Patrice Marcotte, Laura Wynter
Transportation Science
This paper presents a method for estimating missing real-time traffic volumes on a road network using both historical and real-time traffic data. The method was developed to address urban transportation networks where a non-negligible subset of the network links do not have real-time link volumes, and where that data is needed to populate other real-time traffic analytics. Computation is split between an offline calibration and a real-time estimation phase. The offline phase determines link-to-link splitting probabilities for traffic flow propagation that are subsequently used in real-time estimation. The real-time procedure uses current traffic data and is efficient enough to scale to full city-wide deployments. Simulation results on a medium-sized test network demonstrate the accuracy of the method and its robustness to missing data and variability in the data that is available. For traffic demands with a coefficient of variation as high as 40%, and a real-time feed in which as much as 60% of links lack data, we find the percentage root mean square error of link volume estimates ranges from 3.9% to 18.6%. We observe that the use of real-time data can reduce this error by as much as 20%. © 2011 Elsevier Ltd.
Patrice Marcotte, Laura Wynter
Transportation Science
Jingrui He, Wei Shen, et al.
IJCAI 2013
José R. Correa, Roger Lederman, et al.
Operations Research Letters
Fabian Lim, Laura Wynter, et al.
ICML 2022