摘要
深度学习在用于预测非线性时间序列时表现出色,且无须考虑变量之间的内生性问题。将集成经验模态分解(ensemble empirical mode decomposition,EEMD)方法与卷积神经网络(convolutional neural networks,CNN)、长短期记忆模型(long short-term memory,LSTM)、门控循环单元(gated recurrent units,GRU)相结合,构建基于集成分解的农产品期货价格预测模型。以中国玉米、棉花和大豆期货价格为例,对原始期货价格信号进行EEMD分解,然后将分解向量分别输入深度学习模型中进行训练,最终得出EEMD-GRU模型为最优价格预测模型。结果显示,与单独的深度学习模型相比,该文所提基于集成分解的组合模型在预测准确性方面优势明显,具有更强的泛化能力。
Deep learning performs excellent in predicting nonlinear time series,and it does not consider the endogeneity between variables.This paper integrates the Ensemble Empirical Mode Decomposition(EEMD) method with Convolutional Neural Network(CNN),Long Short-Term Memory(LSTM),and Gated Recurrent Unit(GRU),and a model for forecasting agricultural commodity futures prices based on integrated decomposition is constructed.Taking Chinese corn,cotton and soybean futures prices as examples,the original futures price signal is decomposed by EEMD,and then the decomposed vectors are input into the deep learning models for training.Finally,it is concluded that EEMD-GRU model is the optimal price prediction model.The results demonstrate that compared with the individual deep learning models,the proposed integrated EEMD model has obvious advantages in predictive accuracy and stronger generalization ability.
作者
张博群
孙倩
沈虹
ZHANG Boqun;SUN Qian;SHEN Hong(School of Business,Yangzhou University,Yangzhou 225127,China)
出处
《扬州大学学报(自然科学版)》
CAS
2024年第4期47-55,共9页
Journal of Yangzhou University:Natural Science Edition
基金
国家自然科学基金资助项目(61803331,92371116)
江苏省自然科学基金资助项目(BK20170515)。
关键词
农产品期货
集成经验模态分解
深度学习
agricultural commodity futures
ensemble empirical mode decomposition(EEMD)
deep learning