A New Adaptive, Multi-scale Traffic Simulation
Michael Mahut, INRO, Michael Florian, INRO, Daniel Florian, INRO
We propose a different approach to scalable traffic simulation modeling which adapts continuously to the level of traffic congestion. This is in contrast with approaches that apply different traffic models to different areas of the network.
In recent years there has been a motivation to introduce greater realism and fidelity to traffic models on larger applications by employing traffic simulation approaches. This can be attributed to increasing traffic congestion, which may be captured more realistically in simulation models, and to increasingly complex mechanisms being investigated and the corresponding model sensitivities required, e.g. for departure time choice models, peak spreading, congestion-based tolling schemes, intelligent transportation systems and others.
There is now considerable industry experience in the application of traffic simulation models on larger scales and the associated challenges related to the scalability of these models. In addition to increased computational requirements (run times) that may become impractical or even infeasible, there is also the fundamental question of model stability and robustness. The typical approach for larger applications has been to adopt two distinct models, microscopic and mesoscopic, in what may be known as a ‘multi-resolution’ or ‘hybrid’ approach. The underlying models are distinct and operate at only one scale. However, model outputs can change significantly depending on where the micro/meso boundaries are drawn, and the lower fidelity of some mesoscopic models may mean that important details are ignored.
We propose a different approach to scalable dynamic traffic modeling which continuously adapts to the level of traffic congestion in order to prevent non-linear responses such as cascading queues and gridlock. In this way, model stability is ensured even when there are significant demand/supply imbalances in certain areas of the network. The new approach described here is referred to as “multi-scale” because it uses a single model, with a consistent level of detail throughout the network, and an adaptive simulation approach to provide enhanced scalability for larger geographies, higher demands, and more congested conditions.
The multi-scale simulation model was implemented in the Dynameq traffic simulation software and evaluated on several congested real-world networks. The tests examined the impacts of the multi-scale simulation on model convergence, stability and run time, and sensitivity to random seeds. Tests demonstrate that the multi-scale approach is effective in improving model stability in congested scenarios and in reducing run time to convergence for congested dynamic traffic assignments. Compared with hybrid approaches, the consistent level of detail throughout the network provides transparency even across wide areas. The model furthermore provides lane- and turn-based diagnostics which identify critical bottlenecks, with significant demand/supply imbalances, for further investigation and model refinement.
The tests demonstrate that the multi-scale approach using a single, adaptive traffic simulation model scales well and avoids key drawbacks characteristic of approaches which rely on fundamentally different and inconsistent models on separate parts of the network.
Association for European Transport