基于注意力时间卷积网络的农产品期货分解集成预测  

Forecasting agricultural commodity futures with decomposition and ensemble strategy based on attentional temporal convolution network

作　　者：张大斌[1] 黄均杰 凌立文[1] 林锐斌 ZHANG Dabin;HUANG Junjie;LING Liwen;LIN Ruibin(College of Mathematics and Informatics,South China Agricultural University,Guangzhou 510642,China)

机构地区：[1]华南农业大学数学与信息学院,广州510642

出　　处：《南京信息工程大学学报》2024年第3期311-320,共10页Journal of Nanjing University of Information Science & Technology（Natural Science Edition）

基　　金：国家自然科学基金面上项目(71971089);国家自然科学基金青年项目(72001083);广东省自然科学基金面上项目(2022A1515011612)。

摘　　要：针对农产品期货时间序列数据受多方面因素影响,非线性、非平稳数据特征难以提取而导致预测准确性不高的问题,基于“分解-集成”的预测思想,本文提出一种基于自适应噪声完备经验模态分解(CEEMDAN)与Transformer-Encoder-TCN的农产品期货预测方法.首先,使用CEEMDAN将时间序列分解为多尺度多频率的本征模态分量(IMF)与残差,降低了序列建模复杂度;其次,使用融合多阶段自注意力单元Transformer-Encoder的时间卷积网络(TCN)对各个分量子序列进行特征提取与预测,优化了序列显著特征建模权重;最后,将各个子序列预测值线性相加集成得到最终预测结果.以南华期货公司农产品指数中的大豆期货指数为研究对象,采用时序交叉验证与参数迁移的方式进行模型重训练,消融和对比实验结果表明,提出的新模型在RMSE、MAE和DS三个评价指标上具有良好的效果,验证了该模型对农产品期货预测的有效性.To address the low prediction accuracy in agricultural commodity futures due to their nonlinear and non-smooth features resulting from various influencing factors,this paper proposes a decomposition and ensemble forecas-ting approach based on CEEMDAN and Transformer-Encoder-TCN.First,the Complementary Ensemble Empirical Mode Decomposition with Adaptive Noise(CEEMDAN)is used to decompose the time series into multiscale Intrin-sic Mode Function(IMF)and residuals,reducing the complexity of series modeling.Second,each subseries is pre-dicted via Temporal Convolutional Network(TCN)incorporating multi-stage self-attention unit(Transformer-En-coder),which optimizes the modeling weights of significant features.Finally,the prediction results of each subseries are linearly summed and integrated to obtain the final prediction results.The soybean futures revenue index in the agricultural commodity index of South China Futures Company is used as the research object.The model is retrained by time-series cross-validation and parameter transfer.The ablation and comparison experimental results show that the proposed model has superiority in RMSE,MAE and DS,verifying its effectiveness in predicting agricultural com-modity futures.

关 键 词：农产品期货 自适应噪声完备经验模态分解 自注意力机制 Transformer-Encoder 时间卷积网络 

分 类 号：F830.91[经济管理—金融学] TP183[自动化与计算机技术—控制理论与控制工程]