nextuppreviouscontents
Next:Publications Up:Stochastic Estimation and Control Previous:Relations Between Optimal Risk-Sensitive

Maximum Likelihood Parameter Estimation for Systems with Noisy Dynamics and Observations

his project is concerned with estimating unknown parameters for nonlinear systems which are subject to noisy dynamics and observations. The parameters are computed using Maximum-Likelihood techniques via the Expectation-Maximization (EM) algorithm. The methodology is based on computing the unnormalized conditional density of the nonlinear filtering problem, followed by computing the parameters via the EM algorithm. In the special case of Gauss-Markov models, the parameters entering the Kalman-Filter are obtained in closed-form, while for general nonlinear models a Galerkin's approximation scheme is introduced which computes the conditional density. The ML parameter estimates are expressed in terms of the coefficients of the Galerkin scheme.

C.D. Charalambous, M. Demetriou
 


Annual Report

Mon Jun 26 21:22:20 GMT 2000