Radial Basis Function Neural Network Optimized by Particle Swarm Optimization Algorithm Coupling with Prior Information

The traditional training algorithms of RBF neural network converge slowly and are easy to fall into local optimum. The swarm intelligent algorithms such as particle swarm optimization (PSO) can conquer these shortcomings because of their good capability of global search. To further overcome the inadequacies of standard PSO, an improved PSO algorithm coupling with prior information is proposed in this paper. For classification, the relation between the properties and the category labels implied in the sample data can be abstracted as prior information. The prior information is used to adjust the position of particles and the fitness formula. In the proposed algorithm, RBF neural network is first trained by improved PSO and then by gradient descent. The prior information shrinks the search space and guides the movement direction of the particles, so the convergence rate and the generalization performance are both improved. In order to demonstrate the good performance of the new algorithm, a lot of experiments for classification have been carried out. The simulation results show that the proposed method is more effective than traditional ones.