Desensitized Optimal Filtering
Christopher D. Karlgaard, Haijun Shen · 2011
This paper discusses the development of a desensitized optimal filtering technique for systems subject to plant model parameter uncertainties. Desensitized state estimates are obtained by minimizing a cost function consisting of the posterior covariance matrix trace penalized by a weighted norm of the state estimate error sensitivities. The resulting filter is non-minimum variance but exhibits reduced sensitivity to deviations in the assumed plant model parameters. Solutions are obtained for discrete and continuous time linear systems and for discrete and mixed continuous-discrete nonlinear systems using an Extended Kalman Filter-like formulation. Two example problems with parameter uncertainties are provided to illustrate the effectiveness of the proposed filtering technique.