Switch detection and robust parameter estimation for slowly switched Hammerstein systems

Abstract

Switch detection and robust identification for slowly switched Hammerstein systems are considered in this paper. The switching law is slow, arbitrary and cannot be observed online. A two-identifier-based switch detection scheme is proposed, in order to achieve fast adaptability and robustness of parameter estimation. Specifically, at first, a recursive identification method based on long-horizon iteration is exploited under impulsive noise, and its convergence for time-invariant systems is verified. Secondly, to follow the changes of real processes, a forgetting factor is introduced, and two recursive identifiers with different horizon lengths are developed. Identifier (I) with the long horizon length can resist the influence of outliers, and Identifier (II) is responsible for process rapid reactions. Then, the estimated difference between two identifiers is analyzed to distinguish possible switching points from impulsive noise. Consequently, the two-identifier-based switch detection scheme is formed. Finally, a simulation example demonstrates the effectiveness of the proposed scheme.