Adaptive Identification of Autoregressive Processes

Abstract

Adaptive identification problems on coefficient matrices of an autoregressive model for multivariate and one‐dimensional nonstationary Gaussian random processes are investigated by applying the extended Kalman filter incorporated with a weighted global iteration. The major contributions of the present paper are the use of the extended Kalman filter for estimating time‐varying model parameters recursively, and the development of an effective method in terms of computer time. The results indicate that the coefficients of this recursive equation are identified simply, but extremely well, at the stage of their stable convergence to optimum. It is suggested that the method be applied for the analysis of multiple‐point array data of earthquake ground motion.