Radial Basis Function Neural Network Based on PSO with Mutation Operation to Solve Function Approximation Problem

摘要

This paper presents a novel learning algorithm for training and constructing a Radial Basis Function Neural Network (RBFNN), called MuPSO-RBFNN algorithm. This algorithm combines Particle Swarm Optimization algorithm (PSO) with mutation operation to train RBFNN. PSO with mutation operation and genetic algorithm are respectively used to train weights and spreads of oRBFNN, which is traditional RBFNN with gradient learning in this article. Sum Square Error (SSE) function is used to evaluate performance of three algorithms, oRBFNN, GA-RBFNN and MuPSO-RBFNN algorithms. Several experiments in function approximation show MuPSO-RBFNN is better than oRBFNN and GA-RBFNN.