山区高速公路交通事故的

贝叶斯网络模型诊断及推理研究

卢瑶

摘 要

山区高速公路受其地形、地质和气候等特殊因素影响，一旦交通事故发生，山区损失相对平原地区往往较为惨重。本文以京昆高速公路西汉段户县辖区、宁陕辖区、秦岭辖区三个路段在2016-2018年发生的2365起交通事故为背景，以文献统计、解释结构模型法和贝叶斯网络研究为核心，以贝叶斯网络建模分析山区高速公路交通事故的原因和预测分析为研究思路，本文主要研究工作如下：

首先，基于文献统计对一定年限内有关山区高速公路交通事故的影响因素分析的中英文文献进行检索分析，依据影响因素在文献中出现的频次及其所在文章的价值和利用率识别事故的关键影响因素，按照人、车、环境、道路一级因素分类，在检索的13个二级因素中最终选取其中9个为关键影响因素。

然后，基于解释结构模型法，针对山区高速公路交通事故的9个关键影响因素，分析因素间的存在的因果关联性建立了层次结构模型，模型分为四层，因素间逐层影响。是否桥隧路段、道路线形、长大纵坡、日均车流量四个变量因素在模型的底层，是交通事故的最根本致因因素。

之后，阐述贝叶斯网络原理、结构学习和参数学习方法，基于解释结构模型和K2算法建立山区高速公路交通事故混合贝叶斯网络结构模型，利用Netica软件进行参数学习得出节点条件概率，用K折交叉验证法验证贝叶斯网络的稳定性和精确度。

最后，对山区高速公路交通事故贝叶斯网络进行事故预测分析和事故的原因诊断，结合交通事故实例进行最大可能性解释，分析网络节点的敏感性。对课题研究的不足和创新点展望，为未来山区高速公路交通事故、贝叶斯网络研究提供借鉴。

关键词：交通事故、山区高速公路、影响因素、概率预测、贝叶斯网络

Abstract

Mountainous highways are affected by special factors such as their topography, geology and climate. Once a traffic accident occurs, the loss of mountainous areas is often more severe than that of plain areas. This paper takes 2365 traffic accidents in 2016-2018 in the three sections of Huxian District, Ningshan District and Qinling District of the West Han Dynasty of Jingkun Expressway as the background. This paper takes the bibliometric method, the interpretative structural model method and the Bayesian network research as the core of the research. This paper takes Bayesian network modeling to analyze the causes and prediction analysis of mountain highway traffic accidents as research ideas. The main research work of this paper is as follows:

Firstly, based on the bibliometric method, the Chinese and English literatures on the analysis of the influencing factors of mountainous highway traffic accidents within a certain period of time are searched and analyzed. Then, according to the frequency of the influencing factors appearing in the literature and the value and utilization rate of the article, the key influencing factors of the accident are identified. Finally, according to the first-class factors of people, vehicles, environment and roads, 9 of the 13 secondary factors retrieved are finally selected as key influencing factors.

Then, based on the interpretation structure model method, a hierarchical structure model is established for the nine key influencing factors of mountain highway traffic accidents and the causal correlation between the analysis factors. The model is divided into four layers, and the factors are affected layer by layer. Whether the four factors of bridge and tunnel section, road line shape, long longitudinal slope and average daily traffic flow are at the bottom of the model are the most fundamental causes of traffic accidents.

After that, the Bayesian network principle, structure learning and parameter learning methods are explained. Based on the interpretative structural model and K2 algorithm, a hybrid Bayesian network structure model for mountainous highway traffic accidents is established. Using Netica software for parameter learning, the node conditional probability is obtained. The stability and accuracy of the Bayesian network are verified by the K-fold cross-validation method.

Finally, the accident prediction analysis and the cause diagnosis of the accident are carried out on the Bayesian network of mountain highway traffic accidents. Combine the traffic accident instance to explain the maximum possibility and analyze the sensitivity of the network node. The shortcomings of the research on the subject and the prospect of innovation provide reference for the future highway traffic accidents and Bayesian network research in mountainous areas.

Keywords: Traffic accidents, mountain highways, influencing factors, probability prediction, Bayesian network