Study of RBF Neural Network Based on PSO Algorithm in Nonlinear System Identification

Abstract

Development of neural network provided new thought for nonlinear system identification. RBF neural network was widely studied in nonlinear system identification by good approximation ability and fast convergence thereof. In the paper, RBF neural network based on PSO algorithm was proposed, global searching property of PSO algorithm was utilized for remedying RBF local approximation, initial weights of RBF neural network and the base width were globally optimized, insufficiency in RBF neural network random initialization weights and base width was remedied, and identification precision of RBF neural network on nonlinear system was improved aiming at problems of RBF neutral network in nonlinear system identification application, such as local approximation and base width random initialization. The simulation results showed that RBF neural network based on PSO algorithm, proposed in the paper, had prominently better identification precision on nonlinear system than identification of RBF neural network based on GA algorithm and the traditional RBF neural network, and it had great significance on identification of nonlinear systems.