Another Approach to On-Line Parameter Estimation

An on-line parameter estimator for linear, continuous time systems is derived by approximating the optimal non-linear filter and by keeping terms contributing to convergence. To obtain closure of the moment equations, fourth order moments are approximated by second order moments. Convergence is shown using the ordinary differential equation method for the system in which only the process noise variance is to be estimated. The parameter estimator differs from the extended Kalman filter and the Gaussian-optimal approximation in that a key non-zero third order moment is computed.