Urban path travel time estimation using GPS trajectories from high-sampling-rate ridesourcing services

Abstract

Link-Travel-Time (LTT) estimation is essential for the planning and operations of a variety of transportation services. Given the random sampling of a very large number of GPS-points over a highly complex urban network, the task of organizing these individual GPS readings to estimate LTTs requires the development and implementation of a novel comprehensive data processing and path-finding methodology which is described in detail in this paper. As part of this novel methodology, an innovative data-driven matching-algorithm to estimate urban LTT from high-sampling-rate GPS data projected onto the Open-Street-Map network is developed and implemented. Then, using these LTTs, we construct Path-Travel-Time (PTT) between major origin-destination pairs. PTT of Actual-Paths (AP) followed by GPS-enabled vehicles are compared with k-Shortest-Paths (SP), allowing us to better understand route-choice behavior and overall traffic conditions. We compare PTT from observed-trips (OD-trips), map-matched AP, and SP paths with Free-Flow (FF). Results show that OD-trips, AP, and SP exceed FF by 15%, 41%, and 15%, respectively. The difference in PTT between OD-AP is ∼5%, which means the map-matching process works well and does not create bias in our analysis. People using the shortest-path varies with the distance; for ∼3-mile-paths, 50% of users do not use it. For ∼6-mile-paths, the percentage reduces to 35%, and for ∼9-mile, the percentage is 25%. A relatively high number of trips spend more time than the average and much longer than the shortest PTT.