Continuum-flow-model-guided deep learning for freeway traffic state estimation
-
摘要:
针对现有物理信息深度学习(PIDL)交通状态估计研究在网络设计、交通流模型参数设定及匝道流量处理方面的不足,提出了新的PIDL交通状态估计研究框架;在数据驱动层面,引入非线性扩展层、注意力机制和点权重机制,提升网络表达能力;在模型驱动层面,构建专用的交通流模型参数估计网络,通过在线学习实现模型参数自适应辨识;在数模融合层面,设计匝道源项估计器,通过流量守恒正则化损失函数,增强交通流模型和数据的融合适配;基于3种典型宏观交通流模型,分别设计PIDL交通状态估计器;针对一条17.47 km长的城市快速路进行实证研究,评估对比3种PIDL交通状态估计器与5种代表性基线估计方法的性能,并完成相关消融试验。研究结果表明:PIDL交通状态估计性能显著优于基线方法,流量和速度的平均绝对百分比估计误差为10.22%和8.13%;所设计的非线性扩展层、注意力机制、点权重机制、模型参数估计网络及匝道源项估计器均与PIDL交通状态估计性能密切相关;使用匝道源项估计器将流量估计精度提升68%,使用模型参数估计网络可将速度估计精度提升51%。研究方法可适用于大规模高速公路网络交通状态估计。
Abstract:To address the shortcomings of existing physics-informed deep learning (PIDL) studies on traffic state estimation in network design, traffic flow model parameter settings, and ramp flow handling, a new PIDL-based traffic state estimation framework was proposed. At the data-driven level, nonlinear extension layers, an attention mechanism, and a point-weighting mechanism were introduced to enhance the network's representational capacity. At the model-driven level, a dedicated traffic flow model parameter estimation network was constructed to achieve adaptive identification of model parameters through online learning. At the data-model fusion level, a ramp source term estimator was designed to enhance the fusion adaptability between the traffic flow model and data via a flow-conservation-based regularization loss function. Based on three typical macroscopic traffic flow models, three PIDL-based traffic state estimators were designed, respectively. An empirical study was conducted on a 17.47-km urban expressway to evaluate and compare the performance of the three PIDL-based traffic state estimators with five representative baseline estimation methods, and the relevant ablation experiments were completed. The results show that the PIDL-based traffic state estimation performance is significantly superior to the baseline methods, with mean absolute percentage errors of 10.22% for flow and 8.13% for speed. The designed nonlinear extension layers, attention mechanism, point-weighting mechanism, model parameter estimation network, and ramp source term estimator are all closely related to the PIDL-based traffic state estimation performance. The ramp source term estimator improves flow estimation accuracy by 68%, and the model parameter estimation network improves speed estimation accuracy by 51%. The research methodology is applicable to traffic state estimation for large-scale freeway networks.
-
表 1 PIDL方法细节
Table 1. Details of the PIDL methods
估计方法 相关网络 对应网络深度/层 每层参数量/个 交通流模型参数及范围 PIDL-LWR PUNN、RSTE、MPEN 6、4、4 100 453、35 606、9 311 $ {v}_{\mathrm{f}} $∈[60, 120] km·h-1,$ {\rho }_{\mathrm{c}} $∈[15, 65] veh·km-1,$ a $∈[1,2.5] PIDL-ARZ 6、4、4 100 453、35 606、9 376 $ {v}_{\mathrm{f}} $∈[60, 120] km·h-1,$ {\rho }_{\mathrm{c}} $∈[15, 65] veh·km-1,$ a $∈[1,2.5],$ \tau $∈ [0.001,0.02] h PIDL-PW 6、4、4 100 453、35 606、9 441 $ {v}_{\mathrm{f}} $∈[60, 120] km·h-1,$ {\rho }_{\mathrm{c}} $∈[15, 65] veh·km-1,$ a $∈[1,2.5],$ \tau $∈[0.001,0.02] h,$ {c}^{2} $∈[10, 100](km·h-1)2 表 2 TSE方法性能
Table 2. Performance of TSE methods
方法 流量 速度 E1/% E2/
(veh∙h-1)E3/% E1/% E2/
(km∙h-1)E3/% Inter2d 22.18 403.3 18.11 11.89 7.56 10.84 AS 14.14 351.0 15.73 12.81 8.33 11.94 SVR 27.79 525.8 23.61 15.58 9.31 13.34 RF 13.73 343.8 15.39 15.27 10.17 14.57 NN 17.64 405.6 18.17 12.60 10.61 15.21 PIDL-LWR 11.61 184.2 8.26 8.60 7.78 11.15 PIDL-ARZ 10.22 160.8 7.22 9.28 7.39 10.59 PIDL-PW 10.81 166.8 7.47 8.13 7.01 10.04 表 3 不同天气条件下TSE方法的性能
Table 3. Performance of TSE methods under different weather conditions
方法 6月20日(小雨) 6月29日(大雨) 流量E3/% 速度E3/% 流量 E3/% 速度E3/% Inter2d 9.65 9.15 17.68 12.55 AS 9.23 10.25 10.09 9.81 SVR 21.82 13.42 21.04 12.98 RF 10.52 10.61 10.37 10.40 NN 11.45 11.14 15.51 11.75 PIDL-LWR 8.22 9.07 7.66 8.35 PIDL-ARZ 8.00 8.69 7.55 8.33 PIDL-PW 8.03 8.55 7.53 8.29 表 4 消融试验结果
Table 4. Results of ablation studies
方法 评价指标 训练用时/s 流量 速度 E1/% E2/(veh∙h-1) E3/% E1/% E2/(km∙h-1) E3/% PIDL-PW-1 11.21 177.0 7.92 9.55 8.98 12.86 1 467 PIDL-PW-2 11.37 170.4 7.63 9.04 7.45 10.66 1 590 PIDL-PW-3 13.18 201.0 9.00 16.64 15.47 22.15 1 256 PIDL-PW-4 33.98 585.6 26.22 9.45 8.07 11.56 840 PIDL-PW 10.81 166.8 7.47 8.13 7.01 10.04 1 651 -
[1] SEO T, BAYEN A M, KUSAKABE T, et al. Traffic state estimation on highway: A comprehensive survey [J]. Annual Reviews in Control, 2017, 43: 128-151. doi: 10.1016/j.arcontrol.2017.03.005 [2] LIGHTHILL M J, WHITHAM G B. On kinematic waves Ⅱ. A theory of traffic flow on long crowded roads [J]. Proceedings of the Royal Society of London Series A Mathematical and Physical Sciences, 1955, 229(1178): 317-345. [3] DAGANZO C F. The cell transmission model: A dynamic representation of highway traffic consistent with the hydrodynamic theory [J]. Transportation Research Part B: Methodological, 1994, 28(4): 269-287. doi: 10.1016/0191-2615(94)90002-7 [4] PAYNE H J. Freeway traffic control and surveillance model [J]. Transportation Engineering Journal of ASCE, 1973, 99(4): 767-783. doi: 10.1061/TPEJAN.0000375 [5] MESSMER A, PAPAGEORGIOU M. METANET: A macroscopic simulation program for motorway networks [J]. Traffic Engineering and Control, 1990, 31(9): 466-470. [6] ZHANG H M. A non-equilibrium traffic model devoid of gas-like behavior [J]. Transportation Research Part B: Methodological, 2002, 36(3): 275-290. doi: 10.1016/S0191-2615(00)00050-3 [7] NAHI N E, TRIVEDI A N. Recursive estimation of traffic variables: Section density and average speed [J]. Transportation Science, 1973, 7(3): 269-286. doi: 10.1287/trsc.7.3.269 [8] WANG Y, PAPAGEORGIOU M. Real-time freeway traffic state estimation based on extended Kalman filter: A general approach [J]. Transportation Research Part B: Methodological, 2005, 39(2): 141-167. doi: 10.1016/j.trb.2004.03.003 [9] WANG Y, ZHAO M, YU X, et al. Real-time joint traffic state and model parameter estimation on freeways with fixed sensors and connected vehicles: State-of-the-art overview, methods, and case studies [J]. Transportation Research Part C: Emerging Technologies, 2022, 134: 103444. doi: 10.1016/j.trc.2021.103444 [10] 王亦兵, 胡然, 余宏鑫, 等. 面向高速公路非检测点位的全域交通状态预测方法[J]. 交通运输工程学报, 2025, 25(1): 274-294. doi: 10.19818/j.cnki.1671-1637.2025.01.020 WANG Yi-bing, HU Ran, YU Hong-xin, et al. Global traffic state prediction method for non-sensing locations on freeways[J]. Journal of Traffic and Transportation Engineering, 2025, 25(1): 274-294. doi: 10.19818/j.cnki.1671-1637.2025.01.020 [11] WANG R, FAN S, WORK D B. Efficient multiple model particle filtering for joint traffic state estimation and incident detection [J]. Transportation Research Part C: Emerging Technologies, 2016, 71: 521-537. doi: 10.1016/j.trc.2016.08.003 [12] 刘伟, 钟灿, 曹文明. 基于数据驱动的路网连续交通流短时预测方法综述[J]. 交通运输工程学报, 2026, 26(2): 24-43. doi: 10.19818/j.cnki.1671-1637.2026.141 LIU Wei, ZHONG Can, CAO Wen-ming. Review of data-driven short-term prediction methods for continuous traffic flow in road networks[J]. Journal of Traffic and Transportation Engineering, 2026, 26(2): 24-43. doi: 10.19818/j.cnki.1671-1637.2026.141 [13] LI L, LI Y, LI Z. Efficient missing data imputing for traffic flow by considering temporal and spatial dependence [J]. Transportation Research Part C: Emerging Technologies, 2013, 34: 108-120. doi: 10.1016/j.trc.2013.05.008 [14] TREIBER M, HELBING D. Reconstructing the spatio-temporal traffic dynamics from stationary detector data [J]. Cooperative Transportation Dynamics, 2002, 1(3): 1-24. [15] BAE B, KIM H, LIM H, et al. Missing data imputation for traffic flow speed using spatio-temporal cokriging [J]. Transportation Research Part C: Emerging Technologies, 2018, 88: 124-139. doi: 10.1016/j.trc.2018.01.015 [16] AYDILEK I B, ARSLAN A. A hybrid method for imputation of missing values using optimized fuzzy c-means with support vector regression and a genetic algorithm [J]. Information Sciences, 2013, 233: 25-35. doi: 10.1016/j.ins.2013.01.021 [17] BENKRAOUDA O, THODI B T, YEO H, et al. Traffic data imputation using deep convolutional neural networks [J]. IEEE Access, 2020, 8: 104740-104752. doi: 10.1109/ACCESS.2020.2999662 [18] XU D, WEI C, PENG P, et al. GE-GAN: A novel deep learning framework for road traffic state estimation [J]. Transportation Research Part C: Emerging Technologies, 2020, 117: 102635. doi: 10.1016/j.trc.2020.102635 [19] 贾兴利, 曲远海, 朱浩然, 等. 交通预测中的时空图神经网络研究综述: 从模型解构到发展路径[J]. 交通运输工程学报, 2026, 26(1): 46-74. doi: 10.19818/j.cnki.1671-1637.2026.01.003 JIA Xing-li, QU Yuan-hai, ZHU Hao-ran, et al. Research review on STGNN in traffic prediction: From model deconstruction to development path[J]. Journal of Traffic and Transportation Engineering, 2026, 26(1): 46-74. doi: 10.19818/j.cnki.1671-1637.2026.01.003 [20] ZHANG Z, YUAN Y, YANG X. A hybrid machine learning approach for freeway traffic speed estimation [J]. Transportation Research Record, 2020, 2674(10): 68-78. doi: 10.1177/0361198120935875 [21] YUAN Y, ZHANG Z, YANG X T, et al. Macroscopic traffic flow modeling with physics regularized Gaussian process: A new insight into machine learning applications in transportation [J]. Transportation Research Part B: Methodological, 2021, 146: 88-110. doi: 10.1016/j.trb.2021.02.007 [22] 刘婧, 陆良, 郑芳芳, 等. 基于卷积自适应平滑网络的高速公路交通状态估计方法[J]. 中国公路学报, 2024, 37(11): 275-288.LIU Jing, LU Liang, ZHENG Fang-fang, et al. Freeway traffic state estimation based on convolutional adaptive smoothing network [J]. China Journal of Highway and Transport, 2024, 37(11): 275-288. [23] RAISSI M, PERDIKARIS P, KARNIADAKIS G E. Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations [J]. Journal of Computational Physics, 2019, 378: 686-707. doi: 10.1016/j.jcp.2018.10.045 [24] DI X, SHI R, MO Z, et al. Physics-informed deep learning for traffic state estimation: A survey and the outlook [J]. Algorithms, 2023, 16(6): 305. doi: 10.3390/a16060305 [25] KARNIADAKIS G E, KEVREKIDIS I G, LU L, et al. Physics-informed machine learning [J]. Nature Reviews Physics, 2021, 3(6): 422-440. doi: 10.1038/s42254-021-00314-5 [26] REMPE F, LODER A, BOGENBERGER K. Estimating motorway traffic states with data fusion and physics-informed deep learning [C]//IEEE. 2021 IEEE International Intelligent Transportation Systems Conference (ITSC). New York: IEEE, 2021: 2208-2214. [27] USAMA M, MA R, HART J, et al. Physics-informed neural networks (PINNs)-based traffic state estimation: An application to traffic network [J]. Algorithms, 2022, 15(12): 447. doi: 10.3390/a15120447 [28] HUANG A J, AGARWAL S. Physics-informed deep learning for traffic state estimation: Illustrations with LWR and CTM models [J]. IEEE Open Journal of Intelligent Transportation Systems, 2022, 3: 503-518. doi: 10.1109/OJITS.2022.3182925 [29] SHI Z Y, CHEN Y Z, LIU J C, et al. Physics-informed spatiotemporal learning framework for urban traffic state estimation [J]. Journal of Transportation Engineering, Part A: Systems, 2023, 149(7): 04023056. doi: 10.1061/JTEPBS.TEENG-7545 [30] 王天昊. 车辆轨迹数据与交通流模型协同驱动的高速公路交通状态鲁棒重构[D]. 扬州: 扬州大学, 2024: 61-81.WANG Tian-hao. Robust traffic state reconstruction for freeways driven by vehicle trajectory data and traffic flow model [D]. Yangzhou: Yangzhou University, 2024: 61-81. [31] ZHANG J, MAO S, YANG L, et al. Physics-informed deep learning for traffic state estimation based on the traffic flow model and computational graph method [J]. Information Fusion, 2024, 101: 101971. doi: 10.1016/j.inffus.2023.101971 [32] SHI R, MO Z, HUANG K, et al. A physics-informed deep learning paradigm for traffic state and fundamental diagram estimation [J]. IEEE Transactions on Intelligent Transportation Systems, 2021, 23(8): 11688-11698. [33] ZHAO C, YU H. Observer-informed deep learning for traffic state estimation with boundary sensing [J]. IEEE Transactions on Intelligent Transportation Systems, 2023, 25(2): 1602-1611. [34] YANG Z, LI D, NAI W, et al. Compressible non-Newtonian fluid based road traffic flow equation solved by physical-informed rational neural network [J]. IEEE Access, 2024, 12: 12992-13009. doi: 10.1109/ACCESS.2024.3356173 [35] 王挺, 王洪刚, 马昌喜, 等. 基于物理信息自适应深度学习的交通状态估计[J]. 兰州交通大学学报, 2024, 43(4): 37-44, 97.WANG Ting, WANG Hong-gang, MA Chang-xi, et al. Physics-informed adaptive deep learning-based traffic state estimation [J]. Journal of Lanzhou Jiaotong University, 2024, 43(4): 37-44, 97. [36] DAHMEN V, LODER A, TILG G, et al. Traffic state estimation with loss constraint[C]//IEEE. 2022 IEEE 25th International Conference on Intelligent Transportation Systems (ITSC). New York: IEEE, 2022: 1907-1912. . [37] MO Z, FU Y, DI X. Quantifying uncertainty in traffic state estimation using generative adversarial networks [C]//IEEE. 2022 IEEE 25th International Conference on Intelligent Transportation Systems (ITSC). New York: IEEE, 2022: 2769-2774. [38] BARREAU M, AGUIAR M, LIU J, et al. Physics-informed learning for identification and state reconstruction of traffic density [C]//IEEE. 2021 60th IEEE Conference on Decision and Control (CDC). New York: IEEE, 2021: 2653-2658. [39] LU J, LI C, WU X B, et al. Physics-informed neural networks for integrated traffic state and queue profile estimation: A differentiable programming approach on layered computational graphs [J]. Transportation Research Part C: Emerging Technologies, 2023, 153: 104224. doi: 10.1016/j.trc.2023.104224 [40] CHEN X, CHEN R, WAN Q, et al. An improved data-free surrogate model for solving partial differential equations using deep neural networks [J]. Scientific Reports, 2021, 11(1): 19507. doi: 10.1038/s41598-021-99037-x [41] FUKS O, TCHELEPI H A. Limitations of physics informed machine learning for nonlinear two-phase transport in porous media [J]. Journal of Machine Learning for Modeling and Computing, 2020, 1(1): 19-37. doi: 10.1615/JMachLearnModelComput.2020033905 [42] FAN J, WENG W, CHEN Q, et al. PDG2Seq: Periodic dynamic graph to sequence model for traffic flow prediction [J]. Neural Networks, 2025, 183: 106941. doi: 10.1016/j.neunet.2024.106941 [43] UZUNOVA M S, JOLLY D, NIKOLOV E K. Analytical methods and algorithm of decision LWR-model of traffic flow in a class of distributed parameters systems [J]. Information Technologies and Control Journal, 2005: 19-27. [44] ZOBEIRY N, HUMFELD K D. A physics-informed machine learning approach for solving heat transfer equation in advanced manufacturing and engineering applications [J]. Engineering Applications of Artificial Intelligence, 2021, 101: 104232. doi: 10.1016/j.engappai.2021.104232 [45] HU J, SHEN L, SUN G. Squeeze-and-excitation networks [C]//IEEE. Proceedings of the IEEE conference on computer vision and pattern recognition (CVPR). New York: IEEE, 2018: 7132-7141. [46] WENG J, MENG Q. Modeling speed-flow relationship and merging behavior in work zone merging areas [J]. Transportation Research Part C: Emerging Technologies, 2011, 19(6): 985-996. doi: 10.1016/j.trc.2011.05.001 [47] GREENSHIELD B D, BIBBINS J R, CHANNING W S, et al. A study of traffic capacity [J]. Highway Research Board Proceedings. 1935, 14(1): 448-477. [48] WANG Y, YU X, GUO J, et al. Macroscopic traffic flow modelling of large-scale freeway networks with field data verification: State-of-the-art review, benchmarking framework, and case studies using METANET [J]. Transportation Research Part C: Emerging Technologies, 2022, 145: 103904. doi: 10.1016/j.trc.2022.103904 -
下载: