基于LSTM神经网络的中国原油期货指数预测研究  

The Research on the Prediction of China Crude Oil Futures Index Based on LSTM Neural Network

作　　者：虞文美[1] 方扶星 Yu Wenmei;Fang Fuxing(Anhui University of Finance and Economics)

机构地区：[1]安徽财经大学

出　　处：《哈尔滨师范大学自然科学学报》2020年第2期32-37,共6页Natural Science Journal of Harbin Normal University

基　　金：安徽省教育厅高校优秀青年人才支持计划重点项目:“一带一路”对人民币国际化的总体估计、传导机制和经济互动效应研究(gxyqZD2017047);安徽财经大学校级课题:中国金融周期的度量、成因分析和宏观经济效应研究(ACKY1728);安徽财经大学大学生科研创新基金项目研究成果(XSKY2007ZD)。

摘　　要：针对中国原油期货指数的预测问题,从经济指标和技术指标两个层面分别选取4个预测变量构建指数预测体系,并收集了中国原油期货自2018年3月26日上市以来的相关数据,根据时间窗口划分为短期(5日)、中期(10日)以及长期(30日).分别利用长短期记忆(LSTM)神经网络构建指数预测模型进行实证研究.结果发现:长期时间窗口模型相较于其余两者有着更高的预测精度,训练集与测试集的均方根误差(RMSE)分别下降到了7.8210和6.5274,表明该模型有着良好的预测效果.因此可以利用长期窗口LSTM神经网络构建高效的中国原油期货指数预测模型,为金融机构的智能化投资能力以及监管机构的风险识别与风险测度提供一定的参考依据.In view of the forecast problem of China's crude oil futures index,four forecast variables were selected from the economic indicators and technical indicators to build the index forecast system,and the data collected since the listing of China's crude oil futures on March 26,2018,which are divided into short-term(5 days),medium-term(10 days)and long-term(30 days)according to the time window using long-term and short-term memory(LSTM)neural network to construct an exponential prediction model for empirical research.The results show that the long-term time window model has higher prediction accuracy compared to the other two,and the mean square root error(RMSE)of training set and test set decreased to 7.8210 and 6.5274,respectively,that the model with good prediction effect is indicated.Therefore,the long-term window LSTM neural network can be used to construct an efficient prediction model of China's crude oil futures index.And some reference for the intelligent investment ability of financial institutions and the risk identification and risk measurement of regulatory agencies are provided.

关 键 词：中国原油期货 指数预测 LSTM神经网络 R 

分 类 号：F224[经济管理—国民经济]