Multivariate chaotic time series prediction based on extreme learning machine

Abstract

For multivariate chaotic time series prediction problem, a prediction based on input variable selection and extreme learning machine is proposed in this paper. The multivariate chaotic time series is reconstructed in phase space, and a mutual information based method is used to select the input variables, which have high statistics information with the output variables. The extreme learning machine is conducted to model the multivariate chaotic time series in the phase space by utilizing its approximation capability. In order to improve the prediction accuracy, a model selection algorithm is conducted for extreme learning machine to choose an expected minimum risk prediction model. Simulation results based on Lorenz, Rssler multivariate chaotic time series and Rssler hyperchaotic time series show the effectiveness of the proposed method.