Regional Prediction of Non-stationary Time Series

Based on state space reconstruction and the embedding theory, a new regional prediction method of non-stationary time series is presented. External forcing factors are embedded in the model, and the corresponding space information regarding to the predict phase point in this region is also imported into the model. Using this method, an ideal non-stationary time series from the 33-mode Lorenz system is analyzed. The results show that by embedding external forcing factors, the original dynamic system can be reconstructed efficiently, and the prediction accuracy of the non-stationary time series can be effectively developed. Space information can offset the deficiencies of time series length by embedding both space information and external forcing factors, which will further improve the prediction accuracy.