Transferability analysis of the freeway continuous speed model

Operating speed is often used to evaluate consistency in road geometric design. In the China, the Specifications for Highway Safety Audit includes a spot-based speed model that predicts operating speed by dividing the road into homogeneous segments and observing the speeds at sparsely spaced spots. This paper presents a continuous speed model as a more representative alternative for roads with complex alignments, and can be applied to tunnel sections as one general model. The model considers the road geometric characteristics not only at the vehicle’s current position, but also in its neighborhood by including the effects of adjacent segments. Before such a model can be confidently used, however, its transferability must be confirmed for roads other than those used for the model’s development. This study therefore used data collected at two freeways to demonstrate transferability, as well as the advantages of the continuous speed model over the spot-based model. Results of the spot-based model showed large prediction errors, and changes in the predicted speeds along the road were abrupt and discontinuous. On the other hand, the continuous model’s prediction errors were smaller and the predicted speed profile was, as expected, continuous. The continuous model performed well at estimating operating speed on the studied freeway and, most importantly, it can predict operating speeds for out-of-sample roads of the same type as the studied roads. That is, it passed the transferability test. This finding opens an opportunity for evaluating roads in the design stage while minimizing the number of costly driving simulation experiments. Transferring a continuous speed model is a recommended alternative, particularly when high-priced construction is required for roads with challenging conditions such as mountainous terrain.

Language

  • English

Media Info

Subject/Index Terms

Filing Info

  • Accession Number: 01764614
  • Record Type: Publication
  • Files: TRIS
  • Created Date: Jan 6 2021 3:17PM