基于事故修正系数的山区高速公路事故预测模型

Accident prediction model of mountainous freeway based on crash modification factors

为量化分析山区高速公路线形指标对交通事故发生及严重程度的影响,引入事故修正系数(CMF)概念,提出基于CMF的高速公路交通事故预测模型。验证事故数据服从零膨胀负二项(ZINB)分布,并标定10个不同线形组合路段的基本事故预测模型;在界定理想条件的基础上,借助优势比分析方法,建立10个不同线形组合相对于基本预测模型的CMF模型;通过弹性分析识别出对不同严重程度事故有显著影响的因素,并以此分析山区高速公路线形指标及线形组合的安全效应。研究结果表明:CMF能够定量反映不同线形组合的风险效应,且其值越高,事故率及事故严重程度可能更高;山区高速公路中,平曲线缓坡组合相对安全,平曲线陡坡组合、平曲线竖曲线组合均有较高的事故风险;平曲线与凹型竖曲线组合时的风险会略低于其与凸形竖曲线组合;控制纵坡度、减少平竖曲线的组合均有利于山区高速公路交通安全。

In order to quantitatively analyze the influence of mountain highway alignment index on the occurrence and severity of traffic accidents,the concept of CMF was introduced,and a highway traffic accident prediction model based on CMF was proposed.The accident data were verified to obey the zero-inflated negative binomial distribution,and ten basic accident prediction models for different alignment combinations of road sections were determined.With the help of the dominance ratio analysis method,ten crash modification factor models of different alignment combinations with respect to the basic prediction model were established.The factors having significant influence on accidents of different severity were identified by elasticity analysis,and the safety effects of highway alignment index and combination of alignment were analyzed.The results show that:CMF can quantitatively reflect the risk effect of different alignment combinations.The higher the value,the higher the accident rate and accident severity may be.In the mountain highway,the combination of horizontal curve and gentle slope is relatively safe,while the combination of horizontal curve and steep slope and the combination of horizontal curve and vertical curve have higher accident risk.The risk of combination of horizontal curve and concave vertical curve is slightly lower than that of combination of horizontal curve and convex vertical curve.Controlling vertical slope and reducing the combination of horizontal curve and vertical curve are beneficial to highway safety in mountainous areas.