基于数据驱动的城市快速路段拥堵状态辨识及诱导对象选择方法

交通诱导实施效果不佳的主要原因之一是具有差异性出行特征的出行者无法接受单一的诱导方案。针对城市快速路高峰时段拥堵问题, 研究了考虑车辆出行特征差异的交通诱导对象精准识别方法, 以保障诱导方案的实施效果。利用高德路况数据提取拥堵路段, 根据拥堵路段与相邻路段交通状态的相关性提出拥堵源路段识别方法; 利用车牌识别数据提取使用快速路车辆的出行特征, 包括快速路出行强度、地面道路出行强度、快速路出发时刻离散度和快速路路径选择多样性; 采用K-means++算法对车辆出行特征进行聚类, 识别出显著影响道路交通状态的出行者, 并为出行者推荐适合其出行特征的错峰或绕行诱导方案。以苏州快速路为例, 研究发现: 针对拥堵源路段的交通诱导能有效改善拥堵路段的交通状态; 类型3车辆(高频出行且易绕行)占单月工作日早高峰所有使用快速路车辆总数的14%, 却占单日早高峰总交通量的51%, 是重点诱导对象; 通过精准识别, 可推荐诱导车辆数占总车辆数的47%。 

A Data-driven Method for Identifying Congestion State and Selecting Guided Vehicles for Urban Expressways

The inefficient traffic guidance is that travelers are reluctant to accept a single guidance scheme due to heterogeneous travel characteristics. This work proposes an accurate selection method based on the travel characteristics to ensure guidance performance, thus alleviating the peak-hour congestion of the expressways. The congested sections are extracted from a traffic condition dataset of the Gaode map, and the original congested sections are identified according to the correlation of traffic conditions between the congested sections and its adjacent ones. Besides, the travel characteristics of vehicles passing on the expressways are extracted based on the license plate recognition data, including the travel intensity on the expressways, the travel intensity on the ground roads, the dispersion of expressway departure time, and the diversity of the expressway path selection. The travelers significantly affecting the traffic condition of the expressways are identified by the K-means++ clustering algorithm, and appropriate guidance(i.e. staggered shift and detour)is recommended to the identified travelers based on their traveling characteristics. Taking the Suzhou expressway as a case study, the traffic guidance for the original congested sections can improve the traffic condition of congested sections. Type-3 vehicles(high-intensity travel and easy to detour)are the key targets, accounting for only 14% of the total number of vehicles using expressways in the morning peak of working day in one month. However, they constitute 51% of the total traffic volume. There are 47% of vehicles that can be recommended with personalized traffic guidance after congestion-state identification and guided-object selection.