Game‐Theoretic Multi‐Agent Dynamic Traffic Assignment Using Congestion‐Aware Adaptive Learning

This paper proposes a novel game‐theoretic multi‐agent framework for dynamic traffic assignment (DTA) problems, integrating congestion‐aware adaptive learning to simultaneously reduce individual travel time and improve overall network efficiency under dynamic congestion. First, a multi‐agent DTA framework based on a potential game is formulated. Second, a time‐varying and congestion‐aware learning mechanism is designed to capture the impact of changing network conditions on travel decisions. Third, a proof of global convergence is presented, ensuring that any Nash equilibrium achieves at least 50% of the optimal social welfare under certain conditions, and the best equilibrium aligns with the optimal solution. Finally, SUMO‐based simulations demonstrate that the proposed framework outperforms baseline DTA frameworks by reducing the root mean square error in link flow and link utilization to 580 vehicles and 5%, respectively, in the Sioux Falls network across light‐to‐high disruption levels.