Abstract:A new modeling approach based on frequent and equidistant speed measurements was proposed. Driving simulation experiments were carried out to collect free-flow speed data on mountainous freeways. Operating speed profile was derived and grade, curvature, super-elevation, and dynamic sight distance variables extracted from upstream and downstream segments were used to estimate a generous operating speed prediction model. Results showed that as the expected, the average grade and curvature of the segment, the maximum superelevation and the existence of reverse curve decreased the drivers’ speed and the curvature differential between the downstream and upstream segment also reduce the speed. The adjusted R2 of the operating speed model was 0.74 and the averaged absolute prediction error for training set and testing set were both under 3 km?h-1. This study took the advantage of high resolution speed data collected from driving simulator and provided a new methodology to model continuous speed data. The results indicated the effectiveness and capability of the proposed modeling strategy.
