An improved identification method for a class of time-delay systems

Abstract

This paper mainly discusses the identification of a certain class of time-delay systems with polluted outputs. The expectation maximization (EM) algorithm in conjunction with the Laplace distribution are applied to build the effective identification framework. The explicit equations to infer the unknown delay and parameters are simultaneously derived with the EM algorithm and the abnormality of process data is also handled by the Laplace distribution. In the proposed identification process, the weight which is inversely proportional to the absolute difference between the measured and predicted outputs is adaptively assigned to each outlier and then the negative effect of the outlier is suppressed. The verification tests reveal the validity of the improved method under different noise and outlier levels.