Forecasting the HS300 Stock Index with Radial Basis Function Neural Network Using Two-Stage Particle Swarm Optimization

Abstract

This paper introduces a hybrid radial basis function neural network (RBFNN) using a two-stage particle swarm optimization. In stage one, we initially demonstrate how the PSO learning method simultaneously determines the optimal number of hidden neurons, centres and widths of the radial basis function. Afterwards we apply the least mean squares (LMS) method to calculate the weight between the hidden layer and output layer in stage two. Then we propose a novel fitness function and make forecast simulation of HS 300 stock index. It proves to improve the performance of RBFNN.