Chaotic time series prediction based on wavelet echo state network

Abstract

Chaos is widespread in nature and human society, so the prediction of chaotic time series is very important. In this paper, we propose a new chaotic time series prediction model echo state network based on wavelet, which can effectively overcome the ill-posed problem that exists in traditional echo state networks. And it also has a good generalization ability. Three time series are used to test the new model, i.e., Lorenz time series, Lorenz time series with added noise and batch reactor vessel temperature time series. Results suggest that the new proposed method can achieve a higher predictable accuracy, better generalization and more stable prediction results than traditional echo state networks.