A Combination Model Based on EEMD-PE and Echo State Network for Chaotic Time Series Prediction

Abstract

Prediction of chaotic time series has broad application prospects and becomes a research hotspot. Since chaotic time series is strongly non-stationary and nonlinear, it’s difficult to predict based on any single model. Therefore, the empirical mode decomposition (EMD)-based combination model becomes an important means of prediction. To reduce the scale of prediction models of conventional combination method, this paper proposes a high-efficiency combination model using ensemble EMD (EEMD), permutation entropy (PE) and echo state network(ESN). EEMD decomposes the original time series into a group of intrinsic mode functions (IMFs), and the number of IMFs is consistent with the number of predictors. On account of the complexity of the chaotic time series, there is a large demand of predictors. Through complexity analysis by PE, we combine some IMFs whose complexities are similar, and predict the combined signals instead of the initial ones based on ESNs. Finally, these obtained estimates are assembled as the ultimate prediction results. In the experiment, we use real-world dataset to examine the proposed model. The experimental results confirm that our combination approach outperforms existing single models, and efficiently reduces the scale of prediction models comparing to the EEMD-ESN.