Time series prediction using RNN in multi-dimension embedding phase space

Abstract

In this paper, a multidimension chaotic time series prediction method using recurrent neural network (RNN) in embedding phase space is proposed. This method is to reconstruct a phase space based on the chaotic time series and then embed these data as the phase space points for the training of the RNN. The resultant RNN after the training will be served as the embedding phase space which is capable of recovering the predicted phase space point into time domain. Thus, the predicted chaotic time series data can be obtained. Numerical results have shown that the proposed method is simple, practical and effective in chaotic time series prediction.