Flow-inflated selective sampling: Efficient agent-based dynamic ride-sharing simulations

Agent-based simulations have become a popular and powerful tool for simulating emergent mobil- ity modes. Often times, the memory and computing requirements are daunting. Scaling down agent populations by simulating only a fraction of all agents is a frequently used option to reduce these burdens. However, recent studies have pointed out the difficulty of scaling ride-sharing simulations as these rely heavily on demand density and do not scale linearly. In this study, we introduce a simple yet effective methodology for simulating dynamic ride-sharing services, which we call flow-inflated selective sampling (FISS). The basic operation is that, similar to scaling agent-based populations, only a fraction of the actual agents are explicitly assigned. However, here only trips of private car transport are sampled, while public transport as well as ride-sharing vehicles are fully represented. In contrast to scaling in previous studies, the network capacity is not adjusted. Rather, the capacity consumption of the cars is scaled up to obtain realistic traffic flows. We implement this approach in the MATSim simulation environment for a large scenario in the region of Munich, Germany and show that our approach preserves traffic flows while keeping key performance indicators of a ride-sharing service stable and mostly unbiased. Mode choice decisions based on this approach also remain stable. By introducing our approach, the run-times of the actual assignment can be almost halved