Convergence behavior for traffic assignment characterization metrics

Abstract

Traffic assignment is used for infrastructure planning, based on metrics like total system travel time (TSTT), vehicle-miles traveled (VMT) and link or path flows. Algorithms for traffic assignment converge to an equilibrium solution over multiple iterations, but these metrics converge at different rates. Current guidance indicates that freeway link flows stabilize at a relative gap of roughly . This study generalizes this guidance by testing additional networks and metrics, in more experimental settings. Our results reveal that aggregate metrics (VMT and TSTT) stabilize earlier (relative gap ) than link flows (relative gap ), which in turn stabilize slightly before the set of most likely used paths and flows on these paths (relative gap ). These results are stable across the TAPAS and Algorithm B methods for solving assignment. Our results also show strong linear correlations between alternative gap measures, allowing for the translation of stabilization results across other gap definitions as well.