This study addresses the high accident rate on mountainous highways, driven by complex road alignments, harsh climatic conditions, and heterogeneous driving behaviors, aiming to enhance the accuracy of car-following behavior modelling. Using a typical mountainous curved section in Yunnan Province as the test case, drone-collected vehicle trajectory data were filtered using Kalman filtering to reduce noise. Subsequently, a K-Means algorithm optimized by differential evolution classified driving behaviors into three categories: aggressive, conservative, and standard. This revealed significant differences in speed, acceleration, and headway between distinct driving styles. To characterize curve dynamics, this study introduced a curve-radius parameter to enhance the Intelligent Driving Model (IDM). It calibrated it according to the rules for the three driving styles using genetic algorithms. Validation through macro-level error analysis and micro-level trajectory comparisons demonstrated that the improved model significantly enhances prediction accuracy for curve-following behavior while effectively adapting to diverse driving characteristics. This study pioneers the integration of driving behavior heterogeneity with curve geometry characteristics, providing a theoretical foundation for traffic flow simulation, safety assessment, and intelligent driving system design on mountain roads. It holds significant engineering value for reducing the risk of following-distance accidents and optimizing traffic management in mountainous regions.
Fu et al. (Wed,) studied this question.