交通运输系统工程与信息

交通运输系统工程与信息 ›› 2024, Vol. 24 ›› Issue (1): 159-167.DOI: 10.16097/j.cnki.1009-6744.2024.01.016

• 系统工程理论与方法 • 上一篇    下一篇

基于交通波特征的车道级车流溯源方法

袁见1，刘福强2，安琨1，郑喆1，马万经*1，俞秋田3   

  1. 1. 同济大学，道路与交通工程教育部重点实验室，上海 201804；2. 麦吉尔大学，土木工程系，蒙特利尔 H3A0C3，加拿大； 3. 浙江数智交院科技股份有限公司，杭州 310000
  • 收稿日期:2023-10-26 修回日期:2023-11-15 接受日期:2023-11-20 出版日期:2024-02-25 发布日期:2024-02-12
  • 作者简介:袁见(1995- )，男，浙江嵊州人，博士生
  • 基金资助:
    国家自然科学基金(52325210, 52131204)

Lane-level Traffic Flow Tracing Method Based on Traffic Shockwave Features

YUAN Jian1, LIU Fuqiang2, AN Kun1, ZHENG Zhe1, MA Wanjing*1, YU Qiutian3   

  1. 1. The Key Laboratory of Road and Traffic Engineering of the Ministry of Education, Tongji University, Shanghai 201804, China; 2. The Department of Civil Engineering, McGill University, Montreal, QC H3A 0C3, Canada; 3. Zhejiang Institute of Communication Co. LTD, Hangzhou 310000, China
  • Received:2023-10-26 Revised:2023-11-15 Accepted:2023-11-20 Online:2024-02-25 Published:2024-02-12
  • Supported by:
    National Natural Science Foundation of China (52325210, 52131204)

PDF

摘要： 为支撑城市道路中低渗透率轨迹数据条件下车流溯源分析，本文提出一种基于交通波特征的车道级车流溯源方法。基于NGSIM(Next Generation Simulation)数据集中的真实车辆轨迹信息，解析不同来源车流的交通波特征差异性；结合信号配时方案，从车辆初次停车时刻、集结波起始时空位置、集结波斜率、集结波覆盖长度等多个维度验证了交通波应用于车流溯源的可行性。在此基础上，提取集结波5项特征参数，构建4种基于机器学习的车道级车流实时溯源模型。采用NGSIM数据对模型参数进行训练标定，并对不同归一化方法、不同数据量、不同数据精度下的模型效果进行灵敏性分析。结果表明，在低数据量场景下，特征参数宜采用Min-Max法进行归一化处理，溯源流量平均误差比例最大不超过23.60%；当数据量较为充分时(超过100个信号周期)，特征参数宜采用Z-Score法进行归一化处理，平均误差比例最大不超过9.90%，效果最佳的梯度提升回归模型的平均误差低至0.01%。此外，数据误差对不同模型的影响有所差异，但在误差较大时模型不会出现失效问题。本文所构建方法不依赖于固定检测器数据，未来可进一步研究交通波在网络层面的溯源方法。

关键词: 智能交通, 车流时空溯源, 机器学习, 车辆轨迹, 交通波

Abstract: To support traffic flow tracing analysis under low-penetration trajectory data in urban roadways, this paper proposes a lane-level traffic flow source tracing method based on traffic shockwave features. Using real vehicle trajectory data from the Next Generation Simulation (NGSIM) dataset, the differences in traffic shockwave features from different origins of traffic flow are analyzed. Combined with signal timing schemes, the feasibility of using traffic shockwaves for flow source analysis is validated across multiple dimensions, including initial vehicle stopping time, the spatiotemporal location of shockwave initiation, slope, and coverage length. Based on this analysis, five shockwave features are extracted to develop four machine learning-based real-time lane-level traffic flow tracing methods. These models are trained and calibrated using the NGSIM data, and the sensitivity to different normalization methods, data volume, and data accuracy is analyzed. The results show that in low data volume scenarios, the features should be normalized using the Min-Max method, with a maximum average percentage error not exceeding 23.60%. When data volume is more abundant (exceeding 100 signal cycles), the Z-Score normalization method is preferred, with the maximum average percentage error not exceeding 9.90%. The gradient-boosting regression model performs best with an average error as low as 0.01%. In addition, the effect of data errors varies from model to model, but the models do not have failure problems when the errors are large. This method is independent of fixed detector data. In the future, the study can be extended to the network-level flow tracing based on traffic shockwave features.

Key words: intelligent transportation, spatiotemporal tracing of vehicles, machine learning, vehicle trajectories, traffic shockwave

中图分类号: