Effects of scaling down the population for agent-based traffic simulations

Llorca, Carlos; Moeckel, Rolf · 2019 · OpenAlex-citations

DOI: 10.1016/j.procs.2019.04.106

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This paper addresses the computational challenges inherent in agent-based transport models, which simulate individual travel demand at high resolution. While these models offer detailed insights, simulating large populations across extensive road networks requires significant computational resources and long runtimes. To mitigate this, researchers commonly subsample the population and proportionally scale down road network capacity. However, previous studies have noted inconsistencies in results, such as unexpectedly high travel times, without fully understanding the underlying mechanisms or the optimal scaling parameters. This study aims to systematically quantify the effects of scaling down the synthetic population on simulation outcomes, specifically examining runtime, travel times, and link volumes. The authors conducted experiments using the multi-agent simulator MATSim for the Munich metropolitan area. They generated a synthetic population with 2.9 million daily car trips and utilized two road network resolutions derived from OpenStreetMap: a "fine" network including all roads and a "coarse" network excluding residential roads. The experimental design varied six scale factors (1%, 5%, 10%, 20%, 50%, and 100%), three exponents for the storage capacity factor (0.75, 0.85, and 1.00), and five iteration counts (50 to 500). This matrix allowed for a comprehensive analysis of how different scaling strategies and network details influence model performance and output accuracy. The results indicate that model runtime increases linearly with both the number of simulated agents and the number of iterations. Regarding travel times, the study found a distinct U-shaped relationship with the scaling factor: average travel times were minimized at scale factors of 10% to 20%, while the 1% scale factor produced significantly higher travel times across all conditions. Network resolution also played a critical role; the fine network yielded lower average travel times and less variation across scales compared to the coarse network, which suffered from excessive load and routing inefficiencies for short trips. Furthermore, increasing the number of iterations reduced average travel times, particularly for full-sample simulations, by allowing agents to better distribute across routes and avoid random saturation of specific links. The exponent used for storage capacity adjustment had a comparatively minor impact on results. The significance of this work lies in providing empirical guidance for configuring large-scale agent-based simulations. The findings suggest that subsampling at 10% to 20% offers an optimal balance between computational efficiency and result accuracy, avoiding the distortions seen at very low scales (1%) or the high computational costs of full samples. Additionally, the study highlights the importance of using detailed road networks to prevent artificial congestion and routing errors. These insights help researchers and practitioners design more reliable and efficient transportation models, ensuring that scaling strategies do not compromise the validity of the simulated traffic dynamics.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• traffic density