A novel crude oil prices forecasting model based on secondary decomposition. (15th October 2022)

• Record Type: Journal Article
• Authors: Li, Guohui; Yin, Shibo; Yang, Hong
• Is Part Of: Energy. Volume 257(2022)
• Publication Date: 2022-10-15
• Subjects: Crude oil prices forecasting -- Improved complementary ensemble empirical mode decomposition with adaptive noise -- State space correlation entropy -- Improved variational mode decomposition by tunicate swarm algorithm -- Improved kernel based extreme learning machine by artificial gorilla troops optimizer; Power resources -- Periodicals; Power (Mechanics) -- Periodicals; Energy consumption -- Periodicals; 333.7905
• DOI: 10.1016/j.energy.2022.124684
• Languages: English
• ISSNs: 0360-5442

Abstract

Abstract: Crude oil prices have risen sharply and plummeted in recent decades. Therefore, its effective forecasting faces great difficulties and challenges. In this paper, a novel crude oil prices forecasting model based on secondary decomposition with improved complementary ensemble empirical mode decomposition with adaptive noise (ICEEMDAN), state space correlation entropy (SSCE), improved variational mode decomposition by tunicate swarm algorithm (TVMD), and improved kernel based extreme learning machine by artificial gorilla troops optimizer (GTO-KELM), named ICEEMDAN-SSCE-TVMD-GTO-KELM, is proposed. Firstly, ICEEMDAN is applied to decompose the original data and the state space correlation entropy is applied to determine the components with higher complexity. Secondly, the component with the highest complexity is decomposed again, which can decompose the nonstationary, nonlinear and highly complex time series into multiple more regular sub-series. Finally, the KELM with parameter optimization is applied to forecast all the components, and the forecast values are reconstructed to obtain the final forecast results. The data of West Texas Intermediate (WTI) and Brent oil are used in the experiments and Diebold Mariano test. Experiments show that the proposed secondary decomposition forecasting model has high forecasting accuracy. Taking WTI as an example, RMSE, MAE, MAPE and R 2 of the proposed model are 0.2947, 0.2133, 0.3665 and 0.9939, respectively, which are better than those of all 9 forecasting models including the one-time decomposition model and the direct forecasting model at the 95% confidence level, demonstrating the promising application of the proposed method in crude oil prices forecasting. Highlights: Propose improved variational mode decomposition by tunicate swarm algorithm. Propose improved kernel based extreme learning machine by gorilla troops optimize. Propose a novel crude oil prices forecasting model based on secondary decomposition.