Estimating Magnitude and Duration of Incident Delays

Abstract

Traffic congestion is a major operational problem on urban freeways. In the case of recurring congestion, travelers can plan their trips according to the expected occurrence and severity of recurring congestion. However, nonrecurring congestion cannot be managed without real-time prediction. Evaluating the efficiency of intelligent transportation systems (ITS) technologies in reducing incident effects requires developing models that can accurately predict incident duration along with the magnitude of nonrecurring congestion. This paper provides two statistical models for estimating incident delay and a model for predicting incident duration. The incident delay models showed that up to 85% of variation in incident delay can be explained by incident duration, number of lanes affected, number of vehicles involved, and traffic demand before the incident. The incident duration prediction model showed that 81% of variation in incident duration can be predicted by number of lanes affected, number of vehicles involved, truck involvement, time of day, police response time, and weather condition. These findings have implications for on-line applications within the context of advanced traveler information systems (ATIS).