System Identification by the General Parameter Neural Networks with Fuzzy Self-Organization

Abstract

A novel system identification method based on variable structure neural networks is proposed. The background of the developed approach is General Parameter (GP) method of complex system identification which has effective training features and provides information about model approximation accuracy. Improved structure backpropagation and radial basis function networks are developed. A fuzzy procedure is designed for quazi-optimal network structure determination based on learning stage information. Simulations are performed to confirm the feasibility of proposed networks for practical applications. Next, these are experimentally applied to nonlinear time series prediction, fault diagnosis and control problems resulting in higher accuracy and faster learning rate in compare to conventional networks.