Warm-starting dynamic traffic assignment with static solutions

Abstract

Dynamic traffic assignment (DTA) convergence can be accelerated by providing an initial feasible solution from static traffic assignment (STA). While this strategy seems natural, it has received little attention in the literature despite a number of choices which must be made, including STA stopping criteria, determining STA demand from dynamic demand, how to create a path-based DTA solution from the STA solution and the choice of STA algorithm itself. This paper studies these variations in the context of a cell-transmission based DTA algorithm. Experimental results on two networks of different size indicate that the STA solution can provide significant improvements in convergence despite the time-varying and flow model differences. A sensitivity analysis provides evidence that these benefits to convergence are likely to persist over varying inputs. The analysis further suggests that although the optimal parameter choice depends on the network, practitioners will observe significant improvements in convergence for a fairly wide range of reasonable parameters.