EEMD and evolutionary KPCR based adaptive prediction modeling on complex time series

Aiming to some nonlinear, non-stationary, multi-scale characteristics of time series, an adaptive prediction modeling method based on ensemble empirical mode decomposition (EEMD) and evolution kernel principal component regression (KPCR) was proposed. Firstly, the original time series was decomposed into different scales by EEMD according to its composition characteristics, and then C-C method was applied to make the phase space reconstruction in every scale, where KPCR with a composite kernel was used to build a prediction function; at the same time, KPCR model was optimized with a given criteria by particle swarm optimization (PSO) algorithm in every scale, and finally the prediction results in different scales were integrated into the predicted value of time series. The results of the empirical prediction analysis for the international crude oil price show that this method can effectively describe the trend of time series in different scales and adaptively obtain the optimal prediction model, compared with the existing method, which has strong adaptive modeling capabilities and higher prediction accuracy. ©, 2014, Systems Engineering Society of China. All right reserved.