An Empirical Investigation of Speed Patterns on S-Curves Using Naturalistic Driving Data and Mixed Logit Model

Abstract

S-curves are common and complex curves in road networks, consisting of two curves in opposite directions connected by a short segment. Drivers’ behavior on S-curves reflects their ability to navigate complex environments, which is not yet fully understood. This study examines speed patterns on five S-curves using a naturalistic driving dataset of 2676 vehicle trajectories. A hierarchical clustering method is proposed to identify typical speed patterns on S-curves, and a mixed logit model is developed to analyze the influence of curve radius, slope, and lighting conditions on drivers’ speed pattern choices. The empirical findings reveal that (1) each S-curve has dominant speed patterns, and these patterns vary across different S-curves; (2) speed patterns generally exhibit three types of shapes: S-shaped, U-shaped, and anti-S-shaped; and (3) the curve radius, the difference in radii between the two adjacent curves, the slope, and lighting conditions significantly influence drivers’ speed pattern choices. These results provide valuable insights for S-curve alignment design, speed guidance strategies, and speed planning algorithms for autonomous vehicles.