卢建涛

  在恶劣天气情况下，高速公路行车环境受到极大影响，机动车的操作稳定性会降低，使得发生事故的机率增大，甚至导致多车连续追尾相撞的恶性交通事故及二次事故。经常有因恶劣天气干扰而直接或间接引发的重大交通事故，恶劣天气下高速公路交通安全存在极大隐患。本文以恶劣天气下高速公路交通风险为研究对象，深入分析不同恶劣天气下对高速公路交通风险的影响要素以及诸要素之间相互联系、相互作用的机理，在此基础上建立恶劣天气下高速公路交通风险评估模型，提出恶劣天气下高速公路交通风险预警方法。

       首先，本文梳理了不同恶劣天气对交通环境的影响机理，从能见度和路面附着系数两方面，分析恶劣天气对于高速公路行车条件的影响。以G15沈海高速断面线圈采集数据和相应的天气数据为例，分析不同恶劣天气条件下交通流运行特性的变化情况。研究表明，在恶劣天气条件下，高速公路交通流量和平均运行速度变化趋势相同，随着降雨强度增加而减小，随着能见度减小而减小。随着天气恶劣程度的增加，高速公路交通运行也会越发的不稳定。当天气恶劣到一定程度（降雨强度大于30mm/h；能见度小于100m）时，对不同的驾驶人员同时产生严重的影响，高速公路交通运行状态会趋于稳定状态。根据恶劣天气条件下对交通运行特征的影响分析，本文选用断面交通流特征指标表征高速公路交通风险，为恶劣天气下高速公路交通风险评估提供数据支撑和理论依据。

       之后，本文利用随机森林方法对相关变量进行筛选,并运用贝叶斯Logistic回归模型构建恶劣天气下的高速公路交通风险评估模型。基于国内外高速公路的各种数据分布对交通流数据、天气数据和交通事故数据进行数据预处理和数据融合，通过数据匹配构建交通风险评估样本数据。针对不同时间段的样本数据分别构建随机森林模型，运用袋外数据错误率对恶劣天气下的交通流特征指标进行变量筛选，探索恶劣天气下影响交通风险的重要交通流变量。根据筛选得到的重要变量，运用贝叶斯Logistic回归模型构建恶劣天气下高速公路交通风险评估模型，并对模型预测精度进行检验。研究表明，利用事故发生前10-15分钟的数据样本建立的交通风险评估模型的非事故预测和整体预测准确度预测能力更好。对比包含天气因素与不含天气因素的交通风险评估模型，发现在交通风险评估模型中考虑天气因素，模型的各项指标均体现更好的预测效果。

        最后，本文介绍恶劣天气下高速公路交通风险预警方法和流程。根据高速公路交通风险按照风险等级划分的标准，将恶劣天气下高速公路交通风险划分为安全状态、可允许风险、中度风险、重大风险、不可接受风险5个等级。利用模糊C-均值聚类算法，对恶劣天气下高速公路交通风险值进行聚类分析。依据交通风险值的各聚类类别的最大值、最小值和风险等级划分的标准，确定恶劣天气下高速公路不同交通风险预警等级所对应的交通风险值范围。依据恶劣天气下高速公路交通风险评估模型计算不同恶劣天气条件下的交通风险分布，依照交通风险预警等级划定安全范围，从而确定不同恶劣天气条件下的安全车速。提出恶劣天气下高速公路主动安全管控建议，给出交通安全引导措施的布设建议，改善恶劣天气下高速公路行车环境。

关键词：恶劣天气，交通风险评估，风险预警，安全车速

Abstract

       Under adverse weather conditions, the driving environment of freeway is greatly affected, and the operational stability of motor vehicles will be reduced, which will increase the probability of accidents, and even lead to malignant traffic accidents and secondary accidents of multi-vehicle rear-end collision. Taking the traffic risk of freeway in adverse weather as the research object, this study deeply analyzes the influence of different weather conditions on the freeway traffic risk and the mechanism of the interrelation and interaction between various factors. On this basis, a weather-related traffic risk assessment model of freeway is established, and a weather-related traffic risk early-warning method is proposed.

       First of all, this study reviews the influencing mechanism of different adverse weather conditions on the traffic environment, and analyzes the influence of the adverse weather on the freeway driving conditions from two aspects: the visibility and the road adhesion coefficient. Taking G15 Shenhai Freeway loop data and related weather data as an example, the change of traffic flow characteristics under different weather conditions id analyzed. The research shows that the changing trend of traffic flow and average speed is the same in adverse weather, which means that the traffic flow and average speed increase with the increase of rainfall intensity, and decrease with the decrease of visibility. As the weather worsens, the traffic operation of freeway will become more unstable. When the weather id severe to a certain extent (the rainfall intensity is more than 30mm/h, and the visibility is less than 100m), it will cause a serious influence on different drivers, and the traffic operation state will tend to be stable. According to the analysis of the traffic under adverse weather conditions, this study selects the traffic flow characteristic indices to characterize the freeway traffic risk, and provides data support and theoretical basis for freeway traffic risk assessment under adverse weather conditions.

       After that, this study conducts data preprocessing and data fusion on traffic flow data, weather data and traffic accident data, and constructs traffic risk assessment sample data through data matching. A random forest model is constructed for the sample data of different time periods. It utilizes the error rate of out-of-bag data to filter the traffic flow characteristics indices, and to explore the important traffic flow variables influencing traffic risk under adverse weather conditions. According to the important variables, the Bayesian logistic regression model is used to construct the freeway traffic risk assessment model under adverse weather conditions, and the model prediction accuracy is tested. The result shows that the traffic risk assessment model established based on the data samples 10-15 minutes before the accident can predict better. Comparing the traffic risk assessment model with and without weather factors, it is found that considering the weather factors in the assessment model has better prediction results for each index of the model.

       Finally, this study introduces the early-warning process of freeway traffic risk assessment and the function of each part. According to the criteria of freeway traffic risk classification based on the risk grade, the freeway traffic risk in adverse weather is divided into five grades: safety, allowable risk, medium risk, major risk and unacceptable risk. The fuzzy C-means clustering algorithm is used to cluster the freeway traffic risk values under adverse weather. According to the maximum, minimum and risk classification criteria of each cluster category of traffic risk value, the range of traffic risk values corresponding to different traffic risk warning grades in adverse weather is determined. Then, the distribution of traffic risk under different adverse weather conditions is calculated using the traffic risk assessment model. And the safety range is defined in accordance to the traffic early-warning grades, thus the safety speed under different adverse weather conditions is determined. Furthermore, some early-warning measures for freeway under adverse weather conditions is are given to help reduce the influence of adverse weather on freeway traffic risk.

Key Words: Adverse weather, Traffic risk assessment, Risk early-warning, Safety speed