Abstract: This paper proposes a method to quantify the impact of large vehicles on urban roads based on a large amount of license plate recognition (LPR) data. The vehicle headways are analyzed in three situations: on the left-turn lane, on the through lane at the intersection, and on the road segments. The headway analysis provides the input for the analysis of large vehicle impact. The LPR data is divided into two parts according to the acquisition location. A differentiated data preprocessing process is proposed to obtain the headway datasets to investigate four types of vehiclefollowing combinations under different lane conditions. 13 sub-models from the Gaussian Mixed Model (GMM), the Lognormal Mixture Model, and the Gaussian/lognormal Mixture Model are applied to fit all the headway datasets. The Expectation Maximization algorithm is used to solve the parameters. The Kolmogorov- Smirnov test excludes the models that do not meet the requirements, and Akaike Information Criterion and Minimal Description Length are combined to select the optimal model. The impact of large vehicles on different types of travel lanes is quantitatively evaluated based on the optimal model parameters. The amounts of LPR data collected by many checkpoints and electronic police equipment are used to verify the effectiveness of the method. The results show that the headways from different lane types follow different distributions. It is appropriate to model the headways corresponding to lane types. The GMM with three density branches performs best in fitting all types of headway datasets, while other models appear to be unadapted in different stages. Under various conditions, large vehicles have varying impacts on the mean andstandard deviation of the headways of the relevant vehicle- following combinations. Large vehicles have significant impact on traffic flow on the road segments. Then is the situation on the left-turn lane and then the through lane at the intersection. The fitting results provide references for evaluating the impact of large vehicles.

Key words: traffic engineering, impact of large vehicle, mixture distribution model, headway, license plate recognition

摘要： 为量化大型车对城市道路交通运行的影响，提出基于大量车牌识别(License Plate Recognition, LPR)数据研究路段、交叉口左转、交叉口直行这3类车头时距，分析大型车影响的方 法。首先，将LPR数据按采集位置划分，提出差异化数据预处理流程，得到用于考察不同车道条 件下4类过车组合的车头时距集合；然后，以高斯混合模型(Gaussian Mixed Model, GMM)、对数正 态混合模型及高斯/对数正态混合模型这3类共13个子模型分别对上述所有集合建模，以最大期 望算法求解参数；之后，以Kolmogorov-Smirnov检验排除不满足要求的模型，综合赤池信息准则 与最小描述长度准则对剩余模型择优；最后，基于最优模型参数定量评价大型车对不同类型车道 的影响。以某城市区域多个卡口与电子警察设备采集的大量LPR数据验证方法有效性。结果表 明：路段与交叉口、交叉口各功能车道的车头时距不符合同一分布，宜区分建模；3个密度分支的 GMM拟合各类车头时距集合均有最佳表现，其他模型在不同阶段体现出不适应性；各种车道条 件下，大型车对相关过车组合的车头时距均值及标准差均有不同程度的影响，且该影响按照路 段、交叉口左转、交叉口直行的顺序依次递减。拟合结果可供大型车影响评价借鉴。

关键词: 交通工程, 大型车影响, 混合分布模型, 车头时距, 车牌识别数据