翟犇

雾引起的低能见度条件下，高速公路行车环境受到极大影响，使得发生事故的几率增加，甚至导致多车相撞的恶性交通事故发生。以团雾为代表的部分高速公路低能见度事件具有突发性、局地性、流动性的特点，而现有低能见度预测及预警措施存在预测不准确、时间尺度大、预警不及时等不足，因此迫切需要对高速公路低能见度的精细化预测和主动性管理。本课题围绕高速公路低能见度预测与预警技术展开研究，解析高速公路低能见度事件特征和形成机理，建立高速公路低能见度短临预测模型及交通安全预警系统，并提出低能见度下高速公路可变限速控制策略。

首先，分析了高速公路低能见度事件的影响因素。以重庆G50S路段为研究对象，利用安装在路侧的气象监测站采集的数据提取了1年范围内的低能见度事件，并建立了用于相关性分析的样本数据集。结果表明，低能见度事件的形成过程中，能见度的变化曲线往往呈现“象鼻形”的特点。从时间分布来看，清晨（4:00-9:00）是低能见度事件的多发时段，59%的低能见度事件往往仅持续不到半小时。此外，通过气象要素间相关性分析可以得出，低能见度事件的形成需要一定的气象条件，能见度等级与当前时刻及过去时刻的风速、湿度、气温、降雨量具有一定的相关性。

之后，建立了高速公路低能见度短临预测模型和预警系统。在分析了能见度等级与气象要素的相关性基础上，基于两种集成学习方法，在五个预测时间上分别建立了预测模型，并与传统方法进行了比较。模型预测结果表明，基于随机森林算法建立的预测模型在四种模型中表现出更好的预测能力。模型对能见度等级1（能见度小于100m）的预测能力最强，并且在提前15min预测时的整体表现较好。因此，基于集成学习算法建立预测模型可以被用于高速公路低能见度的短临预测。在此基础上，构建了低能见度下高速公路安全预警系统。提出了组成整个预警系统的五个子系统及其之间的逻辑流程。梳理了可变限速控制、入口匝道控制、交通安全引导及交通诱导四种主动交通安全管控措施，以及支撑主动管控措施需要布设的交通安全预警设施。

最后，提出了低能见度下高速公路可变限速控制策略。设计了可变限速控制流程和关键参数。综合考虑影响交通安全风险的交通流状态、气象条件、道路几何特征等动静态因素，利用低能见度下交通安全风险评估模型用于评估可变限速控制对安全的提升。选择元胞传输模型用于模拟交通流状态，建立了基于元胞传输模型的可变限速控制策略，并提出了结合最小二乘法和线性逼近约束优化各自优势的元胞传输模型参数组合标定的方法。在此基础上，综合考虑安全和效率两方面的要求，建立可变限速控制参数优化问题并利用遗传算法求解最优参数组合。以实际高速公路为例，给出了具体的可变限速控制策略并对控制效果进行了评估。结果表明，本研究提出的可变限速控制策略能够综合考虑高速公路安全和效率两方面的要求，在不显著降低通行效率的情况下，可以在一定程度上降低交通安全风险。

关键词：高速公路，能见度预测，交通安全预警，可变限速控制

Under low visibility conditions caused by fog, the driving environment of freeways is greatly affected, which will increase the probability of accidents, and even lead to malignant traffic accidents and secondary accidents of multi-vehicle rear-end collision. Some low visibility events of freeways have the characteristics of suddenness, locality, and mobility. However, the existing low-visibility prediction and early warning measures have inaccurate predictions, large time granularity, and untimely early warnings. Therefore, there is a need for refined prediction and active  management of low visibility on freeways. This study focuses on low-visibility prediction and early warning technology on freeways, analyzes the characteristics and formation mechanism of low-visibility events on freeways, establishes a short-term prediction model for low-visibility, and builds a low-visibility traffic safety early warning system. And develops a variable speed limit control strategy of freewayunder low visibility.

First of all, the influencing factors of low visibility events are analyzed. Taking a section of the G50S freeway as an example, one-year data detected by the weather monitoring station installed on the roadside is used to extract low-visibility events, and a dataset for correlation analysis is established. The results show that during the formation of low-visibility events, the change curve of visibility often presents a "trunk-shaped" characteristic. Early morning (from 4:00 to 9:00) is the frequent periods of low-visibility events, and 59% of the low-visibility events usually last less than half an hour. In addition, according to the correlation analysis of the meteorological elements, it can be concluded that the formation of low-visibility events requires certain meteorological conditions, and the visibility level has a strong correlation with wind speed, humidity, temperature, and rainfall at the current and past moments.

After that, a short-term prediction model for low visibility on freeway segment is established. Based on the relationship between visibility and other meteorological elements, visibility level prediction models are established at five prediction times using two ensemble learning algorithms and two popular benchmarked classifiers are selected to compared with the developed models. The results show that the prediction model based on the random forest algorithm shows better prediction ability among the four models. The model has the strongest ability to predict visibility level 1 (visibility is less than 100 meters), and the overall prediction performance of the model is better when low visibility is predicted 15 minutes in advance. Therefore, the prediction model based on ensemble learning algorithm is used for short-term prediction of low visibility on freeways. On this basis, a low-visibility traffic safety early warning system on freeways is developed. Five subsystems are proposed and the logical flow between the various subsystems is designed. Four active traffic safety control measures are sorted out and the traffic safety early warning facilities to support control measures are summarized.

Finally, a variable speed limit control strategy on freeways under low visibility conditions is proposed. The variable speed limit control process and key parameters are developed. A traffic safety risk assessment model under low visibility is established to evaluate the safety improvement of variable speed limit control, which considers factors that affect traffic safety risks, such as traffic flow status, meteorological conditions, and road geometric characteristics. The cell transmission model is selected to simulate the traffic flow state, and a combined calibration method for the cell transmission model is proposed. On this basis, an optimization problem is established to search the optimal parameters of the variable speed limit control, and the genetic algorithm is used to solve it. Taking a section of freeway as an example, a specific variable speed limit control strategy is given and the control effect is evaluated. The results show that the proposed variable speed limit control strategy can reduce traffic safety risks without significantly reducing traffic efficiency.

Key Words: Freeways, Low visibility prediction, Traffic safety early-warning, Variable speed limit control