Network-Wide Link Flow Estimation Through Probe Vehicle Data Supported Count Propagation

Network-wide dynamic link flow estimation is one of the challenging questions in transportation research. Most of the previous approaches rely on static or dynamic OD matrices. The proposed data-driven approach tackles the problem of link flow estimation as a local network propagation problem between cross-section measurement sites. Distinct propagation rules consider time-dependent travel speeds, turning fractions at intersections and vehicle gain-loss ratios between links. The rules are derived from recorded vehicle paths originating from a probe vehicle data (PVD) system including data from thousands of vehicles from several different fleets. The proposed approach introduces a algorithm with dedicated propagation rules and measures for evaluating propagation quality. Our approach is evaluated using Austria's nation-wide road network (including freeways, urban and rural roads) with approximately 224,443 links and 16,566 intersections as well as traffic count data from 664 cross-section measurement sites. Results show the general applicability of the approach, but also reveal several challenging situations, which have to be treated with suitable propagation strategies.