Abstract: In order to describe the alertness state and level of highway drivers on mountain roads, a quantitative method for the alertness level on mountainous roads was proposed by comprehensively considering human and environmental factors. The initial observation indicators were determined by analyzing the drivers eye movement and behavior data, the dimensionality of multi-source data was reduced based on the kernel principal component analysis method, and the reduced principal components were classified into high and low alertness levels by using K-means clustering method. Taking the probability of low alertness level samples as the target variable, the alertness level quantitative factors were obtained by screening human and environmental factors. Based on the score card method, the quantitative model of the alertness level on mountain roads was established, which was verified through practical examples. The research results show that the AUC value and KS value of the quantitative model for the alertness level on mountainous roads considering human and environmental factors are 0.907 and 71.86%, respectively. The proposed model has good performance and can provide reference for the dynamic evaluation of alertness level of drivers on mountainous roads and the construction of safety supporting facilities.

Key words: traffic engineering; alertness level; mountain roads; score card model; real vehicle test; kernel principal component analysis

摘要： 为描述山区公路驾驶者的警觉状态及水平，综合考虑人因-环境因素，提出了一种山区公路警觉水平量化方法。通过分析驾驶者的眼动及行为数据来确定初始观测指标，基于核主成分分析法对多源数据进行降维，对降维后的主成分使用K-means聚类方法划分高低警觉水平；以低警觉水平样本概率为目标变量，通过筛选人因-环境因素得到了警觉水平量化因子；基于评分卡方法建立了山区公路警觉水平量化模型，并对模型进行了实例验证。研究结果表明：基于人因-环境因素的山区公路警觉水平量化模型的AUC值为0.907，KS值为71.86%，模型效果较好，能为山区公路驾驶者警觉水平动态评价及安全配套设施建设提供参考依据。

关键词: 交通工程；警觉水平；山区公路；评分卡模型；实车试验；核主成分分析