwithmultiplemodels,thealgo-eterestimation.First,thereareproblemsthatcanbeformulatedrithmcanperformsimultaneousmodelselectionandparameteres-ashavingparametersonlyandforwhichon-lineestimatesoftimation.Inthispaper,weusearejuv
