Factors Affecting Convergence and Stability in Simulation-based Dynamic Traffic Assignment



Factors Affecting Convergence and Stability in Simulation-based Dynamic Traffic Assignment

Authors

M Mahut, M Florian, N Tremblay, INRO Consultants Inc., CA

Description

Various factors have a significant impact on the convergence and stability of simulation-based DTA models, such as the size and complexity of the network, overall level of congestion, temporal discretization and stochastic effects.

Abstract

As a network modeling approach, Dynamic traffic assignment (DTA) can be positioned between static assignment models and traffic micro-simulation models, both in terms of the level of modelling detail and also the appropriate network size. Compared to static models, simulation-based DTA models have a more realistic representation of traffic flow, and a more detailed representation of the transportation system. They are particularly useful when the level of congestion in the network, or the requirement for modeling more complex system components (e.g. adaptive control schemes, providing guidance information to drivers) make the application unsuitable for static assignment methodology.

Although the higher level of realism results in the loss of certain well-known mathematical properties associated with static models, such as the existence and uniqueness of the equilibrium assignment solution, simulation-based DTA models have proven to be fairly well-behaved in practice when used with correct input data, and they generally tend to converge to a stable solution that is within a reasonable distance (as measured by the relative gap) of true equilibrium.

This paper examines various factors that have a significant impact on the convergence and stability of simulation-based DTA models. These factors include the size and complexity of the network, overall level of congestion, temporal discretization and stochastic effects. We present some insights into these factors obtained through numerical experiments on actual networks, and also compare the performance of different assignment algorithms in the presence of these factors.

One factor is not considered explicitly is the level of fidelity of the embedded traffic model -- which can vary considerably between different DTA models ? as the tests are carried out using a single modeling methodology. Although this DTA model is built around a relatively high-fidelity traffic simulator, the findings and insights are generally applicable to all DTA models.

Publisher

Association for European Transport