A hybrid prediction model of multivariate chaotic time series based on error correction

Abstract

Considering the problem that simply modifying the reservoir algorithm cannot significantly improve the prediction accuracy of chaotic multivariate time series, in this paper we propose a hybrid prediction model based on error correction. The observed data includes both linear and nonlinear features. First, we use autoregressive and moving average model to capture the linear features, then build a regularized echo state network to portray the dynamic nonlinear features. Finally, we add the predicted nonlinear value to the predicted linear value, in order to improve forecasting accuracy achieved by either of the models used separately. The experimental results of Lorenz and Sunspot-Runoff in the Yellow River time series demonstrate the effectiveness and characteristics of the proposed model herein.