Methodology for O-D matrix estimation using the revealed paths of floating car data on large-scale networks

Abstract

The increasing availability of historical floating car data (FCD) represents a relevant chance to improve the accuracy of model-based traffic forecasting systems. A more precise estimation of origin-destination (O-D) matrices is a critical issue for the successful application of traffic assignment models. The authors developed a methodology for obtaining demand matrices without any prior information, but just starting from a data set of vehicle trajectories, and without using any assignment model, as traditional correction approaches do. Several steps are considered. A data-driven approach is applied to determine both observed departure shares from origins to destinations and static assignment matrices. Then the O-D matrix estimation problem is formulated as a scaling problem of the observed FCD demand and carried out using as inputs: a set of traffic counts, the FCD revealed assignment matrix and the observed departure shares as an a-priori matrix. Four different optimisation solutions are proposed. The methodology was successfully tested on the network of Turin. The results highlight the concrete opportunity to perform a data-driven methodology that, independently from the reliability of the reference demand, minimises manual and specialised effort to build and calibrate the transportation demand models.