A PREDICTIVE ALGORITHM BASED ON DIFFERENCE RADIAL BASIS FUNCTION NEURAL NETWORKS MODEL FOR TIME SERIES AND ITS APPLICATION

Because the performance of the classical radial basis function(RBF) predictor for non-stationary time series is less satisfactory, a modified structure called Difference RBF(DRBF) and its predictive algorithm are presented. Through the application to the prediction of the Shanghai stock index, the simulation results confirm the superior performance of the DRBF over the classical RBF, and the former is more fit for the non stationary time series problems.