Recursive Least M-estimate for robust identification of ARX LPV systems with impulsive noise

Abstract

This paper deals with the presence of impulsive noise in identification data for linear systems with exogenous inputs and varying parameters. For this, we revisit the algorithm proposed by Bamieh and Giarrè (2002), which uses the Recursive Least Square rule. It is proposed to replace its learning rule with a variant, the Recursive Least M-estimate, which uses M-Estimators that are robust to outliers. Different levels of contamination are applied. The results of the computational experiments show a significant improvement in the algorithm’s performance.