Performance improvements for large-scale traffic simulation in MATSim

Abstract

In contrast to aggregated macroscopic models of traffic simulation, multi-agent microscopic models, such as MATSim, enable modeling of individual behavior and facilitate more detailed traffic analysis. However, such detailed modeling also leads to an increased computational burden, such that simulation performance becomes critical. This paper looks specifically at the MATSim simulation framework and proposes several ways to improve its performance. This is achieved through a combination of several approaches, including reducing disk access, decoupling computational tasks, and making use of parallel computing. Additionally, for the traffic simulation, an event-based model is adopted instead of a fixed-increment time advance approach. Experiments show that by applying these methods, a simulation speedup of four times and more is achieved (depending on the scenario) when compared to the current Java micro-simulation in MATSim. Initial simulation experiments on a high-resolution navigation network of Switzerland – containing around one million roads and 7.3 million agents – demonstrate that real-world scenarios can now be executed in around one-and-a-half weeks using the improved model. Ways to further shorten the computational time of MATSim are also described.