基于分解集成的LSTM神经网络模型的油价预测  

FORECASTING WTI CRUDE OIL PRICE WITH AN EEMD-BASED LSTM NEURAL NETWORK ENSEMBLE LEARNING PARADIGM

作　　者：高海翔 胡瑜 余乐安[1] Gao Haixiang;Hu Yu;Yu Lean(School of Economics and Management,Beijing University of Chemical Technology,Beijing 100029,China)

机构地区：[1]北京化工大学经济管理学院,北京100029

出　　处：《计算机应用与软件》2021年第10期78-83,共6页Computer Applications and Software

基　　金：国家杰出青年科学基金项目(71025005)。

摘　　要：为了提高油价的预测效果,提出一种基于EEMD分解、小波阈值去噪、fine-to-coarse法重构和LSTM神经网络的组合预测方法。EEMD对油价原始时间序列分解,利用小波阈值去噪法获取第一高频模态分量的有效信息;分解出的模态分量运用fine-to-coarse法重构,得到从高到低的重构分量;使用LSTM神经网络预测重构分量;对重构序列简单加和得到最终结果。实证结果表明,与其他基准模型比较,在水平预测和趋势预测上该方法能有效地预测原油价格。In order to improve the performance of the oil price,a novel learning paradigm is put forward based on the combining forecasting methods of EEMD,fine-to-coarse method,the wavelet threshold denoising and LSTM neural network.EEMD decomposed the original time series of oil price to a number of intrinsic mode functions(IMFs)and extracted the effective information of the first high-frequency component by using the wavelet threshold in the proposed method.Using the fine-to-coarse method to refactor all IMFs,then LSTM neural network was used to predict all reconstructed components.The predicted results by LSTM were integrated into an ensemble value as final prediction.Empirical results show that the proposed model significantly improves the effect on oil price prediction compared with other benchmark models in level prediction and directional prediction.

关 键 词：长短期记忆神经网络 集合经验模态分解 小波阈值去噪 fine-to-coarse法 油价预测 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]