Designing Neural Networks Using Hybrid Particle Swarm Optimization

Abstract

Evolving artificial neural network is an important issue in both evolutionary computation (EC) and neural networks (NN) fields. In this paper, a hybrid particle swarm optimization (PSO) is proposed by incorporating differential evolution (DE) and chaos into the classic PSO. By combining DE operation with PSO, the exploration and exploitation abilities can be well balanced, and the diversity of swarms can be reasonably maintained. Moreover, by hybridizing chaotic local search (CLS), DE operator and PSO operator, searching behavior can be enriched and the ability to avoid being trapped in local optima can be well enhanced. Then, the proposed hybrid PSO (named CPSODE) is applied to design multi-layer feed-forward neural network. Simulation results and comparisons demonstrate the effectiveness and efficiency of the proposed hybrid PSO.