| 261 | 0 | 53 |
| 下载次数 | 被引频次 | 阅读次数 |
【背景】时空图神经网络是当前交通流预测领域中最具代表性的建模方法，但现有方法多依赖于单一的静态图、自适应图或动态图机制，难以区分不同来源的空间依赖，不能兼顾稳定空间关系与动态空间交互的建模需求。【目标】针对单一空间建模机制难以有效刻画复杂时空相关性的问题，构建一种使用多空间建模机制的时空预测模型。【方法】提出一种时空差分记忆注意力模型STDMA，通过引入记忆原型存储长期典型交通模式，并结合静态图卷积与差分记忆注意力，实现静态空间依赖与短期动态空间交互的解耦建模。此外，使用时间卷积捕捉时间相关性并压缩序列长度，同时使用时间嵌入自适应地捕获时间序列的趋势与周期模式。【数据】采用SD、PEMS03、PEMS04和PEMS08四个来源于加州交通局性能测量系统的真实交通流数据集。【结果】实验结果表明，STDMA在平均绝对误差、均方根误差和平均绝对百分比误差这三个常用评价指标上整体优于11种先进基线方法。同时，在计算效率和GPU 内存占用上均具有显著优势。
Abstract：[Background] Spatio-temporal graph neural networks (STGNNs) are among the most representative modeling approaches in traffic flow prediction. However， existing methods mostly rely on a single mechanism， static graphs， adaptive graphs， or dynamic graphs， which makes it difficult to distinguish spatial dependencies from different sources and to simultaneously model stable spatial relationships and dynamic spatial interactions. [Objective] To address the limitation of single spatial modeling mechanisms in effectively capturing complex spatiotemporal correlations， this study aims to develop a spatiotemporal forecasting model that integrates multiple spatial modeling mechanisms. [Method] A Spatio-temporal Differential Memory Attention Model (STDMA) is proposed that explicitly decouples static spatial dependencies from short-term dynamic spatial interactions. Specifically， STDMA introduces a memory prototype module to store long-term representative traffic patterns， combines static graph convolution with differential memory attention， and integrates temporal convolution to capture temporal dependencies and compress sequence length. Furthermore， a temporal embedding mechanism is employed to adaptively capture trend and periodic patterns in the time series. [Data] Four real-world traffic flow datasets， namely SD， PEMS03， PEMS04， and PEMS08， collected from the California Department of Transportation Performance Measurement System (PeMS) are adopted. [Results] Experimental results demonstrate that STDMA consistently outperforms eleven state-of-the-art baseline methods across three standard evaluation metrics: Mean Absolute Error， Root Mean Square Error， and Mean Absolute Percentage Error. Moreover， it also exhibits significant advantages in both computational efficiency and GPU memory consumption.
[1] 程泽阳， 孙凌霞， 丁恒， 等. 车路协同环境下道路交通安全研究进展[J]. 交通运输工程与信息学报， 2024， 22(3): 14-33.
[2] 李鹏博，林之喆，张衡，等.用于交通流预测的元参数特征融合动态图网络[J].交通运输工程与信息学报:1-22 (2025-10-25)[2025-12-08]. https://doi.org/10.19961/j.cnki.1672-4747.2025.09.019.
[3] 殷炽磊，林之喆，周腾，等.融合自适应图与时空Transformer的交通流预测模型[J].交通运输工程与信息学报.DOI:10.19961/j.cnki.1672-4747.2025.09.002.
[4] 汤银英，朱星龙，李龙.基于SARIMA模型的铁路月度客运量预测[J].交通运输工程与信息学报，2019，17(1):25-32.
[5] LIU A， ZHANG Y. Spatial–temporal dynamic graph convolutional network with interactive learning for traffic forecasting[J]. IEEE Transactions on Intelligent Transportation Systems， 2024， 25(7): 7645-7660.
[6] 胡晓伟， 吴则洋， 卢泓博， 等. 基于遗传算法优化深度神经网络的站点客流预测[J]. 交通运输工程与信息学报， 2025， 23(1): 72-84.
[7] ZHANG J， ZHENG Y， QI D. Deep spatio-temporal residual networks for citywide crowd flows prediction[J]. Proceedings of the AAAI Conference on Artificial Intelligence， 2017， 31(1).
[8] ZHANG H， LIN Z， ZHOU J， et al. A comprehensive review of traffic flow prediction: from traditional models to deep learning architectures[J]. Digital Transportation and Safety， 2025， 4(4): 281-297.
[9] YU B， YIN H， ZHU Z. Spatio-temporal graph convolutional networks: a deep learning framework for traffic forecasting[C]//Proceedings of the 27th International Joint Conference on Artificial Intelligence， 2018: 3634-3640.
[10] LI Y， YU R， SHAHABI C， et al. Diffusion convolutional recurrent neural network: data-driven traffic forecasting [C]//International Conference on Learning Representations， 2018.
[11] BAI L， YAO L， LI C， et al. Adaptive graph convolutional recurrent network for traffic forecasting[J]. Advances in Neural Information Processing Systems， 2020， 33: 17804-17815.
[12] LI F， FENG J， YAN H， et al. Dynamic graph convolutional recurrent network for traffic prediction: benchmark and solution[J]. ACM Transactions on Knowledge Discovery from Data， 2023， 17(1): 1-21.
[13] LIU H， DONG Z， JIANG R， et al. Spatio-temporal adaptive embedding makes vanilla transformer sota for traffic forecasting[C]//Proceedings of the 32nd ACM International Conference on Information and Knowledge Management， 2023: 4125-4129.
[14] JIANG J， HAN C， ZHAO W X， et al. Pdformer: propagation delay-aware dynamic long-range transformer for traffic flow prediction[C]//Proceedings of the AAAI Conference on Artificial Intelligence， 2023， 37(4): 4365-4373.
[15] LIU X， XIA Y， LIANG Y， et al. Largest: A benchmark dataset for large-scale traffic forecasting[J]. Advances in Neural Information Processing Systems， 2023， 36: 75354-75371.
[16] Wright L， Demeure N. Ranger21: a synergistic deep learning optimizer[J]. arXiv preprint arXiv:2106.13731， 2021.
[17] WU Z， PAN S， LONG G， et al. Graph wavenet for deep spatial-temporal graph modeling[C]//Proceedings of the 28th International Joint Conference on Artificial Intelligence. 2019: 1907-1913.
[18] FANG Z， LONG Q， SONG G， et al. Spatial-temporal graph ode networks for traffic flow forecasting[C]//Proceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery & Data Mining， 2021: 364-373.
[19] LAN S， MA Y， HUANG W， et al. Dstagnn: dynamic spatial-temporal aware graph neural network for traffic flow forecasting[C]//International Conference on Machine Learning. PMLR， 2022: 11906-11917.
[20] SHAO Z， ZHANG Z， WEI W， et al. Decoupled dynamic spatial-temporal graph neural network for traffic forecasting[J]. Proceedings of the VLDB Endowment， 2022， 15(11): 2733-2746.
[21] CAO L， WANG B， JIANG G， et al. Spatiotemporal-aware trend-seasonality decomposition network for traffic flow forecasting[C]//Proceedings of the AAAI Conference on Artificial Intelligence， 2025， 39(11): 11463-11471.
[22] JIANG R， WANG Z， YONG J， et al. Spatio-temporal meta-graph learning for traffic forecasting[C]//Proceedings of the AAAI Conference on Artificial Intelligence， 2023， 37(7): 8078-8086.
[23] JIN G， LAI S， HAO X， et al. M3-net: a cost-effective graph-free mlp-based model for traffic prediction[C]//Proceedings of the 34th ACM International Conference on Information and Knowledge Management， 2025: 4847-4851.
基本信息:
DOI：10.19961/j.cnki.1672-4747.2026.01.006
中图分类号:TP18;U491.14
引用信息:
[1]杨博文,林之喆,谢海,等.用于交通流预测的时空差分记忆注意力模型[J].交通运输工程与信息学报().DOI:10.19961/j.cnki.1672-4747.2026.01.006.
基金信息:
国家自然科学基金项目（62462021）
2026-01-27
2026-01-27
2026-01-27