A novel decompose-ensemble methodology with AIC-ANN approach for crude oil forecasting. (1st July 2018)

• Record Type: Journal Article
• Authors: Ding, Yishan
• Is Part Of: Energy. Volume 154(2018)
• Page Start: 328
• Page End: 336
• Publication Date: 2018-07-01
• Subjects: Crude oil price prediction -- Ensemble empirical mode decomposition -- Akaike's information criterion -- Hybrid model -- Predicting accuracy -- Stability; Power resources -- Periodicals; Power (Mechanics) -- Periodicals; Energy consumption -- Periodicals; 333.7905
• DOI: 10.1016/j.energy.2018.04.133
• Languages: English
• ISSNs: 0360-5442

Abstract

Abstract: Forecasting international crude oil is a well-known issue. The hybrid modeling principle tells us that combining different methods could take full advantage of all the merits and leave out the shortcomings. Therefore, hybrid methodology has been widely used in current research. In this study, a novel decompose-ensemble prediction process combining the ensemble empirical mode decomposition (EEMD) and artificial neural network (ANN) is proposed. Moreover, this method, i.e., EEMD-ANN-ADD method, adds the decompose-ensemble to the single AI model to further improve the predicting accuracy. The overall process can be divided into four steps: model selection via Akaike's information criterion (AIC), data decomposition via EEMD, individual prediction via ANN and ensemble prediction through addition ensemble method. To verify the results, we use the official data of oil price to conduct the predicting. The result confirms that "decompose-ensemble" models are better than the normal hybrid one, in terms of prediction accuracy (both level and directional measurement) and modified Diebold-Mariano test. What's more, back to the decompose-ensemble models, the EEMD-based one outperforms the empirical mode decomposition (EMD) one. At last but not the least, AIC gives us reasonable and convincing statement about determining the value of lag. Generally speaking, this novel forecasting technique is a prominent insight for the price of crude oil. Highlights: We propose novel decompose-ensemble model for crude oil forecasting. We compare the decompose-ensemble model with the normal hybrid one. Information criteria acts as the model selection rule for the prediction process. We elaborate more on the training ratio of ANN method.