ROBUST RECURSIVE-IDENTIFICATION OF LINEAR-SYSTEMS USING PRIOR INFORMATION

The problem of recursive robust identification of linear discrete-time dynamic stochastic systems, when the a priori disturbance statistics are incomplete, is discussed. Min-max robust identification algorithms of stochastic gradient type are derived, making use of a priori data, such as the class of distributions to which the unknown disturbance d istribution belongs. Making use of prior statistical information on th e estimated parameters, the rate of estimates convergence at initial s teps is improved. The convergence of the developed algorithms is estab lished theoretically, using the martingale theory. The results of simu lation demonstrating the robustness of the proposed algorithms are als o included.