A neural network learning algorithm based on hybrid particle swarm optimization

Abstract

A hybrid learning algorithm based on simplex method and particle swarm optimization is proposed to train the feedforward neural network in this paper. In the given hybrid algorithm the simplex method which has expansion function and contraction function is embedded in the particle swarm optimization as an operator. Through cross-training mode to train neural network, this hybrid algorithm selects limited elitist particles and executes simplex operator for local searching during each generation of particle swarm optimization, which can make the neural network learning approximate to the global optimum region rapidly and find more excellent solution. The simulation experiments show that comparing with some traditional learning methods this hybrid algorithm enhances the convergence speed and training precision, and improves network performance. It is an effective neural network learning method.