Traffic Simulation Performance Optimization through Multi-Resolution Modeling of Road Segments

In an agent-based traffic simulation the level of detail is crucial to the system's runtime performance as well as the fidelity of the results. Therefore, different model abstractions have been used throughout literature. Macroscopic, mesoscopic and microscopic models have their use-cases and benefits. Microscopic traffic simulations have a high level of detail but at the same time require a large amount of computational resources. In a large traffic network of a mega-city or an entire country, the use of a complete microscopic simulation is just not feasible. The resource required to do so are for most use-cases in no relation to the actual outcome. We propose a hybrid traffic simulation model that uses both, a high-resolution agent based microscopic simulation alongside a lower resolution flow-based macroscopic simulation for specific road segments. The problem with using different simulation models is the fidelity at the boundary between such simulation models. This fidelity discrepancy is caused by the difficulties with aggregation and disaggregation passing through the boundary. We show, in this paper, that the computational performance (simulation time) can be improved by $20\%$ while maintaining a relative high accuracy of below $5\%$ deviation from a pure microscopic simulation.