基于微观仿真与极值理论的城市交叉口安全评价方法

Safety Evaluation Method Based on Traffic Simulation and Extreme Value Theory

为了提升中国的道路安全评价水平,提出一种基于微观交通仿真与极值理论相结合的城市交叉口安全评价方法。首先提出一种新的微观交通仿真标定方法,它是以优化冲突参数后侵占时间(Post-encroachment Time,PET)分布作为标定目标,同时将平均排队长度和冲突数量作为约束条件,并使用遗传算法寻找仿真模型中最优驾驶人行为参数的系统性方法。然后引入极值理论中的极大阈值方法(Peaks-over-threshold,POT),选用PET值作为模型输入指标,对交叉口年平均事故频率进行预测。选取上海市奉贤区的10个交叉口作为案例分析,对交叉口仿真建模后,分别使用基于平均排队长度的传统标定方法、考虑冲突数量的两阶段标定方法和优化PET分布的多约束标定方法进行标定,并使用POT方法获得各交叉口的事故预测值。结果表明:所提方法在相关性检验、相对安全检验和事故预测3个方面显著优于其他2种标定方法;将其与基于实际观测冲突的极值预测进行对比,发现两者在相交冲突和追尾冲突的事故预测上表现均较为良好,而对于换道冲突的预测较为一般;提出的基于微观交通仿真与极值理论的方法是一种可行的、有潜力的城市交叉口安全评价方法。

To improve the performance of safety evaluation methods in China, this paper puts forward a new safety evaluation method for urban intersections by integrating microscopic traffic simulation and the extreme value theory （EVT）. First, a new systematic simulation calibration method was proposed. The post encroachment time （PET） distribution was used as the calibration objective, constrained with the average queue length as a measure of effectiveness （MOE） and total conflict counts. The genetic algorithm （GA） was applied to find the optimal values of driver behavior models in simulation packages. Then, a peak-over-threshold （POT） based EVT was employed to predict the annual crash frequency of intersections, using the PET as input. Ten intersections in Fengxian, Shanghai were chosen for a case study. Three calibration methods were used to develop the simulation models, including a traditional method considering the queue length, a two-stage method considering both the MOE and conflict counts, and the proposed multi-constrained method for optimizing the PET distribution. Crash predictions were derived based on POT-based EVT and the three methods. The results show that the combination of POT and the proposed calibration method outperform the other two methods in terms of correlation, Spearman rank correlation, and crash prediction. Moreover, both field-based POT and simulation-based POT achieve a good crash prediction performance for crossing and rear-end conflict types, and an acceptable performance for a lane change conflict type. Thus, the proposed method can be considered a feasible and promising safety evaluation method for urban intersections.