基于贝叶斯网络的重特大交通事故关键因素分析

Research on influencing factors of major and extra serious traffic accidents based on Bayesian network

基于2013—2021年97起重特大交通事故数据,探究重特大交通事故关键致因。采用改进的灰色关联法构建关键影响因素集,以关键因素为节点变量建立了贝叶斯网络,通过网络结构学习和节点条件概率学习,分析了14个与事故相关的因素,并从中提取了15个单、多车事故死伤人数因素组合链,基于区间数理论对影响因素的危险性进行了排序。结果表明:单、多车事故致因存在差异,超载、存在大型客车、无物理隔离等因素对单、多车事故均有影响,但这些因素对单、多车事故死伤人数的影响程度不同,整体来看,多车事故后果更严重。贝叶斯网络可以反映各因素之间的真实关系,且具有较好的预测精度,研究结论有助于管理部门制定相应的预防策略以减少重特大交通事故发生频率。

Based on the data of 97 major and extra-serious traffic accidents from 2013 to 2021 in China,the relationship between accident impact factors and severity indicators was extracted to explore the key causes of major and extra-serious traffic accidents.Firstly,29 factors containing information about drivers,vehicles,roads,and the environment were extracted from the accident reports.Each category of factors was sorted in descending order of occurrence frequency,and the influencing factors were preliminarily screened based on a cumulative frequency greater than 90%.The Pearson correlation coefficient was used to analyze the correlation between the above factors and the index of accident severity(the number of deaths and injuries in single and multiple-vehicle accidents).Secondly,an improved grey correlation method was proposed to calculate the weighted grey correlation degree of the remaining factors,and the key influencing factors set was constructed by taking the average weighted grey correlation degree greater than 0.75 as the standard.Then,a Bayesian network was established by taking the key factors as node variables.Through the network structure learning based on search scoring and the node conditional probability learning based on Bayesian estimation,14 factors related to accidents were obtained,and 15-factor combination chains were extracted from them.Finally,the risk of single factors was ranked based on interval number theory.The results show that there are differences in the influencing factors of single and multiple-vehicle accidents,with factors such as overloading,large passenger cars,wet roads,no physical isolation,working days,and adverse weather affecting both.However,the degree of impact is different.In addition,fatigued driving and heavy trucks only have an impact on multi-vehicle accidents.Overall,the consequences of multi-vehicle accidents are more severe.The Bayesian network can reflect the true relationship between various factors and has good prediction accuracy.The findings of the study can help the management departments to develop appropriate prevention strategies to reduce the frequency of major and extra serious traffic accidents.