Risk assessment of underground road accidents based on driver behavior preferences

Underground roads, due to their limited space and enclosed environment, impact drivers' psychology and complicate emergency rescue in the event of accidents. Accurate evaluation of traffic accidents improves the safety of underground road traffic. Based on car-following and lane-changing behavior data from a typical underground road in Shanghai, this paper combines an unsupervised feature selection model based on mutual information with the K-means clustering algorithm to classify drivers into three categories: aggressive, stable, and cautious. Considering the behavior preferences of drivers with different styles, this study modifies existing microscopic driver behavior models, including models of car-following behavior and lane-changing decisions. From the perspective of the road system, consisting of people, vehicles, roads, and environments, the study considers the influencing factors of underground road traffic accidents, including driver characteristics, traffic flow states, and the geometric alignment of underground roads. Additionally, considering various collision types in both longitudinal and lateral directions, the study establishes an evaluation index system for underground road traffic accidents and determines threshold values using the graphical method of extremum theory. Finally, the study conducts a case analysis on a specific underground road in Shanghai to identify typical high-risk sections and key factors. The findings provide technical support for the safe operation of underground roads.