基于自适应残差动态融合图注意力网络的交通速度预测

Traffic speed prediction study based on adaptive residual dynamic fusion graph attention network

摘要

在现代智能交通系统领域里,准确预测交通速度对缓解交通拥堵、提高道路安全以及优化交通管理有着重要意义。 为提升现有的交通速度预测模型在中长期预测任务中的性能,本文提出一种自适应残差动态融合图注意力网络的交通速度预测方法,该方法中的双模态图架构通过对自适应邻接矩阵与动态邻接矩阵进行并行处理和动态融合,可以捕捉路网静态拓扑和动态时空关联特征;采用门控时间卷积实现特征筛选,并利用多头注意力机制增强时空特征表达能力,设计动态特征融合单元,通过残差连接保留静态拓扑信息,结合跨层多尺度特征融合避免特征退化。 实验结果显示,与 GraphWaveNet 相比,本模型在 METR-LA和 PEMS-BAY 数据集 60 min 预测任务中均方根误差分别减少 22.5%和 22.6%。 该模型能够实时适应交通状态的变化,为交通管理部门提供精准的速度预测,辅助拥堵疏导、动态路径规划和突发事件响应,具备较高的实际应用价值。

Abstract

In the field of modern intelligent transportation systems, accurate prediction of traffic speed is of great significance to alleviate traffic congestion, improve road safety, and optimize traffic management. To improve the performance of existing traffic speed prediction models in medium and long-term prediction tasks, this paper proposes an adaptive residual dynamic fusion graph attention network for traffic speed prediction, in which the bimodal graph architecture can capture static topology and dynamic spatio-temporal correlation features of the road network through parallel processing and dynamic fusion of adaptive and dynamic adjacency matrices. Applying gated temporal convolution to realize the feature screening, and using multi-head attention mechanism to enhance the spatio-temporal feature expression ability, designing dynamic feature fusion unit, retaining static topological information through residual connection, and combining cross-layer multi-scale feature fusion to avoid feature degradation. The experimental results show that the root mean square error of this model is reduced by 22.5% and 22.6% compared with GraphWaveNet in the 60 min prediction task for the METR-LA and PEMS-BAY datasets, respectively. The model can adapt to the changes of the traffic state in real time, provide accurate speed prediction for the traffic management department, and assist in the congestion diversion, dynamic path planning, and emergency response. The model has high practical application value.

关键词

城市交通 / 交通速度预测 / 双模态图融合 / 动态图卷积网络 / 残差机制 / 注意力机制

Key words

city traffic / traffic speed prediction / bimodal map fusion / dynamic graph convolutional network / residual connection / attention mechanism

参考文献

[1] WEI Honglei , SONG Qingbiao , DAN Chonghong , et al. Performance evaluation of ARIMA, autoformer, and symmetric LSTNFCL models for traffic accident emergency prediction[J]. Symmetry, 2025, 17(5): 639— 639.

[2] ZHOU Xiao , GUO Ling , LI Rui , et al. Intelligent teaching recommendation model for practical discussion course of higher education based on naive bayes machine learning and improved k—NN data mining algorithm[J]. Information, 2025, 16(6): 512.

[3] KHALLADI A S , OUESSAI A , KECHE M . Road traffic classification from nighttime videos using the multihead self—attention vision transformer model and the SVM[J]. Automatic Control and Computer Sciences, 2024, 58(5): 544-554.

[4] 薛红军, 陈广交, 李鑫民, 等 . 基于决策树理论的交通流参数短时预测[J]. 交通信息与安全, 2016, 34(3): 64-71.

[5] XUE Hongjun , CHEN Guangjiao , LI Xinmin , et al. Short—term prediction of traffic flow parameters based on decision tree theory[J]. Journal of Traffic Information and Safety, 2016, 34(3): 64-71.

[6] SANHITA D , KUMAR A M , ARKA D . A recurrent neural network model for predicting two—leader car—following behavior[J]. Transportation Letters, 2024, 16(5): 461-475.

[7] XU W , BLANCAFLOR E , ABISADO M . Performance and improvement of deep learning algorithms based on LSTM in traffic flow prediction[J]. Discover Applied Sciences, 2025, 7(4): 278.

[8] PRAKASH N , SHARAN B , ANIL K , et al. Time series prediction for traffic flow forecasting using CNN[J]. Asian Journal of Research in Computer Science, 2025, 18(5): 442-449.

[9] CUI Z Y , KRISTIAN H , RUIMIN K , et al. Traffic graph convolutional recurrent neural network: a deep learning framework for network—scale traffic learning and forecasting[J]. IEEE Transactions on Intelligent Transportation Systems, 2019, 21(11): 4883-4894.

[10] LI G H , MULLER M , THABET K A , et al. Can GCNs go as deep as CNNs?[C]// 2019 IEEE/CVF International Conference on Computer Vision. Seoul: IEEE, 2020: 9266-9275.

[11] WU Zonghan , PAN Shirui , LONG Guodong , et al. Graphwavenet for deep spatial temporal graph modeling[C]// Proceedings of the 28th International Joint Conference on Artificial Intelligence (IJCAI). San Francisco, CA: Morgan Kaufmann, 2019: 1907-1913.

[12] VELIKOVI P , CUCURULL G , CASANOVA A , et al. Graph attention networks[EB/OL]. (2018—02—04)[2025—04—28]. https://doi.org/10.48550/arXiv.1710.10903.

[13] MANESSI F , ROZZA A , MANZO M . Dynamic graph convolutional networks[J]. Pattern Recognition, 2020, 97: 107000.

[14] 王博文, 王景升 . 基于图注意力网络的交通流量时空预测模型[J]. 公路交通科技, 2022, 39(6): 153-160.

[15] WANG Bowen , WANG Jingsheng . A spatio—temporal prediction model of traffic flow based on graph attention network[J]. Highway and Transportation Science and Technology, 2022, 39(6): 153-160.

[16] ZHU Mingting , ZHU Xianqiang , ZHU Cheng . STGATP: a spatio—temporal graph attention network for long—term traffic prediction[J]. Lecture Notes in Computer Science, 2021, 12893: 255-266.

[17] ZHANG Junbo , ZHENG Yu , QI Dekang . Deep spatio—temporal residual networks for citywide crowd flows prediction[EB/OL]. (2017—10—01)[2025—04—29]. https://doi.org/10.48550/arXiv.1610.00081.

[18] WU Yuankai , TAN Huachun , QIN Lingqiao , et al. A hybrid deep learning—based traffic flow prediction method and its understanding[J]. Transportation Research Part C: Emerging Technologies, 2018, 95: 32-47.

[19] CININ, AYDIN Z. A deep ensemble approach for long—term traffic flow prediction[J]. Arabian Journal for Science and Engineering, 2024, 49(9): 12377-12392.

[20] JI Jiahao , WANG Jingyuan , HUANG Chao , et al. Spatio—temporal self—supervised learning for traffic flow prediction[EB/OL]. (2024—03—06)[2025—05—01]. https://doi.org/10.48550/arXiv.2212.04475.

[21] WANG Zhaohuan , XU Yi , HAN Liangzhe , et al. Multivariate long—term traffic forecasting with graph convolutional network and historical attention mechanism[C]// Knowledge Science, Engineering and Management Lecture Notes in Computer Science. Cham: Springer, 2023, 14120: 112-123.

[22] LI Zhonghang , XIA Long , SHI Lei , et al. OpenCity: open spatio—temporal foundation models for traffic prediction[EB/OL]. (2024—08—16)[2025—05—01]. https://arxiv.org/abs/2408.10269.

[23] 孙大盟, 欧阳安杰, 何立明 . 基于改进 STGCN 深度学习框架的交通速度预测[J]. 计算机技术与发展, 2024, 34(11): 133-139.

[24] SUN Dameng , OUYANG Anjie , HE Liming . Traffic speed prediction based on improved STGCN deep learning framework[J]. Computer Technology and Development, 2024, 34(11): 133-139.

[25] ZHANG Dongping , LAN Hao , WANG Mingting , et al. Graph convolutional networks with multi—scale dynamics for traffic speed forecasting[J]. Applied Soft Computing, 2025: 174.

[26] CAI Jiongbiao , WANG H C , HU Kun . LCDFomer: long—term correlations dual—graph—transformer for traffic forecasting[J]. Expert Systems With Applications, 2024, 249: 123721.

[27] TIAN Chenyu , CHAN Waikin . Spatial—temporal attention wavenet: a deep learning framework for traffic prediction considering spatial—temporal dependencies[J]. IET Intelligent Transport Systems, 2021, 15(4): 549-561.

基金资助

国家重点研发计划资助项目(2023YFB4302701)