Design of robust recursive identification algorithms for large-scale stochastic systems

Abstract

The robust recursive algorithms, for identification of decentralized stochastic systems, are developed. It is supposed that stochastic disturbance belongs to a specified class of distributions which include the gross error model suitable for the description of outliers presence. Such an assumption introduces into the recursive algorithms a nonlinear transformation of prediction error. The given algorithms are robust with respect to uncertainty in the disturbance distribution. The individual subsystems are described with SISO (single-input single output) ARMAX model. Two algorithms are considered: the stochastic approximation and the least squares. Their comparison is based on simulations.