Research on neural networks learning algorithm based on PSO and COA

To the shortcoming of BP algorithm that solution is sensitive to initial value and easy to trap in local optima, this paper makes a research on Particle Swarm Optimization (PSO), Chaos Optimization Algorithm (COA) and a modified Chaos Particle Swarm Optimization (CPSO) and applies them in neural networks learning problem. The mechanism of algorithms is explored in depth. A novel method of evaluating the degree of gathering for the swarm is proposed. The performance of algorithms is tested and analyzed by simulation and compared with BP algorithm. The results show that as novel neural networks learning algorithms, PSO and CPSO can overcome the defect of BP algorithm whose solution is sensitive to initial value and have the certain application value.