An Improved Approach Combining Random PSO with BP for Feedforward Neural Networks

Abstract

In this paper, an improved approach which combines random particle swarm optimization with BP is proposed to obtain better generalization performance and faster convergence rate. It is well known that the backpropagation (BP) algorithm has good local search ability but it is easily trapped to local minima. On the contrary, the particle swarm optimization algorithm (PSO), with good global search ability, converges rapidly during the initial stages of a global research. Since the PSO suffers from the disadvantage of losing diversity, it converges more slow around the global minima. Hence, the global search is combined with local search reasonably in the improved approach which is called as RPSO-BP. Moreover, in order to improve the diversity of the swarm in the PSO, a random PSO (RPSO) is proposed in the paper. Compared with the traditional learning algorithms, the improved learning algorithm has much better convergence accuracy and rate. Finally, the experimental results are given to verify the efficiency and effectiveness of the proposed algorithm.