计算机科学

计算机科学 ›› 2025, Vol. 52 ›› Issue (6A): 240700047-8.doi: 10.11896/jsjkx.240700047

• 大数据&数据科学 • 上一篇    下一篇

基于动态图学习与注意力机制的多变量时间序列预测

洪燚1, 申时凯2, 佘玉梅1, 杨斌3, 代飞4, 王鉴潇1, 张力逸1   

  1. 1 云南民族大学数学与计算机科学学院 昆明 650000
    2 昆明学院信息工程学院 昆明 650000
    3 滁州学院计算机与信息工程学院 安徽 滁州 239000
    4 西南林业大学大数据与智能工程学院 昆明 650000
  • 出版日期:2025-06-16 发布日期:2025-06-12
  • 通讯作者: 申时凯(kmssk2000@sina.com)
  • 作者简介:(hongyi0918@163.com)
  • 基金资助:
    国家自然科学基金面上项目(62372076);国家自然科学基金地区科学基金项目(61962033)

Multivariate Time Series Prediction Based on Dynamic Graph Learning and Attention Mechanism

HONG Yi1, SHEN Shikai2, SHE Yumei1, YANG Bin3, DAI Fei4, WANG Jianxiao1, ZHANG Liyi1   

  1. 1 Department of Mathematics and Computer Science,Yunnan Minzu University,Kunming650000,China
    2 School of Information Engineering,Kunming University,Kunming 650000,China
    3 School of Computer and Information Engineering,Chuzhou University,Chuzhou,Anhui 239000,China
    4 College of Big Data and Intelligent Engineering,Southwest Forestry University,Kunming 650000,China
  • Online:2025-06-16 Published:2025-06-12
  • About author:HONG Yi,born in 2000,master,is a member of CCF(No.E662524639A).Her main research interests include time series prediction and big data analysis.
    SHEN Shikai,born in 1964,master,professor,master supervisor,is a member of CCF(No.22375S).His main research interests include wireless communication network,edge computing,Internet of Things,big data and so on.
  • Supported by:
    National Natural Science Foundation of China(62372076) and National Natural Science Foundation of China(61962033).

PDF (PC)

44

摘要： 多变量时间序列(MTS)预测因变量间复杂的时序依赖和动态相关性而具有挑战性。现有方法大多从单一维度考虑相关影响因素,而未充分考虑多源数据和特征随时间变化的复杂性,这限制了对复杂系统中动态依赖关系的真实反映。针对上述问题,提出了一种基于动态图神经网络(DGNN)的动态关系学习网络(DRLNet)。首先,通过动态更新图邻接矩阵来自适应地建模变量间随时间变化的相关性;然后,设计了一种注意力机制模块,聚焦于重要节点的连接及其随时间的演变;最后,通过评估这些节点与当前时间步的相关程度,引入门控机制选择性地结合历史依赖图。在3个多变量时间序列数据集上的实验结果表明,相较于目前主流的基线方法,DRLNet在预测准确度和稳定性方面表现更优异,能更好地捕捉时序数据中的重要模式和变化,实现多变量时间序列预测。

关键词: 多变量时间序列预测, 时序依赖, 注意力机制, 动态更新图邻接矩阵, 门控机制

Abstract: Multivariate time series(MTS) prediction is challenging due to the complex temporal dependencies and dynamic correlations between variables.Most existing methods focus on single-dimension factors,without fully considering the complexity of multi-source data and evolving feature relationships over time,which limits their ability to capture dynamic dependencies in complex systems.To address these issues,this paper proposes a new model based on dynamic graph neural network(DGNN),which called DRLNet.DRLNet dynamically updates the graph adjacency matrix to adapt to time-varying correlations between variables.Additionally,it includes an attention mechanism that focuses on the evolution of connections between key nodes.A gated mechanism is also introduced to selectively combine historical dependency graphs by evaluating correlations between these nodes and the current time step.Experimental results on three multivariate time series datasets demonstrate that DRLNet outperforms mainstream baseline methods in terms of prediction accuracy and stability.Moreover,it can better capture key patterns and changes in time series data,enhancing its effectiveness for MTS prediction.

Key words: Multivariate time series prediction, Temporal dependencies, Attention mechanism, Dynamically updating graph adjacency matrix, Gated mechanism

中图分类号: 

  • TP391
[1]MATSUBARA Y,SAKURAI Y,VAN PANHUISW G,et al.FUNNEL:automatic mining of spatially coevolving epidemics[C]//Proceedings of the 20th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining.2014:105-114.
[2]PATTON A.Copula methods for forecasting multivariate time series[J].Handbook of Economic Forecasting,2013,2:899-960.
[3]ANGRYK R A,MARTENS P C,AYDIN B,et al.Multivariate time series dataset for space weather data analytics[J].Scientific Data,2020,7(1):227.
[4]TIAN R,LI X,MA Z,et al.LDformer:a parallel neural network model for long-term power forecasting[J].Frontiers of Information Technology & Electronic Engineering,2023,24(9):1287-1301.
[5]WANG J H,LEUJ Y.Stock market trend prediction using ARIMA-based neural networks[C]//Proceedings of International Conference on Neural Networks(ICNN’96).IEEE,1996:2160-2165.
[6]XU D,WANG Y,JIA L,et al.Real-time road traffic state prediction based on ARIMA and Kalman filter[J].Frontiers of Information Technology & Electronic Engineering,2017,18:287-302.
[7]BENVENUTO D,GIOVANETTI M,VASSALLO L,et al.Application of the ARIMA model on the COVID-2019 epidemic dataset[J].Data in Brief,2020,29:105340.
[8]CAO L J,TAY F E H.Support vector machine with adaptive parameters in financial time series forecasting[J].IEEE Transactions on neural networks,2003,14(6):1506-1518.
[9]ROBERTS S,OSBORNE M,EBDEN M,et al.Gaussian processes for time-series modelling[J].Philosophical Transactions of the Royal Society A:Mathematical,Physical and Engineering Sciences,2013,371(1984):20110550.
[10]MASINI R P,MEDEIROS M C,MENDESE F.Machine learning advances for time series forecasting[J].Journal of Economic Surveys,2023,37(1):76-111.
[11]KUMARI S,SINGH S K.Machine learning-based time seriesmodels for effective CO2 emission prediction in India[J].Environmental Science and Pollution Research,2023,30(55):116601-116616.
[12]CASTÁN-LASCORZ M A,JIMÉNEZ-HERRERA P,TRON-COSO A,et al.A new hybrid method for predicting univariate and multivariate time series based on pattern forecasting[J].Information Sciences,2022,586:611-627.
[13]WANG R,PEI X,ZHU J,et al.Multivariable time series forecasting using model fusion[J].Information Sciences,2022,585:262-274.
[14]OZYEGEN O,ILIC I,CEVIK M.Evaluation of interpretability methods for multivariate time series forecasting[J].Applied Intelligence,2022,52:4727-4743.
[15]SHIH S Y,SUN F K,LEE H.Temporal pattern attention for multivariate time series forecasting[J].Machine Learning,2019,108:1421-1441.
[16]LAI G,CHANG W C,YANG Y,et al.Modeling long-and short-term temporal patterns with deep neural networks[C]//The 41st International ACM SIGIR Conference on Research & Development in Information Retrieval.2018:95-104.
[17]BRUNA J,ZAREMBA W,SZLAM A,et al.Spectral networks and locally connected networks on graphs[J].arXiv:1312.6203,2013.
[18]KIPF T N,WELLING M.Semi-supervised classification withgraph convolutional networks[J].arXiv:1609.02907,2016.
[19]FENG F,HE X,WANG X,et al.Temporal relational ranking for stock prediction[J].ACM Transactions on Information Systems,2019,37(2):1-30.
[20]FU Q,BA B,HUANG C J,et al.Dynamic spatiotemporal graph convolutional network for short-term traffic flow prediction[J].Journal of Hunan University of Science and Technology,2024,39(1):70-79.
[21]WANG W T,WANG X Q,LI L X,et al.Review on the construction and application of spatio temporal graph neural networks in traffic flow prediction[J].Computer Engineering and Applications,2024,60(8):31-45.
[22]LI J,MA H,ZHANG Z,et al.Spatio-temporal graph dual-attention network for multi-agent prediction and tracking[J].IEEE Transactions on Intelligent Transportation System,2022,23(8):10556-10569.
[23]GAO Z,LI Z,ZHANG H,et al.Dynamic spatiotemporal interactive graph neural network for multivariate time series forecasting[J].Knowledge-Based Systems,2023,280:110995.
[24]WU Z,PAN S,LONG G,et al.Connecting the dots:multivariate time series forecasting with graph neural networks[C]//Proceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining.2020:753-763.
[25]WU N,GREEN B,BEN X,et al.Deep transformer models for time series forecasting:the influenza prevalence case[J].arXiv:2001.08317,2020.
[26]YOO J,KANG U.Attention-based autoregression for accurate and efficient multivariate time series forecasting[C]//Procee-dings of the 2021 SIAM International Conference on Data Mi-ning(SDM).Society for Industrial and Applied Mathematics.2021:531-539.
[27]LIU S,YU H,LIAO C,et al.Pyraformer:low-complexity pyramidal attention for long-range time series modeling and forecasting[C]//International Conference on Learning Representations.2021.
[28]HAN L,HUO W G,ZHANG Y H,et al.Multivariatetime series prediction based on multi-scale feature fusion and dual attention mechanism[J].Computer Engineering,2023,49(9):99-108.
[29]FUNABASHI S,ISOBE T,HONGYI F,et al.Multi-fingered in-hand manipulation with various object properties using graph convolutional networks and distributed tactile sensors[J].IEEE Robotics and Automation Letters,2022,7(2):2102-2109.
[30]LAI G,CHANG W C,YANG Y,et al.Modeling long-andshort-term temporal patterns with deep neural networks[C]//International ACM SIGIR Conference on Research and Deve-lopment in Information Retrieval.ACM,2018.
[31]ZHOU H,ZHANG S,PENG J,et al.Informer:beyond efficient transformer for long sequence time-series forecasting[C]//Proceedings of the AAAI Conference on Artificial Intelligence.2021:11106-11115.
[32]YE J,LIU Z,DU B,et al.Learning the evolutionary and multi-scale graph structure for multivariate time series forecasting[C]//Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining.2022:2296-2306.
[33]LIU M,ZENG A,CHEN M,et al.SCINet:time series modeling and forecasting with sample convolution and interaction[J].Advances in Neural Information Processing Systems,2022,35:5816-5828.
Viewed
Full text


Abstract

Cited
  Shared   
  Discussed   
No Suggested Reading articles found!
渝ICP备16013121号-4
地址：重庆市渝北区洪湖西路18号    邮编：401121  
电话：023-63500828（编辑部） 023-67039612（会议/专题） 023-67039623（财务/发票）
E-mail：jsjkx12@163.com
本系统由北京玛格泰克科技发展有限公司设计开发