基于LSTM的棉花期货价格预测方法  

Cotton Futures Price Forecasting Method Based on LSTM

作　　者：刘洋 黎玉寒 窦宝明[3] LIU Yang;LI Yuhan;DOU Baoming(Tianjin Chengrui Technology Consulting Co.,Ltd.,Tianjin 300300;School of Economics and Management,Tianjin Agricultural University,Tianjin 300384;School of Economics and Management,Tianjin Tianshi College,Tianjin 301723)

机构地区：[1]天津诚瑞科技咨询有限公司,天津300300 [2]天津农学院经济管理学院,天津300384 [3]天津天狮学院经济管理学院,天津301723

出　　处：《软件》2024年第7期8-12,86,共6页Software

基　　金：2022年教育部人文社科青年基金项目(22YJC630185)。

摘　　要：传统的统计学和计量学时间序列预测模型在复杂的金融市场中存在限制。深度学习的长短期记忆(LSTM)网络被认为能克服这些限制。本研究利用我国2019年-2024年的棉花期货价格数据构建多层LSTM网络预测模型。结果表明,调整LSTM网络模型参数对预测效果优化显著,尤其是对迭代次数、学习率、窗口大小和网络层数的调整。相较于K近邻算法(KNN)、多元线性回归(MLR)以及支持向量回归(SVR)模型,LSTM网络预测准确性更高。以平均绝对百分比误差(MAPE)衡量,LSTM网络相较于KNN、MLR、SVR模型误差降低89.28%、85.92%和17.8%。研究结果表明,LSTM网络在价格预测中表现出色,为棉花期货价格预测提供了新思路。Traditional statistical and econometric time series forecasting models have limitations in complex financial markets.Long Short Term Memory(LSTM)networks in deep learning are believed to overcome these limitations.This study constructs a multi-layer LSTM network prediction model using cotton futures price data from 2019 to 2024 in China.The results indicate that adjusting the parameters of the LSTM network model significantly optimizes the prediction performance,especially in terms of the number of iterations,learning rate,window size,and network layers.Compared to K-nearest neighbor algorithm(KNN),multiple linear regression(MLR),and support vector regression(SVR)models,LSTM networks have higher prediction accuracy.Measured by the Mean Absolute Percentage Error(MAPE),the LSTM network reduces errors by 89.28%,85.92%,and 17.8%compared to KNN,MLR,and SVR models.The research results indicate that LSTM networks perform well in price prediction,providing new ideas for cotton futures price prediction.

关 键 词：LSTM神经网络 价格预测 棉花期货 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]