Research on Neural Network Based on Improved PSO Algorithm

Abstract

A modified particle swarm optimization algorithm is proposed for artificial neural network studying, in which a window is introduced to detect the change of the environment. Through re-initialization of the parameters, the modified particle swarm optimization algorithm can enhance the diversity of the particles, as will solve the problems that the convergence result is determined by the distribution of the particles in the initialization and the particle swarm optimization often converges prematurely when solving the problems in complicated situations. Another parameter will be tuned according to conditions of the search process and this tuning will balance the global and local search capability. Experiments were made to verify the improved algorithm and the results showed that the modified particle swarm optimization algorithm gains more advantages over the original one.