基于贝叶斯网络的城市道路交通事故分析  

Bayesian network-based urban road traffic accidents analysis

作　　者：赵金宝[1] 邓卫[1] 王建[1] 

机构地区：[1]东南大学交通学院,南京210096

出　　处：《东南大学学报（自然科学版）》2011年第6期1300-1306,共7页Journal of Southeast University：Natural Science Edition

基　　金："十一五"国家科技支撑计划资助项目(2006BAJ18B03);江苏省普通高校研究生科研创新计划资助项目(CXZZ11_0165)

摘　　要：以5190起交通事故数据为分析依据,基于专家知识和数据融合方法建立了城市道路交通事故分析的贝叶斯网络结构.利用服从Dirichlet分布的贝叶斯方法对贝叶斯网络进行参数学习.结合网络模型,应用联合树引擎推断了在车辆类型、事故地点和交通参与者等因素的影响下交通事故类型概率分布.结果表明:客货车等大型车辆发生侧面碰撞的可能性为39.96%,高于其他车型;助力车和自行车在正面碰撞引发事故的可能性分别为39.01%和39.44%;因制动不当引发尾随碰撞事故的可能性为46.12%;转向不当而引发的侧面碰撞可能性为55.72%;随交叉口进口道和出口道数量的增加,发生侧面碰撞的概率会增加.贝叶斯网络模型具有较高的精确度,相关研究可以为城市道路管理部门深入了解交通事故诱发因素和提高城市道路交通系统安全水平提供依据.On the basis of 5 190 recorded urban road accidents,the topological structure of BN（Bayesian network） is formed with references to expert knowledge and data fusion method.Bayesian method is used to complete the process of parameter learning with Dirichlet prior distribution.Under the influences of some factors,such as the vehicle type,accident location,and traffic participant,the probability of different traffic accident type are inferred using junction tree engine based on BN structure and recorded accidents.Inference results indicate that the probability of side collision caused by heavy vehicles is 39.96%,higher than other vehicle types.The probabilities of frontal collision caused by electric bike and bicycle are 39.01% and 39.44% respectively.Brake failure may cause the occurrence of rear-end collision and the inferred probability is 46.12%.Steering failure may cause side collision with a inferred probability of 55.72%.The more the accesses of an intersection,the higher the side collision probability is.Moreover,BN method has a high accuracy.The results of this paper can provide basis for road management department to study the characteristics of urban road traffic accidents and improve safety level of urban road traffic.

关 键 词：贝叶斯网络 城市道路 交通事故 Dirichlet分布 

分 类 号：U491.3[交通运输工程—交通运输规划与管理]