Application of Parallel RBF Network on Iterative Prediction of Chaotic Time Series

Abstract

An application of Parallel Radial Basis Function (PRBF) network model on prediction of chaotic time series is presented in this paper. The PRBF net consists of a number of radial basis function (RBF) subnets connected in parallel. The number of input nodes for each RBF subnet is determined by different embedding dimension based on chaotic phase -space reconstruction. The output of PRBF is a weighted sum of all RBF subnets and represents the prediction value for each new input vector. The chaotic time series data from Lorenz simulation signal and hydraulic pump vibration signal was used to verify the proposed method. Both GrassbergerProcaccia (G-P) algorithm and Takens' method were employed to calculate the minimum embedding dimension of chaotic time series. Finally, the prediction accuracy and result were compared between RBF and PRBF. It is shown that PRBF network is more effective and feasible for the iterative prediction of chaotic time series.