基于广义加性模型的高速公路交通事故影响因素分析

Analysis of the Crash Influence Factors of Motorway Based on Generalized Additive Model

为准确识别事故影响因素并充分反映各因素对事故的影响,基于8条高速公路的事故、道路设计及交通运行等数据,构建了广义加性模型(GAM),并将其与传统的广义线性模型(GLM)进行对比。结果表明:1)GAM比GLM具有更高的拟合优度与预测精度,且能识别出更多对事故具有显著影响的货车比例与纵坡坡度等因素,其中货车比例与事故次数呈显著相关性,且影响程度仅次于交通量;2)GAM模型可鉴别出事故与交通量、货车比例、平曲线曲率、纵坡坡度及连续下坡长度间的非线性关系,且各因素在不同数值范围内对事故的影响趋势和程度存在明显差异。结论是GAM具有挖掘更多事故影响因素的潜力,且模型输出结果可以更合理地解释各因素对事故的作用机理。

In order to exactly identify the crash influence factors and fully reflect the influence of each factor on crash,the Generalized Additive Model(GAM)was set up based on the crash,road design and traffic operation data of eight motorways,and was compared with the traditional Generalized Linear Model(GLM).The results showed that GAM had higher goodness of fit and predicting accuracy than GLM.GAM could identify more factors with significant influence on crash:the percentage of heavy vehicles and longitudinal grade,among of them,the percentage of heavy vehicles was significantly correlated with the crash frequency,and the influence degree was second only to traffic volume.GAM model has identified the nonlinear relationships between crash and traffic volume,percentage of heavy vehicles,curvature,longitudinal grade as well as distance along composite descending grade,and the influence trend and degree of each factor on crash were obviously different in different value range.In conclusion,GAM has the potential to explore more influence factors of crash,and the output of the model can explain the mechanism of each factor on crash more reasonably.