A cooperative evolutionary system for designing neural networks

Abstract

A novel cooperative evolutionary system, i.e., CGPNN, for automatic design artificial neural networks (ANN's) is presented where ANN's structure and parameters are tuned simultaneously. The algorithms used in CGPNN combine genetic algorithm (GA) and particle swarm optimization (PSO) on the basis of a direct encoding scheme. In CGPNN, standard (real-coded) PSO is employed to training ANN's free parameters (weights and bias) and binary-coded GA is used to find optimal ANN's structure. In the simulation part, CGPNN is applied to the predication of tool life. The experimental results show that CGPNN has good accuracy and generalization ability in comparison with other algorithms. © Springer-Verlag Berlin Heidelberg 2006.