Filtering discrete time nonlinear systems with unknown parameters: a nonparametric approach. (Q557113)

A particle filtering technique is proposed for discrete-time nonlinear systems with unknown parameters. It is well known that particle filtering is in general not well suited for the case of unknown parameters when the analytic form of the likelihood function is not available. The method proposed in the paper overcomes the traditional difficulties and exploits a kernel-based nonparametric estimation of the joint conditional probability density of the system state and parameters. \(L_1\)-convergence of the approximate conditional density to its true counterpart when the number of generated particles diverges is proved. Furthermore, if in addition time asymptotics are considered, the conditional density of the parameters concentrates around their true values.