Abstract: Considering the potential correlation between driver behavior, physiological changes and hazardous road features on the highland roadway, this paper proposes a method to identify hazardous road sections based on driving status. First, the driver's behavior and physiological data were collected through the real vehicle test, and the driving behavior data was obtained through Density Based Spatial Clustering of Applications with Noise. The driving behavior was classified into different groups based on the behavior characteristics. Then, a hazard recognition model was developed using convolutional neural network, the Bidirectional Long and Short-Term Memory neural network, and Attention mechanism. The identification of hazardous road sections was realized by (Global Positioning System) GPS point correspondence. The causative analysis of hazardous road sections was performed in consideration of driver perception, manipulation, and physiology based on different driving behaviors. Physiological indicators and linear parameters were considered as optimization factors, and multiple regression analysis was performed with vehicle speed as the dependent variable, and the recommended interval of vehicle speed was determined based on the safety domain values of physiological indicators. The results show that drivers can be classified as "cautious", "steady", and "aggressive" based on the behavioral characteristics, and the dangerous sections for the three types of drivers are mostly combined sections with curved slopes. The distribution of the dangerous sections closely related to the driving behaviors. The increase in altitude can accelerate the emergence of dangerous driving states. Drivers became nervous when driving up mostly due to the increase in the combination of the curved slope and the turning degree. Aggressive drivers become highly tense when drive on straight and longitudinal slopes with a slope greater than 6% . When descending, cautious and aggressive drivers are likely to be in a dangerous state when the gradient of straight and vertical slopes is greater than 3%. Aggressive drivers also have driving risks when the turning degree is greater than 80° and the combined value of the curved slope is greater than 50. The research results provides references for human accident prevention on plateau highways, roadway geometric design, and traffic management measures.

Key words: traffic engineering, dangerous road section identification, deep learning, driving status, driving safety, driving style

摘要： 针对高原环境中驾驶人风格、生理变化与危险路段特征之间的潜在关联，提出一种基于驾驶状态的危险路段识别方法，辨识和分析不同风格驾驶人具有潜在风险的路段，并提出优化方案。首先，通过实车实验采集驾驶人行为及生理指标数据，使用DBSCAN(Density Based Spatial Clustering of Applications with Noise)得出驾驶风格类型，并依据行为特征对驾驶风格进行差异性 分析；其次，采用卷积神经网络、双向长短时记忆神经网络与注意力机制搭建危险状态识别模型，通过GPS(Global Positioning System)点位对应实现危险路段辨识，并基于驾驶风格差异，从驾驶人感知、操纵与生理角度对危险路段进行致因分析；最后，将生理与道路线形作为优化参考，以车速建议为着力点进行多元回归分析，并按照生理舒适域确定车速建议区间。结果表明：驾驶人根据行为特点分为谨慎、稳健和激进型，3类驾驶人在上行和下行途中的危险路段多为具有弯坡特征的组合型路段；海拔提升可加速危险驾驶状态的出现，各类驾驶人在上行时的紧张状态多源于弯坡组合值和转角值的增长，激进型驾驶人在坡度大于6%的直纵坡路段时亦会开始高度紧张；下行时，谨慎与激进型驾驶人在直纵坡坡度大于3%时易出现危险状态，激进型驾驶人在转角值大于80°且弯坡组合值大于50时亦存在驾驶风险。研究成果可满足高原公路人因事故预防的需求，为线形设计与交通管理措施制定提供理论依据。

关键词: 交通工程, 危险路段识别, 深度学习, 驾驶状态, 行车安全, 驾驶风格