Incident-Induced Delays on Freeways

Abstract

Freeway incidents cause nonrecurrent congestion, which contributes significantly to overall travel delays. This paper provides a method to estimate the magnitude of delays resulting from incidents. Incident-induced delays can be estimated by establishing an incident-free reference travel time profile as a basis from which the incident-induced delay is measured. The main objective of this paper is to show how to establish the reference profile from a subset of candidate profiles of similar traffic patterns. To identify similar traffic patterns, a hybrid of travel time and volume profiles was used. To establish the reference profile, three methods for candidate profile identification were investigated: day of the week based, cluster analysis based, and k–nearest neighbor based. Weighted summations of the candidate profiles were also investigated: arithmetic average, Shepard's inverse distance method, and rank exponent method. A ground truth experiment was conducted, and the results showed that the best way to formulate the reference traffic profile was by using the k–nearest neighbor in combination with the weighted summation of rank exponent method. The proposed approach provides practical guidance to estimate incident-induced delays more accurately.