Designing for RBF Networks Based on Particle Swarm Optimization and Regularized Orthogonal Least Squares

This paper presents a two-level learning method for designing radial basis function (RBF) networks based on particle swarm optimization (PSO) and regularized orthogonal least squares (ROLS), which is called ROLS-PSO method. The ROLS algorithm is employed at the lower level to construct RBF networks while the two key learning parameters, i.e., the regularized parameter and the RBF width, are optimized using PSO method to obtain the optimal value. The simulation results indicate that the RBF neural network designed with ROLS-PSO method not only has a more parsimonious network models, but also has better generalization ability than the one designed with orthogonal least squares (OLS) and ROLS algorithm, which demonstrates the effectiveness of this new approach