Day-to-Day Traffic Dynamics under Rational Inattention and Limited Memory: Modelling and Efficient Computation
2026-99-2431
11/20/2026
- Content
- This paper develops a day-to-day (DTD) traffic flow evolution model that integrates the Rational Inattention (RI) route-choice probability framework with a geometrically weighted limited-memory perception mechanism. In the proposed model, travellers update their perceived path costs as a decaying weighted average of the past M days of realised costs and adjust their flows proportionally toward the RI-based Logit distribution. We show that the fixed point of this dynamical system is equivalent to the static Rational Inattention User Equilibrium (RIUE), provide local stability conditions via an augmented block-companion Jacobian, and validate on a two-path network that the deterministic mean-field process closely approximates the stochastic agent-based dynamics in the large-population limit. To enable efficient equilibrium computation, we introduce the Adaptive Relaxed Anderson Acceleration (AAmd) algorithm; at each iteration it selects an optimal relaxation parameter without additional function evaluations. Experiments on the Sioux Falls network show that AAmd consistently converges faster than the standard Method of Successive Averages (MSA). AAmd reaches convergence 1.38× faster than MSA and an additional 1.17× faster than classical Anderson Acceleration, while achieving the same solution accuracy.
- Citation
- Liu, Y. and Zhao, T., "Day-to-Day Traffic Dynamics under Rational Inattention and Limited Memory: Modelling and Efficient Computation," 2026 2nd International Conference on Intelligent Transportation and Future Mobility (ITFM2026), Xi'An, China, March 27, 2026, https://doi.org/10.4271/2026-99-2431.