Agent-Based Dynamic Traffic Assignment with Information Mixing

Auld, Joshua; Verbas, Omer; Stinson, Monique · 2019 · Crossref

DOI: 10.1016/j.procs.2019.04.119

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 challenge of achieving efficient convergence in agent-based dynamic traffic assignment (DTA) models. As transportation simulation systems increasingly integrate activity-based demand models with supply-side traffic simulations, accurately modeling the circular interactions between routing decisions and congestion becomes computationally intensive. The authors aim to improve the speed and stability of DTA convergence by developing an individualized approach that utilizes a weighted mix of prevailing and historical network conditions, rather than relying solely on static or purely historical data. The study implements this methodology within the POLARIS agent-based modeling framework. The approach is gap-based, meaning re-assignment decisions are driven by the relative difference between the routed (estimated) travel time and the experienced travel time from the previous iteration. The core innovation is an information mixing mechanism that determines link costs by weighting current prevailing traffic conditions against historical time-dependent conditions. This weight is calculated individually for each agent using a modified two-parameter Weibull survival function. The function incorporates the agent’s relative gap from the previous iteration and the iteration number, effectively allowing agents to "trust" historical data more for future time steps while relying on prevailing conditions for immediate travel. The model uses a mesoscopic traffic simulation based on Newell’s Simplified Kinematic Waves model to provide feedback on experienced travel times. The methodology was tested on a medium-scale network of Bloomington, Illinois, comprising 2,540 nodes, 7,023 links, and approximately 439,249 passenger car trips. The authors compared four scenarios: a static case, a standard gap-based DTA using only historical conditions, a gap-based DTA with information mixing where the gap does not affect weighting, and the proposed gap-based information mixing where the weighting parameter is adjusted by the iteration number. Results showed that the standard gap-based approach converged to a relative gap of approximately 2.5% after seven iterations but exhibited significant fluctuations. In contrast, the proposed method with gap-based information mixing converged after only two iterations with minimal fluctuation, demonstrating superior stability and speed compared to the standard approach. The significance of this work lies in its ability to drastically reduce the number of iterations required for DTA convergence, which is critical for large-scale network applications where computational time per iteration is high. By individualizing the convergence process and dynamically adjusting the reliance on historical versus prevailing data based on individual traveler gaps, the method offers a more efficient path to user equilibrium. The authors conclude that this approach is promising for future large-scale applications, such as modeling the Chicago metropolitan area, and suggest further sensitivity analyses on the Weibull function parameters to optimize the trade-off between convergence speed and accuracy.

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.

• gap acceptance