中文摘要：

针对道路交通事故的形成机理进行定性、定量研究，根据我国道路交通事故记录数据特征，应用贝叶斯网对事故发生概率进行定量分析。引入“驾驶员紧张度”和“道路线形合理度”两个隐节点，建立了事故分析的贝叶斯网多层隐类模型，采用最大似然估计方法确定了模型的边缘概率和条件概率。将贝叶斯网模型应用于国道104二级公路（K1310＋000～K1330＋000）的事故分析中，运用贝叶斯网分析软件包Netica对其历史事故记录数据进行分析。结果表明：贝叶斯网不仅可以定量计算某种道路交通状态下的事故发生概率，而且可以找出影响事故概率的关键原因和最不利状态组合（事故概率最大时的道路交通状态）。

英文摘要：

The paper is to propose a method for qualitative and quantitative analysis on road accidents. As is known, many factors are likely to conduce to traffic accidents, such as unreasonable road alignment, vehicle running away, drivers＇ carelessness and adverse weather influences and so on. Also, the interaction and interdepen- dence among these factors constitute the complexity of accident forma- tion. To describe this interaction and interdependence, Bayesian Net- work （BN） was applied to analyze the mechanism of road accident formation. While introducing two hidden nodes： Y0, which denotes Driver＇s Nervousness and Y1, Alignment Rationality, we have built a Bayesian Network Hierarchical Latent Class model with eight nodes built-in for the road accident analysis. The other six observable nodes include X0 （ level of driver＇ s familiarity with road condition）, Xt （time）, X2 （ traffic volume）, X3 （weather condition）, X4 （ horizontal radius） and X5 （longitudinal gradient）. Then the Marginal Probability and Conditional Probability were determined by Maximum Likelihood Estimation. During the estimation, prior probability was firstly let to be determined by the prior knowledge, such as historical accident data and experts experience. In addition, posterior probability was supposed to be calculated by the prior probability and likelihood function, with Marginal Probability and Conditional Probability being modified. And, finally, trial applications were conducted on the his- torical accidents data of the national highway G104 in China from K1310＋ 000 to K1330 ＋ 000, by using the common soft package of Netiea. Primary applications study were conducted to estimate the probability of road accidents under some certain conditions with spe- cial road geometry, traffic volume, drivers＇ performance and weather condition. Next applications were conducted to test fault diagnosis of the accident reasons. Afterwards, parameter sensitivity analysis and comparative analysis of different factors