Traffic flow prediction based on random matrix-based dynamic spatio-temporal network
Article Text (Baidu Translation)
-
摘要: 为了更灵活地建模交通数据中复杂多变的时空结构并提升其对异常交通模式的识别能力，提出了一种新型的随机矩阵动态时空网络(Random Matrix-based Dynamic Spatiotemporal Network, RM-DTSN)模型，该模型引入了时空随机矩阵嵌入机制，摒弃了对预定义邻接矩阵的依赖，能够根据输入的数据动态调整节点间的空间交互强度，从而更精准地表达节点间异构关系与动态空间结构；为了增强对时间序列依赖的建模能力，RM-DTSN设计了独立注意力机制，能更有效地捕捉不同时间步之间的短期与长期动态特征；此外，模型融合残差分解与门控机制，能够有效提取多层次的时空特征，不仅在保留关键信号的同时抑制噪声干扰，提升了对异常交通信号的鲁棒性，还缓解了深层网络中的梯度消失问题。研究结果表明，在关键交通数据集上，RM-DTSN取得了显著的性能提升：在PeMSD3数据集上，其均方根误差(RMSE)为24.79，相比经典时空图网络STGCN(30.42)降低了18.5%；平均绝对误差为14.38，较当前最优模型DDGCRN(14.63)降低了1.71%；在PeMSD8数据集上，RMSE为23.62，相比广泛应用的时序卷积模型TCN(35.79)显著降低了34.0%。试验结果充分验证了RM-DTSN在不同预测场景中的稳定性和泛化能力。RM-DTSN为复杂交通环境下的流量预测提供了一种高效且鲁棒的解决方案，在智慧交通、路径规划和城市调度等实际场景中展现出广阔的应用前景，特别适用于应对高维交通数据中的突发拥堵、线路异常等复杂预测任务。Abstract: In order to flexibly model the complex and variable spatiotemporal structures in traffic data and enhance the identification ability of abnormal traffic patterns, a novel model named random matrix-based dynamic spatiotemporal network (RM-DTSN) was proposed. A spatiotemporal random matrix embedding mechanism was introduced, and the dependence on a predefined adjacency matrix was discarded to dynamically adjust the spatial interaction strength between nodes according to the input data, thereby more accurately expressing heterogeneous relationships and dynamic spatial structures among nodes. In order to enhance the ability to model temporal sequence dependencies, an independent attention mechanism was designed to capture short-term and long-term dynamic features between different time steps more effectively. In addition, residual decomposition and gating mechanisms were integrated to effectively extract multi-level spatiotemporal features, not only retaining key signals while suppressing noise interference to improve robustness to abnormal traffic signals but also alleviating the gradient-vanishing problem in deep networks. Experimental results show that on key traffic datasets, significant performance improvements are achieved by RM-DTSN. On the PeMSD3 dataset, its RMSE is 24.79, which is 18.5% lower than that of the classical spatiotemporal graph network STGCN (30.42). Its MAE is 14.38, which is 1.71% lower than that of the current state-of-the-art model DDGCRN (14.63). On the PeMSD8 dataset, its RMSE is 23.62, showing a significant reduction of 34.0% compared to that of the widely used temporal convolutional network TCN (35.79). The stability and generalization capability of RM-DTSN in different prediction scenarios are fully verified by the above results. An efficient and robust solution for flow prediction in complex traffic environments is provided, and broad application prospects of RM-DTSN are shown in real scenarios such as intelligent transportation, route planning, and urban dispatching. It is particularly suitable for handling complex prediction tasks such as sudden congestion and route anomalies in high-dimensional traffic data.
-
图 4 PeMSD8数据集上与先进模型预测误差对比
Figure 4. Comparison of prediction errors with advanced models on the PeMSD8 datasets
表 1 数据集描述与统计
Table 1. Description and statistics of the datasets
数据集 节点数 记录数 时间间隔/min 时间跨度 PeMSD3 358 26 208 5 2018.09~2018.11 PeMSD4 307 16 992 5 2018.01~2018.02 PeMSD8 170 17 856 5 2016.07~2016.08 
下载: 导出CSV
表 2 RM-DTSN和3个交通数据集上的基线模型比较
Table 2. Comparison of RM-DTSN and baseline models on three traffic datasets
模型 PeMSD3 PeMSD4 PeMSD8 MAE RMSE MAPE/% MAE RMSE MAPE/% MAE RMSE MAPE/% HA[29] 31.58 52.39 33.78 38.03 59.24 27.88 34.86 59.24 27.88 ARIMA[30] 35.41 47.59 33.78 33.73 48.80 24.18 31.09 44.32 22.73 VAR[29] 23.65 38.26 24.51 24.54 38.61 17.24 19.19 29.81 13.10 FC-LSTM[31] 21.33 35.11 23.33 26.77 40.65 18.23 19.19 29.81 13.10 TCN[32] 19.32 33.55 19.93 23.22 37.26 15.59 22.72 35.79 14.03 GRU-ED[32] 19.12 32.85 19.31 23.68 39.27 16.44 22.00 36.22 13.33 DSANet[33] 21.29 34.55 23.21 22.79 35.77 16.03 17.14 29.96 11.32 STGCN[34] 17.55 30.42 17.34 21.16 34.89 13.83 17.50 27.09 11.29 DCRNN[16] 17.99 30.31 18.34 21.22 33.44 14.17 16.82 26.36 10.92 Graph WaveNet[17] 19.12 32.77 18.89 24.89 39.66 17.29 18.28 30.05 12.15 ASTGCN[35] 17.34 29.56 17.21 22.92 35.22 16.56 18.25 28.06 11.64 MSTGCN[35] 19.54 31.93 23.86 23.96 37.21 14.33 19.00 29.15 12.38 STG2Seq[34] 19.03 29.83 21.55 25.20 38.48 18.77 20.17 30.71 17.32 AGCRN[36] 15.98 28.25 15.23 19.83 32.26 12.97 15.95 25.22 10.09 STFGNN[37] 16.77 28.34 16.30 20.48 32.51 16.77 16.94 26.25 10.60 STGODE[38] 16.50 27.84 16.69 20.84 32.82 13.77 22.59 37.54 10.14 Z-GCNETs[21] 16.64 28.15 16.39 19.50 31.61 12.78 15.76 25.11 10.01 STG-NCDE[39] 15.57 27.09 15.06 19.21 31.09 12.76 15.45 24.81 9.92 STG-NRDE[40] 15.50 27.06 14.90 19.13 30.94 12.68 15.32 24.72 8.90 MAGCRN[41] 15.10 26.28 14.08 19.04 31.20 12.45 15.46 25.03 9.79 DAST-Transformer[42] 18.44 31.30 12.92 14.58 24.90 9.01 DDGCRN[23] 14.63 25.07 14.22 18.45 30.51 12.19 14.40 23.75 9.40 RM-DTSN 14.38 24.79 14.26 18.35 30.44 12.19 14.31 23.62 9.36
下载: 导出CSV
表 3 PeMSD4和PeMSD8数据集的训练时间
Table 3. The computation time on PeMSD4 and PeMSD8 datasets
数据集 模型 训练时间/(s·数据集-1) PeMSD4 AGCRN 6.5 STG-NCDE 118.6 RM-DTSN 59.7 PeMSD8 AGCRN 3.9 STG-NCDE 43.2 RM-DTSN 38.1
下载: 导出CSV
表 4 PeMSD4和PeMSD3的消融试验
Table 4. Ablation experiments on PeMSD4 and PeMSD3
数据集 评估指标 RM-DTSN w/o RM w/o RG w/o DGCN w/o AT PeMSD4 MAE 18.35 18.48 18.40 27.49 18.57 RMSE 30.44 30.72(↓0.92%) 30.50 43.38 30.45 MAPE/% 12.19 12.30 12.28(↓0.74) 19.65(↓61.2) 12.36(↓1.39) PeMSD3 MAE 14.38 14.48 14.50 15.17 14.62(↓1.67%) RMSE 24.79 25.07(↓1.13%) 24.10 27.06(↓9.16%) 24.73 MAPE/% 14.26 14.26 14.81(↓3.86) 14.70 14.42
下载: 导出CSV
-
[1] XUE S, LU J, WU J, et al. Multi-instance graphical transfer clustering for traffic data learning[C]//IEEE. 2016 International Joint Conference on Neural Networks (IJCNN). New York: IEEE, 2016: 4390-4395. [2] 程国柱, 李金禹, 陈永胜, 等. 高速公路异构交通流HDV建模及其特征[J]. 长安大学学报(自然科学版), 2024, 44(4): 97-107.CHENG Guo-zhu, LI Jin-yu, CHEN Yong-sheng, et al. Modeling of human-driven vehicles and characteristics of heterogeneous traffic flow for freeway[J]. Journal of Chang'an University (Natural Science Edition), 2024, 44(4): 97-107. [3] 刘静, 关伟. 交通流预测方法综述[J]. 公路交通科技, 2004, 21(3): 82-85.LIU Jing, GUAN Wei. A summary of traffic flow forecasting methods[J]. Journal of Highway and Transportation Research and Development, 2004, 21(3): 82-85. [4] VAN DER VOORT M, DOUGHERTY M, WATSON S. Combining kohonen maps with ARIMA time series models to forecast traffic flow[J]. Transportation Research Part C: Emerging Technologies, 1996, 4(5): 307-318. doi: 10.1016/S0968-090X(97)82903-8 [5] VAN LINT J W C, VAN HINSBERGEN C. Short-term traffic and travel time prediction models[J]. Artificial Intelligence Applications to Critical Transportation Issues, 2012, 22(1): 22-41. [6] FU R, ZHANG Z, LI L. Using LSTM and GRU neural network methods for traffic flow prediction[C]//IEEE. 2016 31st Youth Academic Annual Conference of Chinese Association of Automation (YAC). New York: IEEE, 2016: 324-328. [7] JEONG Y S, BYON Y J, CASTRO-NETO M M, et al. Supervised weighting-online learning algorithm for short-term traffic flow prediction[J]. IEEE Transactions on Intelligent Transportation Systems, 2013, 14(4): 1700-1707. doi: 10.1109/TITS.2013.2267735 [8] BAI J D, ZHU J W, SONG Y J, et al. A3T-GCN: Attention temporal graph convolutional network for traffic forecasting[J]. ISPRS International Journal of Geo-Information, 2021, 10(7): 485. doi: 10.3390/ijgi10070485 [9] SUN K, ZHU Z, LIN Z. ADAGCN: Adaboosting graph convolutional networks into deep models[C]//ICLR. The 9th International Conference on Learning Representations. Appleton: ICRA, 2021: 1-15. [10] WU Z H, PAN S R, CHEN F W, et al. A comprehensive survey on graph neural networks[J]. IEEE Transactions on Neural Networks and Learning Systems, 2021, 32(1): 4-24. doi: 10.1109/TNNLS.2020.2978386 [11] LIN Z Q, FENG J, LU Z Y, et al. DeepSTN+: Context-aware spatial-temporal neural network for crowd flow prediction in metropolis[C]//AAAI. The AAAI Conference on Artificial Intelligence. Washington DC: AAAI, 2019, 33(1): 1020-1027. [12] ZHANG J B, ZHENG Y, QI D K. Deep spatio-temporal residual networks for citywide crowd flows prediction[C]//AAAI. The 31th AAAI Conference on Artificial Intelligence. Washington DC: AAAI, 2017, 31(1): 1-7. [13] LI F X, 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. [14] DEFFERRARD M, BRESSON X, VANDERGHEYNST P. Convolutional neural networks on graphs with fast localized spectral filtering[C]// NeurIPS. NIPS'16: Proceedings of the 30th International Conference on Neural Information Processing Systems. Denver: NeurIPS, 2016: 3844-3852. [15] KIPF T N, WELLING M. Semi-supervised classifi-cation with graph convolutional networks[C]//ICLR. The 5th International Conference on Learning Representations. Appleton: ICLR, 2016, 1-14. [16] LI Y G, YU R, SHAHABI C, et al. Diffusion convolutional recurrent neural network: data-driven traffic forecasting[C]//ICLR. The 6th International Conference on Learning Representations. Appleton: ICLR. 2017: 1-16. [17] WU Z H, PAN S R, LONG G D, et al. Graph wavenet for deep spatial-temporal graph modeling[C]//ICLR. The 7th International Conference on Learning Representations. Appleton: ICLR, 2019: 1-7. [18] YU B, YIN H T, ZHU Z X. Spatio-temporal graph convolutional networks: A deep learning framework for traffic forecasting[C]// IJCAI. The 27th International Joint Conference on Artificial Intelligence. Boston: IJCAI, 2017: 1-7. [19] SONG C, LIN Y F, GUO S N, et al. Spatial-temporal synchronous graph convolutional networks: A new framework for spatial-temporal network data forecasting[C]//AAAI. The 34th AAAI Conference on Artificial Intelligence. Washington DC: AAAI, 2020, 34(1): 914-921. [20] ZHENG C P, FAN X L, WANG C, et al. GMAN: A graph multi-attention network for traffic prediction[C]//AAAI. The 34th AAAI Conference on Artificial Intelligence. Washington DC: AAAI, 2020, 34(1): 1234-1241. [21] CHEN Y, SEGOVIA I, GEL Y R. Z-GCNETs: Time zigzags at graph convolutional networks for time series forecasting[C]// PMLR. The 38th International Conference on Machine Learning. New York: PMLR, 2021: 1684-1694. [22] WU Z H, PAN S R, LONG G D, et al. Connecting the dots: Multivariate time series forecasting with graph neural networks[C]//ACM. The 26th ACM International Conference on Knowledge Discovery & Data Mining. New York: ACM, 2020: 753-763. [23] WENG W C, FAN J, WU H F, et al. A Decomposition Dynamic graph convolutional recurrent network for traffic forecasting[J]. Pattern Recognition, 2023, 142: 109670. doi: 10.1016/j.patcog.2023.109670 [24] YU C J, MA X, REN J W, et al. Spatio-temporal graph transformer networks for pedestrian trajectory prediction[C]// Springer. Computer Vision-ECCV 2020. Berlin: Springer, 2020: 507-523. [25] CAO D, WANG Y, DUAN J, et al. Spectral temporal graph neural network for multivariate time-series forecasting[J]. Advances in Neural Information Processing Systems, 2020, 33: 17766-17778. [26] SHAO Z Z, ZHANG Z, WANG F, et al. Spatial-temporal identity: a simple yet effective baseline for multivariate time series forecasting[C]//ACM. The 31st ACM International Conference on Information & Knowledge Management. New York: ACM, 2022: 4454-4458. [27] JIANG J W, HAN C K, ZHAO W X, et al. PDFormer: Propagation delay-aware dynamic long-range transformer for traffic flow prediction[C]//AAAI. The 37th AAAI Conference on Artificial Intelligence. Washington DC: AAAI, 2023, 37(4): 4365-4373. [28] CHEN C, PETTY K, SKABARDONIS A, et al. Freeway performance measurement system: Mining loop detector data[J]. Transportation Research Record: Journal of the Transportation Research Board, 2001, 1748(1): 96-102. doi: 10.3141/1748-12 [29] HAMILTON J D. Bivariate granger causality tests[M]. Princeton: Princeton University Press, 2020. [30] WILLIAMS B M, HOEL L A. Modeling and forecasting vehicular traffic flow as a seasonal ARIMA process: Theoretical basis and empirical results[J]. Journal of Transportation Engineering, 2003, 129(6): 664-672. doi: 10.1061/(ASCE)0733-947X(2003)129:6(664) [31] SUTSKEVER I, VINYALS O, LE Q V. Sequence to sequence learning with neural networks[C]//NeurIPS. NIPS'16: Proceedings of the 28th International Conference on Neural Information Processing Systems. Denver: NeurIPS, 2014, 1-9. [32] YU F, KOLTUN V. Multi-scale context aggregation by dilated convolutions[C]//ICLR. The 4th International Conference on Learning Representations. New York: ICLR, 2015: 1-13. [33] HUANG S T, WANG D L, WU X H, et al. DSANet: Dual self-attention network for multivariate time series forecasting[C]// ACM. The 28th ACM International Conference on Information and Knowledge Management. New York: ACM, 2019: 2129-2132. [34] BAI L, YAO L, KANHERE S, et al. Stg2seq: Spatial-temporal graph to sequence model for multi-step passenger demand forecasting[C]//AAAI. The 28th AAAI Conference on Artificial Intelligence. Washington DC: AAAI, 2019: 1-5. [35] GUO S N, LIN Y F, FENG N, et al. Attention based spatial-temporal graph convolutional networks for traffic flow forecasting[C]//AAAI. The 34th AAAI Conference on Artificial Intelligence. Washington DC: AAAI, 2020, 33(1): 922-929. [36] 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. [37] LI M Z, ZHU Z X. Spatial-temporal fusion graph neural networks for traffic flow forecasting[C]//AAAI. The 35th AAAI Conference on Artificial Intelligence. Washington DC: AAAI, 2021, 35(5): 4189-4196. [38] FANG Z, LONG Q Q, SONG G J, et al. Spatial-temporal graph ODE networks for traffic flow forecasting[C]//ACM. The 27th ACM International Conference on Knowledge Discovery & Data Mining. New York: ACM, 2021: 364-373. [39] CHOI J, CHOI H, HWANG J, et al. Graph neural controlled differential equations for traffic forecasting[C]//AAAI. The 36th AAAI Conference on Artificial Intelligence. Washington DC: AAAI, 2022, 36(6): 6367-6374. [40] CHOI J, PARK N. Graph neural rough differential equations for traffic forecasting[J]. Transactions on Intelligent Systems and Technology, 2023, 14(4): 1-27. [41] ZEB A, YE Y, ZHANG S, et al. Meta attentive graph convolutional recurrent network for traffic forecasting[J]. Journal of Latex Class Files, 2015, 14(8): 1-15. [42] 孔文翔. 基于空间信息迟延感知的时空Transformer交通流预测[J]. 软件工程与应用, 2023, 12(2): 293-302.KONG Wen-xiang. Spatio-temporal Transformer traffic flow prediction based on spatial delay-aware information[J]. Software Engineering and Applications, 2023, 12(2): 293-302. -
图(4) / 表(4)
计量
- 文章访问数: 149
- HTML全文浏览量: 74
- PDF下载量: 15
- 被引次数: 0
