基于HighD数据集的高速公路交通冲突预测方法

Prediction Method of Highway Conflict Based on HighD Dataset

为改进高速公路交通冲突模型,基于High D数据集提取某路段交通冲突数据和交通流数据,依据TTC(time to collision,碰撞时间)的大小确定轻微冲突、一般冲突以及严重冲突的阈值。研究不同影响因素与冲突次数之间的关系,并对4种模型的效果进行比较,为冲突预测方法的研究提供思路。首先,针对不同严重程度的交通冲突数分别采用Poisson(泊松)模型、NB(Negative Binomial,负二项)模型、ZINB(Zero-inflated Negative Binomial,零膨胀负二项)模型及ZIP(Zero-inflated Poisson,零膨胀泊松)模型进行回归分析。其次,利用赤池信息量准则、贝叶斯信息准则对4种模型的拟合效果进行比较。结果表明:不同冲突次数预测模型中表现最佳的模型分别为NB模型、ZIP模型以及Poisson模型,3种模型中加速度标准差和平均速度都与冲突次数显著相关。

In order to improve the highway traffic conflict model,the traffic conflict data and traffic flow data of a certain road section were extracted based on HighD dataset,and the thresholds of minor conflict,general conflict and serious conflict were determined according to TTC numerical size.The relationship between different influencing factors and the number of conflicts were studied and the effects of four models were compared.It provides a feasible idea for the study of conflict pre-diction methods.Firstly,Poisson model,NB model,ZINB model and ZIP model are used to conduct regression analysis for the number of traffic conflicts of different severity.Secondly,the fitting effects of the four models are compared by using Akaike Information Criterion(AIC)and Bayesian Information Criterion(BIC).The results show that NB model,ZIP model and Poisson model have the best performance among the conflict prediction models.The velocity standard deviation and average velocity were significantly correlated with the number of conflicts in the three models.