On the Parameter Estimation of Stochastic Switched ARX Model

Abstract

This paper presents a parameter identification technique for a stochastic switched ARX model. The developed algorithm can be regarded as a natural extension of the learning algorithm for Hidden Markov Model (HMM), which is based on the EM algorithm. The results of the estimation of the parameter in the ARX model at each discrete state can be comprehended as a solution for the weighted least mean square estimation problem. The usefulness of the proposed technique is verified through some numerical experiments for the switched impedance model with and without existence of the modelling error.