Prediction of Multivariate Chaotic Time Series Based on Optimized Phase Space Reconstruction

In this paper, a new method is applied for predicting multivariate chaotic time series which based on optimized multivariate phase space reconstruction. The details of the methodology are: the ranges of the dimension and the delay of every variable are set firstly, and the least prediction error indicator for selecting the optimal parameters is employed as the criterion. Then the phase space reconstruction with the optimal parameters is used as the input of the neural network, in the end, the best result of the prediction is obtained. Simulations of the Lorenz system and the real world time series show that the methodology proposed is efficient.