Abstract: In order to more thoroughly explore the influencing factors of highway vehicle conflict risk, both micro and macro factors were taken into account to analyze the possible heterogeneity in them. A random parameter Logit model was established to analyze. Using high-resolution vehicle trajectory data and time to collision (TTC) as the risk identification criterion, the data within 1 second before the occurrence of risk were extracted and processed to obtain three types of characteristics such as the vehicles own motion state, micro interaction with surrounding vehicles and macro traffic flow state on the road section. The features were filtered by using the Pearson correlation coefficient and embedding method. Then, a random parameter Logit model and a random parameter Logit model considering mean heterogeneity were constructed and compared by using the filtered features as model independent factors and the occurrence of vehicle conflict risk as model dependent variables. The research results show that the random parameter model considering mean heterogeneity has the best fitting result, and all three types of variables have significant effect on the conflict risk. The fluctuation of the vehicles own motion state and the speed difference between the vehicle and the vehicle in front are positively correlated with risk, while the proportion of large vehicles in adjacent lanes is negatively correlated with risk. The two types of characteristics, namely the vehicles own motion state and the micro interaction with the vehicle in front, have a significant impact on conflict risk, while the macro traffic characteristics have a relatively small impact.

Key words: traffic engineering; traffic conflict technique; influencing factors; random parameter Logit model

摘要： 为更全面研究高速公路车辆冲突风险的影响因素，笔者考虑微观与宏观因素影响，分析其可能存在的异质性，建立了随机参数Logit模型对其进行分析。使用高分辨率车辆轨迹数据，以碰撞时间（time to collision, TTC）为风险识别标准，提取风险发生前1 s内数据，处理得到车辆自身运动状态、与周围车辆的微观交互以及路段宏观交通流状态三类特征，并基于皮尔逊相关系数和嵌入法进行特征筛选；以筛选后的特征作为模型自变量，以车辆是否存在冲突风险为因变量，分别构建随机参数Logit模型以及考虑均值异质性的随机参数模型并进行对比。研究结果表明：考虑均值异质性的随机参数Logit模型拟合效果最好，且三类变量均对冲突风险有显著影响，其中车辆自身运动状态的波动以及车辆与前方车辆的速度差与风险呈正相关，相邻车道的大车比例与风险呈负相关；车辆自身运动状态以及与前方车辆的微观交互两类特征对冲突风险影响较大，而宏观交通特征的影响相对较小。

关键词: 交通工程；交通冲突技术；影响因素；随机参数Logit模型