RiskSensitive Estimators forInaccurately ModelledSystems
Shovan Bhaumikl, Smita Sadhu · 2005
IEEE Indicon 2005 Conference,~~~~~~~~ ~~~ Abstract - Robustness ofrisk sensitive (RSE)estimators/filters forinaccurately modelled plant areelucidated andexemplified. A theorem whichallows alternative pathway forderiving RSE filter relation andderivation ofdifferent closed formrelations forRSfilters inlinear Gaussian casesisprovided. Consequently, errorsinexpressions in earlier publications havebeendetected andrectified. Properties ofRSfilters arebriefly reviewed and theinterpretation ofrobustness ofRS filters elaborated. Using MonteCarlosimulation, itisshownthatRS filters perform significantly better compared torisk-neutral filters when(i) processnoise covariance isinerror(u)thetruesystem(truth model) contains unmodelled bias (iiO) thestate transition matrix isinaccurately known.Design pragmatics forthechoice ofthe risk sensitive parameter isindicated. Keywords - Kalmanfilter, Modeluncertainty. Risksensitive filter, Robust Estimation