Key detection node identification method for expressways based on multi-scale temporal graph convolutional network model
Article Text (Baidu Translation)
-
摘要: 为提升高速公路交通检测数据质量并优化检测器布局，提出了一种基于多尺度时序图卷积网络(MT-GCN)的关键节点识别方法；融合了多尺度时序分析与自适应扩张卷积以增强模型对短期波动和长期趋势的学习能力，改进了图卷积网络学习交通网络拓扑结构，以捕捉关键节点的空间交互关系，结合梯度重要性分析筛选最具代表性的关键检测节点；设计了2组对比试验以验证方法有效性，并设计了消融试验分析多尺度时序分析与传统图卷积网络(GCN)空间特征学习的具体贡献。研究结果表明：MT-GCN在所有节点覆盖率下均取得最小误差，与Traffic Former结合时组合表现最优，60%节点覆盖率下平均绝对误差为2.08 km·h-1、平均绝对百分比误差为6.25%，80%节点覆盖率下平均绝对误差为1.42 km·h-1、平均绝对百分比误差为4.91%；关键节点覆盖率在60%~65%时，可实现性能与资源的最优平衡；消融试验显示了完整MT-GCN性能优于仅用GCN或多尺度时序分析的模型，如在80%节点覆盖率下与时空图神经网络(ST-GNN)结合时，MT-GCN的平均绝对误差为1.59 km·h-1，而多尺度时序分析模型和GCN模型的平均绝对误差分别为1.89和2.02 km·h-1；与其他方法相比，MT-GCN在全局交通流表征方面更优，即便与性能较弱的估计方法结合仍能保持较低误差率。Abstract: To improve the quality of expressway traffic detection data and optimize the layout of detectors, a key node identification method based on Multi-scale Temporal Graph Convolutional Network (MT-GCN) was proposed; it integrated multi-scale temporal analysis with adaptive dilated convolution to enhance the capacity for learning both short-term fluctuations and long-term trends. It also enhanced the graph convolutional network to learn the topological structure of the traffic network, so as to capture the spatial interaction relationships among key nodes. Combining gradient importance analysis, the network screened out the most representative key detection nodes. Two sets of comparative experiments were designed for the verification of the method's effectiveness, and ablation experiments were performed for the analysis of the contributions of multi-scale temporal analysis and GCN-based spatial feature learning. The research results show that MT-GCN achieves the smallest error under all node coverage rates, and the combination of MT-GCN with Traffic Former performs the best. Under a 60% node coverage rate, the mean absolute error (MAE) is 2.08 km·h-1, and the mean absolute percentage error (MAPE) is 6.25%. Under an 80% node coverage rate, the MAE is 1.42 km·h-1and the MAPE is 4.91%. When the key node coverage rate is in the range of 60%-65%, the optimal balance between performance and resources can be achieved. The ablation experiments show that the performance of the complete MT-GCN is better than that of the models using only GCN or multi-scale temporal analysis. For example, when combined with Spatio-Temporal Graph Neural Network (ST-GNN) under an 80% node coverage rate, the MAE of MT-GCN is 1.59 km·h-1, while the MAEs of the multi-scale temporal analysis model and the GCN model are 1.89 and 2.02 km·h-1, respectively. MT-GCN performs better in representing overall traffic flow than other methods, and can maintain low error rates even when combined with estimation methods that have relatively weak performance.
-
图 3 不同关键节点覆盖率与识别方法对MAE的影响
Figure 3. Impact of different key node coverage rates and identification methods on MAE
图 4 不同关键节点覆盖率与识别方法对MAPE的影响
Figure 4. Impact of different key node coverage rates and identification methods on MAPE
图 5 关键节点识别与交通状态估计方法组合的速度误差对比
Figure 5. Comparison of speed errors under combinations of key node identification and traffic state estimation methods
图 6 不同关键节点覆盖率下速度估计误差对比
Figure 6. Comparison of speed errors under different key node coverage rates
表 1 高速公路检测器统计数据
Table 1. Statistical data of expressway detector
高速公路名称 距离/km 主线检测器数量 进口匝道检测器数量 出口匝道检测器数量 检测器总数量 I5-S 16.1 33 31 14 78 I80-E 7.5 16 9 7 32 SR99-S 5.6 0 4 0 4 US50-E 7.8 17 15 3 35 总计 37.0 66 59 24 149 
下载: 导出CSV
表 2 超参数设计
Table 2. Design of hyperparameters
超参数 卷积核 卷积路径 特征划分段数 特征段长度 GCN层数 隐藏层特征维度 学习率 数值 3, 5, 7, 9, 11 5 6 6, 12, 24, 48, 92, 288 2 128 0.001
下载: 导出CSV
表 3 基于MT-GCN的消融试验MAE结果
Table 3. Results of MAE in ablation experiments based on MT-GCN
km·h-1 方法 60%覆盖率时的速度MAE 80%覆盖率时的速度MAE MT-GCN Multi-scale GCN MT-GCN Multi-scale GCN LSTM 4.05 2.39 2.66 3.07 3.68 3.93 GE-GAN 3.18 3.72 4.13 2.29 2.75 2.93 GraphWaveNet 2.76 3.17 3.51 1.92 2.11 2.30 ST-GNN 2.35 2.66 3.03 1.59 1.89 2.02 TrafficFormer 2.08 2.39 2.66 1.42 1.60 1.79
下载: 导出CSV
表 4 基于MT-GCN的消融试验MAPE结果
Table 4. Results+of MAPE in ablation experiments based on MT-GCN
% 方法 60%覆盖率时的速度MAPE 80%覆盖率时的速度MAPE MT-GCN Multi-scale GCN MT-GCN Multi-scale GCN LSTM 9.85 12.02 12.61 7.82 9.31 10.17 GE-GAN 8.40 9.91 10.50 5.92 6.93 7.99 GraphWaveNet 7.80 9.05 10.06 5.92 6.93 7.99 ST-GNN 7.05 8.46 9.17 5.32 6.38 6.70 TrafficFormer 6.60 8.12 8.32 4.91 6.09 6.38
下载: 导出CSV
-
[1] CHEN H R, ZHOU R Y, CHEN H, et al. A resilience-oriented evaluation and identification of critical thresholds for traffic congestion diffusion[J]. Physica A: Statistical Mechanics and Its Applications, 2022, 600: 127592. doi: 10.1016/j.physa.2022.127592 [2] FUJITO I, MARGIOTTA R, HUANG W M, et al. Effect of sensor spacing on performance measure calculations[J]. Transportation Research Record: Journal of the Transportation Research Board, 2006(1945): 1-11. [3] FEI X, MAHMASSANI H S, EISENMAN S M. Sensor coverage and location for real-time traffic prediction in large-scale networks[J]. Transportation Research Record: Journal of the Transportation Research Board, 2007, 2039(1): 1-15. doi: 10.3141/2039-01 [4] LAI Q, ZHANG H H. Analysis of identification methods of key nodes in transportation network[J]. Chinese Physics B, 2022, 31(6): 068905. doi: 10.1088/1674-1056/ac4a6c [5] GENDREAU M, LAPORTE G, PARENT I. Heuristics for the location of inspection stations on a network[J]. Naval Research Logistics: NRL, 2000, 47(4): 287-303. doi: 10.1002/(SICI)1520-6750(200006)47:4<287::AID-NAV2>3.0.CO;2-R [6] BAN X, HERRING R, MARGULICI J D, et al. Optimal sensor placement for freeway travel time estimation[M]Berlin: Springer, 2009. [7] DANCZYK A, LIU H X. A mixed-integer linear program for optimizing sensor locations along freeway corridors[J]. Transportation Research Part B: Methodological, 2011, 45(1): 208-217. [8] AFRIN T, YODO N. A survey of road traffic congestion measures towards a sustainable and resilient transportation system[J]. Sustainability, 2020, 12(11): 4660. [9] HE Z G, GUO J N, XU J X. Cascade failure model in multimodal transport network risk propagation[J]. Mathematical Problems in Engineering, 2019, 2019: 3615903. [10] GUAN L C, WANG D L, SHAO H, et al. Understanding the topology of the road network and identifying key bayonet nodes to avoid traffic congestion[J]. International Journal of Modern Physics C, 2023, 34(3): 2350031. [11] WANG L W, YAN X D, LIU Y, et al. Grid mapping for road network abstraction and traffic congestion identification based on probe vehicle data[J]. Journal of Transportation Engineering, Part A: Systems, 2021, 147(5): 04021024. [12] BONACICH P. Some unique properties of eigenvector centrality[J]. Social Networks, 2007, 29(4): 555-564. [13] NIRMALA P, NADARAJAN R. Cumulative centrality index: centrality measures based ranking technique for molecular chemical structural graphs[J]. Journal of Molecular Structure, 2022, 1247: 131354. [14] QIN Q, WANG D X. Evaluation method for node importance in complex networks based on eccentricity of node[C]//IEEE. 2016 2nd IEEE International Conference on Computer and Communications (ICCC). New York: IEEE, 2016: 2499-2502. [15] WU X G. Identify influential nodes in complex networks based on modified TOPSIS[C]//IEEE. 2017 36th Chinese Control Conference (CCC). New York: IEEE, 2017: 1474-1479. [16] DU Z Y, TANG J J, QI Y, et al. Identifying critical nodes in metro network considering topological potential: a case study in Shenzhen city: China[J]. Physica A: Statistical Mechanics and Its Applications, 2020, 539: 122926. [17] RAO C J, GAO Y. Evaluation mechanism design for the development level of urban-rural integration based on an improved TOPSIS method[J]. Mathematics, 2022, 10(3): 380. [18] HU P, FAN W L, MEI S W. Identifying node importance in complex networks[J]. Physica A: Statistical Mechanics and Its Applications, 2015, 429: 169-176. [19] AGRYZKOV T, OLIVER J L, TORTOSA L, et al. An algorithm for ranking the nodes of an urban network based on the concept of PageRank vector[J]. Applied Mathematics and Computation, 2012, 219(4): 2186-2193. [20] LIU J, XIONG Q Y, SHI W R, et al. Evaluating the importance of nodes in complex networks[J]. Physica A: Statistical Mechanics and Its Applications, 2016, 452: 209-219. [21] AI X B. Node importance ranking of complex networks with entropy variation[J]. Entropy, 2017, 19(7): 303. [22] WANG L, YAN P Z, LI Y H, et al. Signal sub-control-area division of traffic complex network based on nodes importance assessment[C]//IEEE. 2011 30th Chinese Control Conference (CCC). New York: IEEE, 2011: 5606-5609. [23] HUANG X L, CHEN J, CAI M, et al. Traffic node importance evaluation based on clustering in represented transportation networks[J]. IEEE Transactions on Intelligent Transportation Systems, 2022, 23(9): 16622-16631. [24] HUANG X L, HU S, WANG W, et al. Identifying critical links in urban transportation networks based on spatio-temporal dependency learning[J]. IEEE Transactions on Intelligent Transportation Systems, 2024, 25(6): 5583-5597. [25] YANG F, YAN F, ZHANG C K, et al. Applying the virtual input-output method to the identification of key nodes in busy traffic network[J]. Complexity, 2021, 2021(1): 5559857. [26] WANG L J, ZHANG S C, SZVCS G, et al. Identifying the critical nodes in multi-modal transportation network with a traffic demand-based computational method[J]. Reliability Engineering & System Safety, 2024, 244: 109956. [27] CHEN J, WANG W, YU K P, et al. Node connection strength matrix-based graph convolution network for traffic flow prediction[J]. IEEE Transactions on Vehicular Technology, 2023, 72(9): 12063-12074. [28] LV Z Q, CHENG Z S, LI J B, et al. TreeCN: time series prediction with the tree convolutional network for traffic prediction[J]. IEEE Transactions on Intelligent Transportation Systems, 2024, 25(5): 3751-3766. [29] WEN Y J, LI Z H, WANG X Y, et al. Traffic demand prediction based on spatial-temporal guided multi graph Sandwich-Transformer[J]. Information Sciences, 2023, 643: 119269. [30] ZHANG K P, ZHAO L, DONG C X, et al. AI-TP: attention-based interaction-aware trajectory prediction for autonomous driving[J]. IEEE Transactions on Intelligent Vehicles, 2023, 8(1): 73-83. [31] ZHANG Y P, GILL G S, CHENG W, et al. Exploring influential factors and endogeneity of traffic flow of different lanes on urban freeways using Bayesian multivariate spatial models[J]. Journal of Traffic and Transportation Engineering (English Edition), 2023, 10(1): 104-115. [32] YE Y, JI S H. Sparse graph attention networks[J]. IEEE Transactions on Knowledge and Data Engineering, 2023, 35(1): 905-916. [33] LIANG W W, ZHANG W. Learning social relations and spatiotemporal trajectories for next check-in inference[J]. IEEE Transactions on Neural Networks and Learning Systems, 2023, 34(4): 1789-1799. [34] ZHAO Y, LU W Q, RUI Y K, et al. Classification of the traffic status subcategory with ETC gantry data: an improved support tensor machine approach[J]. Journal of Advanced Transportation, 2023, 2023: 2765937. [35] TIAN C Y, CHAN W K. 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. [36] 曲栩, 甘锐, 安博成, 等. 基于广义时空图卷积网络的交通群体运动态势预测[J]. 交通运输工程学报, 2022, 22(3): 79-88. doi: 10.19818/j.cnki.1671-1637.2022.03.006QU Xu, GAN Rui, AN Bo-cheng, et al. Prediction of traffic swarm movement situation based on generalized spatio-temporal graph convolution network[J]. Journal of Traffic and Transportation Engineering, 2022, 22(3): 79-88. doi: 10.19818/j.cnki.1671-1637.2022.03.006 [37] 史昕, 胡欣倩, 赵祥模, 等. 基于神经常微分方程的自适应图时空同步交通流预测方法[J]. 交通运输工程学报, 2025, 25(2): 170-188. doi: 10.19818/j.cnki.1671-1637.2025.02.011SHI Xin, HU Xin-qian, ZHAO Xiang-mo, et al. Adaptive graph spatio-temporal synchronization for traffic flow prediction based on NODEs[J]. Journal of Traffic and Transportation Engineering, 2025, 25(2): 170-188. doi: 10.19818/j.cnki.1671-1637.2025.02.011 [38] 姚俊峰, 何瑞, 史童童, 等. 基于机器学习的交通流预测方法综述[J]. 交通运输工程学报, 2023, 23(3): 44-67. doi: 10.19818/j.cnki.1671-1637.2023.03.003YAO Jun-feng, HE Rui, SHI Tong-tong, et al. Review on machine learning-based traffic flow prediction methods[J]. Journal of Traffic and Transportation Engineering, 2023, 23(3): 44-67. doi: 10.19818/j.cnki.1671-1637.2023.03.003 -
图(6) / 表(4)
计量
- 文章访问数: 214
- HTML全文浏览量: 57
- PDF下载量: 21
- 被引次数: 0
