Identificação ARX LPV robusta a outliers via Estimador-M

Abstract

This paper deals with the problem of identifying autoregressive linear dynamic systems with exogenous inputs and variant parameters with the presence of outliers in the data. For this, we revisit the algorithm proposed by Bamieh e Giarre (2002), which uses the Least Mean Square rule. It is proposed to replace its learning rule with a variant, the Least Mean M-estimate, which usesM-Estimators that are robust to outliers. Different contamination levels are considered. The results of the computational experiments show a considerable improvement in the algorithm’s performance.