Consistency Between Convergence of Dynamic Assignment and Stochasticity of Microsimulation: Implication for Number of Runs

Abstract

Dynamic transportation models route vehicles by using the principles of dynamic user equilibrium. These models include a dynamic network loading (DNL) module that is used to evaluate link costs. However, an element of stochasticity creeps into the modeling framework when the analytical dynamic assignment (DA) procedure is used along with a stochastic microscopic DNL. A methodologically correct way of approaching this problem is by solving the entire DA with a microscopic DNL (DA-microDNL) model until convergence for a given random seed and then repeating the process with different seed values. This paper proposes an approach to determine the minimum number of replications of the DA-microDNL model to determine a statistically valid estimate of the measure of effectiveness (MOE). The approach was tested on a small and medium-size network having different demand and network characteristics. Results show that running the integrated DA-microDNL framework for a minimum number of replications provides a statistically significant MOE at much lower computation time. The consistent estimates obtained by using this approach would provide robust information to transportation planners and practitioners in evaluating the impacts of several policy decisions on network performance.