Characterizing the settling time for microscopic simulations of urban traffic for application to CAVs

Due to their cost efficiency, safety improvements, and repeatability, traffic simulators are an integral and widely used tool to study the performance of a traffic network with connected and/or autonomous vehicles (CAVs). Such simulations depend significantly on the choice of simulation parameters, and in most cases, it is up to the user to identify the appropriate setting for even a basic usage question: what duration should the simulation be run? It is generally understood that a simulation's duration should be sufficient to ensure that the simulations are fully initialized wherein the user-chosen initial conditions of the simulation do not affect assessments of steady behavior. This duration is generally chosen based on the user's practical experience. This paper presents a method for determining a simulation's initialization interval by examining the settling time for traffic simulations, e.g. the time the simulation requires to reach steady-state conditions. In this work, steady-state behavior is measured via the cumulative means of the speeds of edges on a traffic network. The completion times of vehicle trips were also evaluated. Two examples of virtual and real-world networks were tested with Simulation of Urban MObility (SUMO) traffic simulator, and the results show that: i) for the given networks, the settling times were 0.63 and 0.73 hours for virtual and real-road networks respectively, and ii) the ratio between the mean of settling times and the mean of trip completion times for the real-world network is higher than the one for the virtual network, approximately 20 times versus 10 times longer respectively. An important insight of this work is that the required simulation time for the traffic simulation to converge to steady-state behavior is a large multiplier of the average trip time, and this duration is far longer than what is often seen in the literature.