A Divide-and-Conquer System Based Neural Networks for Forecasting Time Series

This paper presents a Divide-and-Conquer System based Neural Networks (DCSNN) for forecasting time series. This DCSNN is composed of several sub-RBF networks which takes each low-dimensional sub-input as its input. The output of DCSNN is the sum of each sub-RBF networks´ output. The algorithm of DCRBF is given and its forecasting ability also is discussed in this paper. The experimental results have shown that the DCSNN is outperforms the conventional RBF for forecasting time series.