Modified Kolmogorov's neural network in the identification of Hammerstein and Wiener systems.

Abstract

This brief deals with the possibilities of using the modified Kolmogorov's neural network for the identification of non-linear dynamic systems, among them the Wiener and Hammerstein systems. The algorithm of training the network is simple, well convergent and with a small error of approximation. The modified neural network is characterized by a simple computer algorithm; it also omits complicated techniques of back propagation. The simulation results are shown to illustrate the modified Kolmogorov theorem.