Chaotic time series prediction based on fuzzy boundary modular neural networks

Abstract

A fuzzy boundary modular neural network （FBMNN） is proposed for the chaotic time series prediction. First，the reconstructed phase space is divided into several subspaces and the divided points are evaluated by genetic algorithms. Then a fuzzy membership function is defined and the fuzzy boundary is set on the border according to the fuzzy membership. Through this fuzzy treatment，the jumping problem of the predicted data near the divided points are solved. Finally the data points of each module and its fuzzy boundary are input to a recurrent neural network for training and the output predicted points are synthesized by a synthesis forecast module. The effectiveness of FBMNN is evaluated by using three benchmark chaotic time series data sets：the Mackey-Glass series，Lorenz series，and Henon series. The simulation results show that FBMNN improves the performance of chaotic time series prediction.