高速公路大区段交通事故预测模型研究

Study on Traffic Accident Prediction Model for Long Sections of Expressway

针对交通事故记录位置精度较低和路段划分过短可能引起的分析结论偏差问题,探讨了大区段划分条件下高速公路交通事故预测模型的构建方法。结合辽宁省5条高速公路的交通事故数据和道路交通条件数据,提出了基于高速公路天然节点(互通式立交和服务区)的大区段划分方法。构建了处理大区段内指标异质性的模型变量,应用积分-微分方法确定了变量模块,进而利用负二项回归方法建立了高速公路交通事故预测模型。研究表明:在大区段条件下,年平均日交通量、累积纵坡和标志密度等变量的事故预测模块为Hoerl函数,累积曲率为幂函数,而挖方段比例为指数函数;路段长度、年平均日交通量、累积纵坡和标志密度等因素对大区段上的交通事故发生具有显著影响,且考虑这些变量的模块式模型具有较高的预测精度。

To reduce the possible analysis bias caused by low accuracy of recorded traffic accident location and too short divided sections, the approach to establish expressway traffic accident prediction model based on division of long sections is discussed. Based on the traffic accident data and traffic condition data collected from 5 expressways in Liaoning, the method for dividing long sections based on natural notes （ interchanges and service areas） of expressway is put forward. The model variables which can handle the heterogeneity of indexes within long section are established, the building blocks corresponding to the variables are determined by using the integrate-differentiate method, and the expressway traffic accident prediction model is established by using the negative binomial regression method. The result shows that （ 1 ） under the condition of long section, the building blocks of the traffic accident prediction model for annual average daily traffic （AADT）, cumulative longitudinal grade, and density of traffic signs, etc. are Hoerl＇s functions, for cumulative curvature is power function, and for proportion of cut sections is exponential function ; （2） the factors such as length of section, AADT, cumulative longitudinal grade, and density of traffic signs have significant impact on the occurrence of traffic accidents in long sections, and the modular model including these variables has relative high prediction accuracy.