An approach to recursive identification of abruptly changing systems

Abstract

A way to model systems with abruptly changing dynamics is suggested. The parameters of the system are described as realizations of a finite-state Markov chain. It is further discussed how to perform recursive parameter identification for this type of system. A crucial part in the identification algorithm is to estimate the present state of the Markov chain. The effects of some typical rules to do this estimation are examined. Also a procedure which reduces the need for a priori information is given.