Abstract：This research focuses on the issue of chaotic time series prediction. A virtual feature extraction method for forecasting performance improvement is proposed. Firstly, details and smooths of the chaotic time series are extracted by shift invariant wavelet algorithms. Then the relationship between the linear and nonlinear components is extrapolated from additive to functional one. Finally, a novel virtual feature expression is given based on the above wavelet details and smooths for forecasting. Experiment results of forecasting on Mackey-Glass and real Mississippi River flow series show that the proposed method is superior over some existing methods, which also demonstrate the effectiveness of the proposed virtual feature extraction method. Moreover, the results may provide a decision-making reference for a variety of chaotic areas, such as control, hydrology, and meteorology.
2014, Vol. 36 
