Comparison of dynamic-equilibrium assignment methods for traffic simulation models.

This paper reports on the evaluation of alternative dynamic-equilibrium assignment methods for use in conjunction with traffic simulation models. Since the realism of a traffic simulation model results in an assignment map which is discontinuous and difficult to characterize analytically, algorithms for the static assignment problem are not directly applicable. The method of successive averages (MSA) has been applied to this problem (in the space of paths) and over the last few years has consistently performed well on a variety of real-world networks of considerable size. This paper evaluates two variants of this model inspired by gradient-based methods specifically, the projected gradient and reduced gradient and further tests these methods in conjunction with a dynamic step-size adjustment techinique. These path-based methods are compared with a recently developed splitting-rate model that is based on fundamental properties of traffic flow. The tests are executed on several congested urban networks of significant size (up to 6500 links) from North America and Europe. This paper reports on the evaluation of alternative dynamic-equilibrium assignment methods for use in conjunction with traffic simulation models. Since the realism of a traffic simulation model results in an assignment map which is discontinuous and difficult to characterize analytically, algorithms for the static assignment problem are not directly applicable. The method of successive averages (MSA) has been applied to this problem (in the space of paths) and over the last few years has consistently performed well on a variety of real-world networks of considerable size. This paper evaluates two variants of this model inspired by gradient-based methods specifically, the projected gradient and reduced gradient and further tests these methods in conjunction with a dynamic step-size adjustment techinique. These path-based methods are compared with a recently developed splitting-rate model that is based on fundamental properties of traffic flow. The tests are executed on several congested urban networks of significant size (up to 6500 links) from North America and Europe. For the covering abstract see ITRD E137145.