Outlier Robust One-Step-ahead Adaptive Predictor for Hammerstein Models

Abstract

In paper considered outliers-robust recursive stochastic approximation algorithm for adaptive prediction of MIMO (multiple input multiple output) Hammerstein models. The static nonlinear block has polynomial form and linear block is output-error model. It is supposed that a priori known class of distributions to which belongs the real disturbance. In that situation we can use Huber`s methodology for design of robust algorithm which introduces nonlinear transformation of prediction error. Model transformation allows representation of unknown matrix parameters in the form of vector. The problem is not considered before in the field of adaptive prediction. Simulation study presents the practical behaviour of algorithm.