[论文解读] Joint Estimation of Multi-phase Traffic Demands at Signalized Intersections Based on Connected Vehicle Trajectories

Chaopeng Tan, Jiarong Yao+3 more|arXiv (Cornell University)|Oct 19, 2022

Traffic Prediction and Management Techniques被引用 5

一句话总结

本文提出了一种基于联网车辆（CV）轨迹的周期性联合估计方法（JO-MAP），用于信号交叉口的多相位交通需求估计。通过构建联合加权似然函数，并将历史CV数据作为先验分布，该方法在自由流和过载条件下均提升了估计精度，在实证验证中实现了12.73%的MAPE，显著降低了对高CV渗透率的依赖。

ABSTRACT

Accurate traffic demand estimation is critical for the dynamic evaluation and optimization of signalized intersections. Existing studies based on connected vehicle (CV) data are designed for a single phase only and have not sufficiently studied the real-time traffic demand estimation for oversaturated traffic conditions. Therefore, this study proposes a cycle-by-cycle multi-phase traffic demand joint estimation method at signalized intersections based on CV data that considers both undersaturated and oversaturated traffic conditions. First, a joint weighted likelihood function of traffic demands for multiple phases is derived given real-time observed CV trajectories, which considers the initial queue and relaxes the first-in-first-out assumption by treating each queued CV as an independent observation. Then, the sample size of the historical CVs is used to derive a joint prior distribution of traffic demands. Ultimately, a joint estimation method based on the maximum a posteriori (i.e., the JO-MAP method) is developed for cycle-based multi-phase traffic demand estimation. The proposed method is evaluated using both simulation and empirical data. Simulation results indicate that the proposed method can produce reliable estimates under different penetration rates, arrival patterns, and traffic demands. The feature of joint estimation makes our method less demanding for the penetration rate of CVs and the consideration of prior distribution can significantly improve the estimation accuracy. Empirical results show that the proposed method achieves accurate cycle-based traffic demand estimation with a MAPE of 12.73%, outperforming the other four methods.

研究动机与目标

• 解决利用联网车辆（CV）数据在信号交叉口实现实时、多相位交通需求估计的空白。
• 开发一种在自由流和过载交通条件下均能可靠运行的方法。
• 通过联合估计与先验分布整合，降低对高CV渗透率的依赖。
• 通过利用历史CV数据作为先验分布，提升估计精度。
• 实现基于周期的实时交通需求估计，以支持动态信号优化。

提出的方法

• 基于实时CV轨迹观测，构建多相位交通需求的联合加权似然函数。
• 通过将每辆排队的CV视为独立观测，放松了先进先出假设。
• 利用历史CV样本量推导交通需求的联合先验分布。
• 开发一种基于最大后验概率（MAP）的估计框架（JO-MAP），用于周期性多相位交通需求估计。
• 整合实时CV数据与历史数据，以增强鲁棒性与精度。
• 将JO-MAP方法应用于模拟与实证数据集进行验证。

实验结果

研究问题

• RQ1 如何利用联网车辆轨迹在信号交叉口实现多相位交通需求的联合估计？
• RQ2 所提出的方法在多大程度上降低了对高联网车辆渗透率的依赖？
• RQ3 历史CV数据的整合在多大程度上提升了实时估计的精度？
• RQ4 该方法在过载交通条件下是否仍能保持高精度？
• RQ5 与现有单相位或非联合估计方法相比，JO-MAP方法表现如何？

主要发现

• 实证验证中，JO-MAP方法实现了12.73%的平均绝对百分比误差（MAPE），优于四种基准方法。
• 仿真结果表明，该方法在不同CV渗透率、到达模式和交通需求水平下均表现出可靠的估计性能。
• 联合估计框架显著降低了实现准确估计所需的CV渗透率。
• 基于历史CV数据的先验分布整合显著提升了估计精度。
• 该方法在自由流和过载交通条件下均保持了稳健的性能。
• JO-MAP方法在实时动态信号控制与交通管理应用中展现出强大潜力。