Short-term forecasting of natural gas prices by using a novel hybrid method based on a combination of the CEEMDAN-SE-and the PSO-ALS-optimized GRU network

作     者：Wang, Jun Cao, Junxing Yuan, Shan Cheng, Ming 

作者机构：Chengdu Univ Technol State Key Lab Oil & Gas Reservoir Geol & Exploita Chengdu 610059 Peoples R China Chengdu Univ Technol Sch Geophys Chengdu 610059 Sichuan Peoples R China Chengdu Univ Technol Sch Earth Sci Chengdu 610059 Sichuan Peoples R China 

出 版 物：《ENERGY》 (能)

年 卷 期：2021年第233卷

页      面：121082-121082页

核心收录：

学科分类：0820[工学-石油与天然气工程] 08[工学] 0807[工学-动力工程及工程热物理] 

基　　金：National Natural Sci-ence Foundation of China [41430323  41974160  42030812  42042046] 

主　　题：Deep learning CEEMDAN-SE decomposition PSO algorithm Adaptive learning strategy GRU neural networks Natural gas price forecasting 

摘      要：With the continuous growth of the global natural gas trade, the accurate prediction of natural gas prices has become one of the most critical issues in the planning and operation of public utilities. In order to further improve the prediction accuracy of natural gas prices, we have developed a novel hybrid method based on the CEEMDAN-SE (complete ensemble empirical mode decomposition with an adaptive noise sample entropy) and the PSO-ALS-GRU (gated recurrent unit network optimized by the particle swarm optimization algorithm with an adaptive learning strategy (PSO-ALS)) for predicting the natural gas prices. The proposed approach can address the limitations of the traditional forecasting approaches and perform accurate predictions. First, the original natural gas price series is decomposed into a series of sub-sequences with obvious differences in a complex degree by using the CEEMDAN-SE. Then, the forecasting model PSO-ALS-GRU is developed, and each sub-sequence is individually predicted. Finally, the prediction results of each sub-sequence are superimposed and reconstructed in order to form the overall forecast result. This hybrid model combines the methodology of complex systems with deep learning techniques, making it more appropriate for analyzing relationships such as long-term dependences and solving complex nonlinear problems. By finding the key hyperparameters in the GRU network using the PSO-ALS, the data feature of natural gas prices matches the network topology structure, and the prediction accuracy of the model is improved. For illustration and verification purposes, the simulation is performed by using real data. The results show that the novel hybrid model can accurately predict the weekly prices of natural gas. In a comparison of prediction errors with other individual models, the proposed model demonstrates the highest prediction ability among all of the investigated models. (c) 2021 Elsevier Ltd. All rights reserved.