交通违法行为与事故关系研究:基于零截尾负二项和广义泊松模型

Relationship Between Traffic Violations and Crashes Based on Zero-Truncated Negative Binomial and Generalized Poisson Models

为了探究交通违法行为与事故间的相关关系,选用符合交通违法行为和事故数据特点的零截尾负二项模型和广义泊松模型对某城市2007-2016年高频发生的10类交通违法行为与3类事故间的关系进行建模,估计模型各参数,并对比分析两模型的拟合效果和预测精度。研究结果表明,ZTNB模型在拟合效果和预测精度上均优于GP模型;交通事故不具有年份时间平稳性;与其他违法行为相比,跟车过近对追尾事故影响最大,违法变道对刮擦事故影响最大,而违反信号灯对直角事故影响最大。考虑其与交通违法行为间相关关系及相关程度的事故预防方法更具针对性和有效性。

In order to explore the correlation between traffic violations and traffic crashes, the Zero-Truncated Negative Binomial(ZTNB) and Generalized Poisson(Generalized Poisson, GP) models that best fit the traffic crashes data set are selected to describe the relationships between the 10 types of traffic violations and the 3 types of crashes that occurred frequently in a city from 2007 to 2016. The estimated parameters for models are estimated. Afterward, the goodness of fit and the prediction accuracy of the two proposed models are compared and analyzed. The results show that the ZTNB model is better than the GP model in terms of goodness of fit and prediction accuracy;traffic crashes don’t have time stability in terms of year;compared with other traffic violations, following too closely has the greatest impact on rear-end crashes, improper lane change has the greatest impact on side swipe crashes, and failed to obey traffic signal has the greatest impact on right-angle crashes. The crash prevention method that considers the correlative relationships between traffic violations and traffic crashes is more targeted and effective.