[1]张震,李卓,刘博,等.基于双通道时空融合的图神经网络交通流预测[J].郑州大学学报(工学版),2027,48(XX):1-9.[doi:10.13705/j.issn.1671-6833.2026.02.004]
　ZHANG Zhen,LI Zhuo,LIU Bo,et al.Traffic Flow Prediction Based on a Dual-channel Spatio-temporal Graph Neural Network[J].Journal of Zhengzhou University (Engineering Science),2027,48(XX):1-9.[doi:10.13705/j.issn.1671-6833.2026.02.004]
点击复制

基于双通道时空融合的图神经网络交通流预测()
分享到：

《郑州大学学报(工学版)》[ISSN:1671-6833/CN:41-1339/T]

卷:
48
期数:
2027年XX
页码:
1-9
栏目:
出版日期:
2027-12-10

文章信息/Info

Title:
Traffic Flow Prediction Based on a Dual-channel Spatio-temporal Graph Neural Network
作者:
张震 1 , 李卓 1 , 刘博 2 , 马继骏 3 , 孔令涛 3 , 王子昂3
1. 郑州大学 电气与信息工程学院,河南 郑州 450001;2. 郑州大学 河南先进技术研究院,河南 郑州 450001;3. 河南省交通运输厅,河南 郑州 450016
Author(s):
ZHANG Zhen 1 , LI Zhuo 1 , LIU Bo 2 , MA Jj 3 , KONG Lt 3 , WANG Za3
1. School of Electrical and Information Engineering, Zhengzhou University , Zhengzhou 450001, China; 2. School of Henan Institute of Advanced Technology, Zhengzhou University, Zhengzhou 450001, China; 3. Henan Provincial Department of transportation, Zhengzhou 450016, China
关键词:
交通流预测;  频域注意力机制;  离散小波变换;  多阶图卷积;  多头图注意力
Keywords:
traffic flow forecasting ;  frequency-enhanced mechanism;  discrete wavelet transform;  multi-order graph convolution;  spatio-temporal graph neural network
分类号:
TP391
DOI:
10.13705/j.issn.1671-6833.2026.02.004
文献标志码:
A
摘要:
针对现有交通流预测方法未能充分捕获异质时序特征以及在建模动态空间依赖性时难以准确捕捉其时变特性等问题，提出了一种基于双通道时空融合机制的图神经网络模型（DC-STGNN）。在时间维度上，设计基于离散小波变换的解耦层，将交通序列分解为低频与高频分量，并分别采用频域增强模块与多尺度扩张因果卷积进行建模，精准捕获异质时序特征；设计时间门控机制，根据交通动态变化调节长期趋势和短期扰动之间的权重，实现更具针对性的时间特征提取，提升模型对不同时间尺度的适应能力。在空间维度上，融合多阶扩散图卷积与多头图注意力机制进行建模，分别提取交通网络中的静态结构特征和动态空间依赖性，有效刻画了复杂交通网络中动态空间交互模式；设计空间门控机制，自适应地融合静态与动态图信息，有效增强模型对复杂空间结构的建模能力。最后在真实的交通数据集 PEMS-04、PEMS-08、PEMS-BAY 和 METR-LA 上分别进行了 15，30 和 60 min 的交通流量预测实验。实验结果表明：DC-STGNN 相较于多个基线模型，在 MAE、RMSE 和 MAPE 3 项预测指标上基本保持最优，表现出更高预测精度与长期稳定性。
Abstract:
To overcome the limitations of existing traffic flow forecasting methods, which often fail to adequately capture heterogeneous temporal patterns and struggle to accurately model the time-varying characteristics of dynamic spatial dependencies, this study proposes a Dual-Channel Spatio-Temporal Graph Neural Network (DC-STGNN). In the temporal dimension, a decoupling layer based on discrete wavelet transform is designed to decompose traffic sequences into low-frequency and high-frequency components. The low-frequency components are modeled using a frequency-enhanced module, while the high-frequency components were captured through multi-scale dilated causal convolution, enabling precise modeling of heterogeneous temporal patterns. In addition, a temporal gating mechanism is introduced to dynamically balance the contributions of long-term trends and short-term fluctuations according to traffic variations, thereby achieving more targeted temporal feature extraction and improving adaptability across different time scales. In the spatial dimension, multi-order diffusion graph convolution and multi-head graph attention are integrated to separately extract static structural features and dynamic spatial dependencies of the traffic network, effectively capturing the evolving spatial interaction patterns in complex traffic systems. Furthermore, a spatial gating mechanism was developed to adaptively fuse static and dynamic graph information, enhancing the model’s capability in representing complex spatial structures. Extensive experiments on real-world traffic datasets--PEMS-04, PEMS-08, PEMS-BAY, and METR-LA--for 15-, 30-, and 60-minute traffic flow prediction tasks show that DC-STGNN achieves higher prediction accuracy and better long-term stability compared to the best-performing baseline models.

参考文献/References:

[1] Cascetta E. Transportation systems engineering: theory and methods[M]. Boston: Springer, 2001.
[2] Liu Boyi, Tang Xiangyan, Cheng Jieren, et al. Traffic flow combination forecasting method based on improved LSTM and ARIMA[J]. International Journal of Embedded Systems, 2020, 12(1): 22.
[3] Cai Lingru, Zhang Zhanchang, Yang Junjie, et al. A noise-immune Kalman filter for short-term traffic flow forecasting[J]. Physica A: Statistical Mechanics and its Applications, 2019, 536: 122601.
[4] Ma Dongfang, Song Xiang, Li Pu. Daily traffic flow forecasting through a contextual convolutional recurrent neural network modeling inter- and intra-day traffic patterns[J]. IEEE Transactions on Intelligent Transportation Systems, 2021, 22(5): 2627-2636.
[5] Han Xu, Gong Shicai. LST-GCN: long short-term memory embedded graph convolution network for traffic flow forecasting[J]. Electronics, 2022, 11(14): 230.
[6] Yang Yanqun, Lin Jie, Zheng Yubin. Short-time traffic forecasting in tourist service areas based on a CNN and GRU neural network[J]. Applied Sciences, 2022, 12(18): 9114.
[7] Han Lingyi, Zheng Kan, Zhao Long, et al. Short-term traffic prediction based on DeepCluster in large-scale road networks[J]. IEEE Transactions on Vehicular Technology, 2019, 68(12): 12301-12313.
[8] Wang Shun, Zhang Yong, Hu Yongli, et al. Knowledge fusion enhanced graph neural network for traffic flow prediction[J]. Physica A: Statistical Mechanics and its Applications, 2023, 623: 12842.
[9] Zhang Yunuo, Wang Xiaoling, Yu Jia, et al. Adaboosting graph attention recurrent network: a deep learning framework for traffic speed forecasting in dynamic transportation networks with spatial-temporal dependencies[J]. Engineering Applications of Artificial Intelligence, 2024, 127: 107297.
[10] Xu Yi, Han Lianghe, Zhu Tongyu, et al. Generic Dynamic Graph Convolutional Network for traffic flow forecasting[J]. Information Fusion, 2023, 100: 101946.
[11] Yu Bing, Yin Haoteng, Zhan Zhixing. Spatio-temporal graph convolutional networks: a deep learning framework for traffic forecasting[C]//Proceedings of the Twenty-Seventh International Joint Conference on Artificial Intelligence. Sweden: AAAI, 2018: 3634-3640.
[12] Li Yaguang, Yu R, Shahabi C, et al. Diffusion convolutional recurrent neural network: data-driven traffic forecasting[PP/OL]. V1. arXiv(2017-07-06)[2025-06-19]. https://doi.org/10.48550/arXiv.1707.01926.
[13] Fan Xuanxuan, Qi Kaiyuan, Wu Dong, et al. MGHCN: multi-graph structures and hypergraph convolutional networks for traffic flow prediction[J]. Alexandria Engineering Journal, 2025, 111: 221-237.
[14] Wang Bin, Long Zhendan, Sheng Jinfang, et al. Spatial-Temporal Similarity Fusion Graph Adversarial Convolutional Networks for traffic flow forecasting[J]. Journal of the Franklin Institute, 2024, 361(17): 107299.
[15] Alsehaimi B, Alzamzami O, Alowidi N, et al. An adaptive spatio-temporal traffic flow prediction using self-attention and multi-graph networks[J]. Sensors, 2025, 25(1): 282.
[16] Zhou Tian, Ma Ziqing, Wen Qingsong, et al. FEDformer: frequency enhanced decomposed transformer for long-term series forecasting[C]//International Conference on Machine Learning. Baltimore: PMLR, 2022.
[17] Xu Xing, Mao Hao, Zhao Yun, et al. An urban traffic flow fusion network based on a causal spatiotemporal graph convolution network[J]. Applied Sciences, 2022, 12(14): 7010.
[18] Wu Zonghan, Pan Shiri, Long Guodong, et al. Graph WaveNet for deep spatial-temporal graph modeling[PP/OL]. V1. arXiv (2019-05-31)[2025-12-06]. https://doi.org/10.4850/arXiv.1906.00121.
[19] Li Yujie, Shao Zezhi, Xu Yongjun, et al. Dynamic frequency domain graph convolutional network for traffic forecasting[C]//Proceedings of the ICASSP 2024 - 2024 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). Piscataway: IEEE, 2024: 5245-5249.
[20] Li Yaguang, Shahabi C. A brief overview of machine learning methods for short-term traffic forecasting and future directions[J]. SIGSPATIAL Special, 2018, 10(1): 3-9.
[21] Shin Y, Yoon Y. PGCN: progressive graph convolutional networks for spatial-temporal traffic forecasting[J]. IEEE Transactions on Intelligent Transportation Systems, 2024, 25(7): 7633-7644.
[22] Bai Lei, Bai Lei, Yao Lina, et al. Adaptive graph convolutional recurrent network for traffic forecasting[C]//Proceedings of the 34th International Conference on Neural Information Processing Systems. New York: ACM, 2020: 17804-17815.

备注/Memo

备注/Memo:
收稿日期:2026-01-20;修订日期:2026-03-29
基金项目:河南省重点研 发专项(231111211600) ;河南省交通运输厅科技项目(2019G3)
作者简介:张震(1966— ) ,男,河南郑州人,郑州大学教授,博士,博士生导师,主要从事计算机视觉、交通流量预测的研究，E-mail:zhangzhen66@126.com。
更新日期/Last Update: 2026-04-03