Designing Artificial Neural Networks Using MCPSO and BPSO

Abstract

A novel hybrid evolutionary system HPSONN combing an improved particle swarm optimization using multiple swarms(MCPSO) and a binary particle swarm optimization (BPSO) is proposed for joint optimization of three-layer feed-forward artificial neural networks (ANNs). In the proposed method, the topology of neural network is optimized by BPSO and connection weights are training by MCPSO. The experiment results on function approximation problem show that HPSONN can produce compact ANNs with good accuracy and generalization.