Performance diagnosis of local filters in state estimation of nonlinear systems

European Workshop on Advanced Control and Diagnosis

Straka, Ondej^1 ; Duník, Jindich^1 ; Šimandl, Miroslav^1

Department of Cybernetics, Faculty of Applied Sciences, University of West Bohemia, Univerzitni 8, Pilsen

306 14, Czech Republic^1

关键词: Discrete;    time systems;    Filter performance;    Kalman-filtering;    Nongaussianity;    Performance diagnosis;    Recursive computation;    Third order moment;    Tracking motion;   

Others  :  https://iopscience.iop.org/article/10.1088/1742-6596/570/1/012004/pdf
DOI  :  10.1088/1742-6596/570/1/012004

来源: IOP

【 摘 要 】

The paper deals with state estimation of nonlinear stochastic discrete-time systems with a special focus on filters providing a point estimate of the state. The aim is concentrated on performance diagnosis of the filters, which should inform about a possible decrease of estimate quality provided by the filter. The paper proposes a diagnostics of the filter performance using the non-Gaussianity measures. More specifically, the paper focuses on a third-order moment based non-Gaussianity measure and provides relations for its recursive computation. It is shown that this measure can effectively be used for the performance diagnosis of the filters, which is illustrated using a numerical example. The example considers tracking motion of a ship using the extended Kalman filter.