Crude Oil Price Forecasting Based On Decomposition Ensemble Learning Paradigm: Mode Reconstruction And Component Forecasting Approach Research

Crude oil as the world's primary energy. The price volatility affects the international political, economic and military status seriously, via variety of transmission mechanism. Especially in the case of the gradual opening up energy prices, and the rising external dependence on Crude oil, volatility in crude oil prices seriously affect the normal operation of the national economy. Under the influence of market supply and demand, global economic and political environment, market speculation, emergencies, crude oil present the characteristics of nonlinear and high-complexity, which makes the crude oil price forecasting becoming the research hotspot and difficulty.To enhance prediction accuracy and reduce computation complexity, This paper would like to improve the reconstruction and component forecasting approach in decomposition ensemble learning paradigm, and proposed two novel decomposition-ensemble methodology.One is the decomposition-ensemble model with data-characteristic-driven reconstruction. This novel proposed model improves the existing decomposition-ensemble techniques, by formulation and incorporating a novel data-characteristic-driven reconstruction based on "data-characteristic-driven modeling". Four main steps are involved, i.e., data decomposition, component reconstruction, individual prediction, and ensemble prediction. In the proposed novel data-characteristic-driven reconstruction, all decomposed modes are thoroughly analyzed to explore hidden data characteristics, and are accordingly reconstructed into some meaningful components. The empirical analysis indicate that the proposed model statistically outperforms all considered benchmark models?The other is an EEMD-based FA-LSSVR Ensemble Learning Paradigm. It is actually an extension on exsiting ensemble models. Three main steps are involved, i.e., data decomposition, individual forecasting, and ensemble forecasting. In particular, Firefly algorithm (FA) is especially introduced to optimal penalty coefficient and kernel function parameters in LSSVR. The empirical results indicate that the novel ensemble learning paradigm achieve highest prediction accuracies and robustness.These approaches proposed by this study have been proved that the accuracy of crude oil price prediction is promoted. Morever, the proved models are effective in forecasting crude oil price with high volatility and irregularity.