基于 CEEMDAN-SE-CNN-BiLSTM 模型的大豆期货价格预测
Soybean Futures Price Forecasting Based on CEEMDAN-SE-CNN-BiLSTM Model
周雅丽 1谭莹莹 2赵玉华3
作者信息
- 1. 安徽建筑大学数理学院,安徽合肥 230601
- 2. 安徽建筑大学数理学院,安徽合肥 230601;安徽建筑大学运筹学与数据科学实验室,安徽合肥 230601
- 3. 合肥师范学院 数学与统计学院,安徽合肥 230601
- 折叠
摘要
为了提高大豆期货价格预测的精确度,综合利用大豆期货市场内外部信息,基于一种"分解—重组—预测—集成"多步期货价格预测模型进行改进.对大豆价格序列进行自适应噪声完备集合经验模态分解(CEEMDAN),得到IMF分量及误差项.筛选后剔除与原序列相关系数小的IMF分量,再用样本熵算法(SE)对分解序列进行重组.用优化的CNN-BiLSTM预测模型对重组序列进行预测,集成后得到最终预测值.实证结果表明:在预测大豆期货价格时,改进后的CEEMDAN-SE-CNN-BiLSTM模型普遍优于LSTM、CNN-LSTM等基准农产品期货预测模型.
Abstract
In order to improve the accuracy of soybean futures price prediction,a multi-step futures price prediction model based on"decomposition-reorganization-prediction-integration"is improved by utilizing internal and external information of soybean futures market.The soybean price series are subjected to adaptive noise-complete ensemble empirical modal decomposition(CEEMDAN)to obtain the IMF components and error terms.The IMF component with small correlation coefficients with the original series is screened and eliminated,and then the decomposed series are reorganized by the sample entropy(SE)algorithm.Using the optimized CNN-BiLSTM prediction model to predict the reconstructed sequence,the final prediction value is obtained after integration.The empirical results show that the proposed CEEMDAN-SE-CNN-BiLSTM model generally outperforms the benchmark agricultural futures prediction models such as LSTM and CNN-LSTM in predicting soybean futures prices.
关键词
大豆期货价格预测/自适应噪声完备集合经验模态分解/样本熵/卷积神经网络/双向长短期神经网络Key words
soybean futures price prediction/CEEMDAN/sample entropy/convolutional neural network/BiLSTM引用本文复制引用
基金项目
安徽省运筹控制与组合优化创新团队项目(2023AH010020)
安徽省教育厅新时代育人质量工程项目(2022xxsfkc031)
安徽省研究生创新实践项目(2022cxcysj150)
出版年
2024
NETL
NSTL
万方数据