SharedEdge

Abstract

Estimating travel time on the highway in real time is of great importance for transportation services. Previous work has been mainly focusing on the city scale for a particular transportation system, e.g., taxi, bus, and metro. Little research has been conducted to estimate fine-grained real-time travel time in state-level highway systems. This is because the traditional solutions based on probe vehicles or loop sensors cannot scale to state-level highway systems due to their large spatial coverage. Recently, the adoption of Electric Toll Collection (ETC) systems (e.g. EZ-pass) brings a new opportunity to estimate the real-time travel time in the highway systems with little marginal cost. However, the key challenge is that ETC data only record the coarse-grained total travel time between a pair of toll stations rather than fine-grained travel time in each individual highway edge. To address this challenge, we design SharedEdge to estimate the fine-grained edge travel time with large-scale streaming ETC data. The key novelty is that we estimate real-time fine-grained travel time (i.e., edge travel time) without using fine-grained data (i.e. GPS trajectories or loop sensor data), by a few techniques based on Bayesian Graphical models and Expectation Maximization. More importantly, we implement our SharedEdge in the Guangdong Province, China with an ETC system covering 69 highways and 773 toll stations with a length of 7, 000 km. Based on this implementation, we evaluate SharedEdge in details by comparing it with some baselines and the state-of-the-art models. The evaluation results show that SharedEdge outperforms other methods in terms of travel time estimation accuracy when compared with the ground truth obtained by 114 thousand GPS-equipped vehicles.