A Modified Particle Swarm Neural Network Based on Local Chaotic Optimization Strategy

BP neural network is an important type of neural network that has good approximation property and generalization ability. However, it often converges to local optima in implementation. the particle swarm optimization (PSO) is a global optimization method. It is integrated into BP neural network to enhance the global problem-solving ability. Premature problems are often encountered in optimization. to alleviate this problem, a chaotic strategy is devised to prevent the combined algorithm from falling into premature state. This hybrid algorithm is named with modified chaotic particle swarm optimization (MCPSO) algorithm. We use it to optimize the parameters of a BP neural network, and then apply this network to Iris data classification problem. Simulation results show that MCPSO-based BP neural network can not only optimize the network´s parameters and thresholds, but also can achieve accurate classification results.