Identificartion for Linear System by Renewal of Models

Abstract

This paper considers the identification for linear system of unknown order.
First, we hypothesize some candidate models which have different orders. For each model, we estimate the unknown system parameters and calculate its a posteriori probability by Bayesian theorem. Then we select the optimal model on basis of the a posteriori probability.
Next, we renew the candiate models in consideration of asymptotic property of a posteriori probability, when we have poorly priori information of system order.