基于深度学习的农产品期货价格预测研究  

Research on Price Prediction of Agricultural Futures Using Deep Learning Techniques

作　　者：刘彦虹 刘合兵[1] 尚俊平[1] LIU Yanhong;LIU Hebing;SHANG Junping(School of Information and Management Science,Henan Agricultural University,Zhengzhou 450046,China)

机构地区：[1]河南农业大学信息与管理科学学院,郑州450046

出　　处：《河南科学》2024年第3期430-439,共10页Henan Science

基　　金：河南省科技攻关项目(212102110204);河南省研究生教育改革与质量提升工程项目(YJS2023AL046);河南省现代农业产业技术体系(S2010-01-G04);河南农业大学信息与管理科学学院大学生创新创业项目(2022-XGDC-11)。

摘　　要：提高农产品期货价格的预测能力可为投资者的投资交易和政府宏观调控提供一定借鉴.在对LSTM、GRU、BiLSTM三种深度学习模型进行对比研究的基础上,通过添加随机种子稳定预测结果、使用一阶差分降低价格预测滞后性、用正则化、回调函数等方法解决过拟合问题,对LSTM模型进行优化.利用大连商品交易所农产品期货数据,将优化后的模型应用于玉米、黄大豆1号、鸡蛋三种农产品期货的价格预测.预测结果评价指标表明,优化LSTM模型的均方根误差为17.04,平均绝对误差为13.94,误差分别降低了38.6%和33.6%.优化的深度学习模型能够用于预测农产品期货价格,为投资交易提供借鉴.Enhancing the predictive capability of agricultural futures prices can offer valuable insights for investors in their transactions and further bolster the theoretical basis for government macroeconomic policies.Through a comparative analysis of three deep learning models(LSTM,GRU,and BiLSTM),the optimization of the LSTM model included the incorporation of random seeds to enhance result stability,utilization of first-order differences to minimize prediction delays,and addressing overfitting issues using regularization techniques and callback functions.Subsequently,the refined model was employed for predicting the prices of three agricultural futures(corn,yellow soybean No.1,and egg)by leveraging data from the Dalian Commodity Exchange.Evaluation of the forecast results indicated that the optimized LSTM model achieved a root mean square error of 17.04 and a mean absolute error of 13.94,resulting in reduced errors of 38.6%and 33.6%,respectively.The enhanced deep learning model developed in this study holds the potential to forecast agricultural futures prices for investment trading purposes.

关 键 词：深度学习 神经网络 非线性预测 时序数据 长短期记忆网络 

分 类 号：TP314[自动化与计算机技术—计算机软件与理论] F323.9[自动化与计算机技术—计算机科学与技术]