Robust Kalman Filter as Parameter Estimator for Output Error Models

Abstract

The data in industry are corrupted with stochastic noise. In the real situations data contain outliers which can create problems to linear algorithms. Because, some kind of prevention must be taken into account. So are developed robust procedures for parameters estimation. In this paper we shall consider output error model and for robust parameters estimation the Masreliez-Martin’s robust filter is used. This filter is generalization of Kalman filter. In this paper we (i) eliminate the transformation factor (ii) nonlinear Masreliez-Martin prediction error transformation we replace with Huber function (iii) Fisher information is replaced with derivative of Huber’s function (iv) generation of input signal (experiment design) is based on ideas from predictive control Also, the intensive simulations are performed.