基于贝叶斯网络模型的道路交通事故链生成与演化研究

Study on Generation and Evolution of Road Traffic Accident Chain Based on Bayesian Network Model

为了研究道路交通事故链的生成和演变规律并全面反映道路交通事故的产生机理,以美国100-Car自然驾驶研究数据为基础,充分考虑了驾驶员状态和驾驶行为对道路交通安全的影响,构建了事故发生前驾驶员状态和行为特征参数等变量,并与其他传统驾驶员特征、道路交通特性以及环境特点等影响因素变量共同建立了关于道路交通事故风险类型的贝叶斯网络模型。在贝叶斯网络模型基础上,引入事故因果链理论,利用有向无环网络中简单路径搜索算法生成事故链集合,并采用信息增益特征选择方法识别关键事故链。通过100-Car自然驾驶数据得到的关键事故链显示,单车事故类型更容易在弯道和坡道条件下驾驶员无注意力转移或进行简单非驾驶任务的情况下发生;而正面、侧面和尾部碰撞事故类型的发生,往往在水平直线道路条件下,伴随着驾驶员注意力由前方道路转移至两侧车窗或内后视镜,以及驾驶员非驾驶任务变复杂的情况。通过改变事故链中各节点的状态概率,可以探索不同类型事故的演化路径和规律,克服了传统方法对每个道路交通事故致因因素进行独立分析的局限性,揭示了事故链中影响因素/事件之间的相关关系,为更好地掌控道路交通风险状态和实现事故链阻断提供了新的思路。

In order to study the generation and evolution rules of road traffic accident chain and fully reflect the mechanism of road traffic accidents,based on the 100-car natural driving research data from the United States,the variables characterizing pre-accident driver status and behavior are constructed considering the impact of driver＇ s status and driving behavior on road traffic safety,which are utilized along with other variables of influencing factors such as traditional driver,roadway,traffic,and environment characteristics for establishing the Bayesian network model for road traffic accident risk type. By introducing the casual chain theory of accidents,a simple path search algorithm for directed acyclic network is used to generate the accident chain set based on the established Bayesian Network,and the information gain feature selection method is used to identify the key accident chain. The result of critical accident chain derived from the 100-car natural driving research data shows that（ 1） single vehicle accidents tend to occur on curved and sloped roadways when drivers are not distracted or not engaged in complex secondary tasks;（ 2） while front,side,and rear-end collision accidents are more usually occurred under horizontal and straight roadway conditions withthe driver＇s attention transferring from the roadway ahead to both sides of the window and rearview mirror or the secondary task of the driver becoming more complex. By changing the status probability of each node in the identified accident chain,the evolution path and rule of different types of accidents can be explored,which has overcome the limitation of traditional methods that analyze the causes of road traffic accidents separately,and reflected the interrelationship of the contribution factors and events of accident chain,and provided a new perspective on better control of road traffic risk and realizing the blockage of accident chain.