OpenStreetMap for Traffic Simulation

Michael Zilske, Technische Universität Berlin

Details about the talk:
Micro-simulations of traffic systems are becoming more important as highly disaggregated mobility data, such as GPS traces, become available.

The traffic demand model used in the research project MATSim is based on individual agents which travel the road network. They are equipped with mobility plans for an entire day. Each plan contains timed trips between georeferenced sites. These plans are concurrently executed in a simulation of the traffic system. At the end of the day, the agents score the outcome of their plans with an individual utility function. Using their experience from the simulated traffic system, they can make adaptations to their plan. Choice dimensions include route choice, changing activity times, or switching to public transit. These two steps, simulation and re-planning, are iterated until an equilibrium is reached. After some validation, the resulting scenario can then be taken as a base case to predict the reaction of the system to policy measures, such as introducing tolls, building new roads or improving public transit.

In order to produce interpretable results, reliable input data is required. On the demand side, this means census data or activity-travel diaries. On the supply side, we need a high-quality model of the road network. Recently, we have started using OpenStreetMap for our case studies, and it has proven a valuable data source, often on par with and sometimes surpassing the quality of proprietary network models provided by adminstrations. The quality of the simulation depends crucially on the reliability of two attributes for each link: mean free-flow speed, and capacity in vehicles per hour. We assume values based on the OSM highway-types and, where available, number of lanes. For a recent case study, we have been able to tweak these attributes by incorporating additional knowledge from floating car data (FCD) collected from taxis, which we map-matched to the OSM network.

A more detailed simulation requires additional attributes which are not yet present in OSM, but which we believe may be of independent interest for other applications. For example, in order to simulate spillback from congested turn lanes, one needs a detailed representation of the intersection layout, including turn lane lengths. Ongoing work includes the development of a tagging scheme and a custom JOSM plugin to collect and encode these data.

MATSim is free software, available under the GPL.

Details about the speaker:
Studied computer science. Worked as a programmer, developing software for public transit scheduling. Now a research associate in the Transportation Systems Planning and Intelligent Transport Systems workgroup at TU Berlin.

Theme: research