高速公路交通事故非线性负二项预测模型

Nonlinear Negative Binomial Regression Model of Expressway Traffic Accident Frequency Prediction

以京港澳高速公路(G4)粤境北段3年发生的1 354起交通事故为研究对象,将基础数据根据路段长度一致、曲线半径一致和坡度一致划分路段单元对基础数据进行处理,从道路线形和环境条件2个方面选取13个自变量,分别采用负二项(Negative Binomial, NB)回归模型和非线性负二项(Nonlinear Negative Binomial, NNB)回归模型建立交通事故起数预测模型,根据模型的拟合优度和预测准确性对比分析负二项回归和非线性负二项回归模型的优劣,并找出影响交通事故起数的显著自变量,分析显著自变量对交通事故起数的影响程度。研究结果表明:无论采用上述何种路段划分方法,非线性负二项回归模型构建的交通事故起数预测模型均优于负二项回归模型;采用坡度一致划分方法明显优于路段长度一致和曲线半径一致划分方法,更适合应用于山区高速公路交通事故数预测研究;从显著变量相关性来看,路段长度、相邻路段坡度变化值、弯坡组合、曲率、是否存在隧道路段以及是否为易结冰和起雾路段均是非线性模型的显著影响因素。

In this paper, the significant influencing factors of expressway traffic accidents were identified, and the relationship between traffic accident frequency and influencing factors such as road geometry and environmental conditions on expressways were analyzed. Accordingly, 1354 accidents that occurred on the Beijing-Hong Kong-Macao Expressway within the northern part of the Guangdong Province in a three-year period were evaluated. Three types of segmentation methods were used to divide the study section: fixed-length, homogeneous horizontal radius, and homogeneous longitudinal grade. Thirteen independent variables were selected from the road geometry and environmental conditions. Further, an accident frequency prediction-model was established using negative binomial(NB) and nonlinear negative binomial(NNB) regression models to explore the significant influencing factors. Elastic analysis was used to determine the degree of influence of the independent variables. The results indicate that the accident frequency prediction model based on the NNB regression model is better than that of the NB regression model for the fixed-length, homogeneous horizontal radius, and homogeneous longitudinal grade segmentation methods. Additionally, homogeneous longitudinal grade segmentation methods are better than the fixed-length and homogeneous horizontal radius methods for studying mountain expressways. From the correlation of significant variables, it was found that the length of section, change value of slope of adjacent sections, combination of curves and slopes, curvature, existence of tunnel sections, and whether the sections are susceptible to freezing and fog are the significant factors of the nonlinear model.