With A Observation Noise Parameter Estimation Of New Methods And Algorithms

Based on the classical least squares method (RLS) in system identification, the several new identification algorithms of parameter estimation for the autoregressive moving average (ARMA) model, are presented. They include univariable and multivariable two-stage recursive least squares-recursive extended least squares (RLS-RELS) and two-stage recursive least squares-pseudo-inverse (RLS-PI) algorithms. Compared with classical recursive extended least squares, their accuracy obviously is improved. Applying these new algorithms to the parameter estimation problem for systems with measurement noises, some new approaches and algorithms of parameter estimation for system with measurement noises, are presented, for example, two-stage RELS-Gevers-Wouters algorithm and three-stage RLS-PI-Gevers-Wouters algorithm, which solve the biased parameter estimation problem by classical least squares method. Applications of their to self-tuning filtering are given, where the steady-state Kalman tracking filter with the position and velocity measurements, and multivariable self-tuning tracking predictor, filter and smoother are presented. Simulation examples show effectiveness of new algorithms.