Quasi-Dynamic Traffic Assignment using High Performance Computing

Traffic assignment is one of the key approaches used to model the congestion patterns that arise in transportation networks. Since static traffic assignment does not have a notion of time dynamics, it is not designed to represent the complex dynamics of transportation networks as usage changes throughout the day. Dynamic traffic assignment methods attempt to resolve these dynamics, but require significant computational resources if modeling urban-scale regions and often take days of compute time to complete. The focus of this work is two-fold: 1) to introduce a new traffic assignment approach: a quasi-dynamic traffic assignment (QDTA) model and 2) to describe how we parallelized the QDTA algorithms to leverage High-Performance Computing (HPC) capabilities and scale to large metropolitan areas while dramatically reducing compute time. We examine and compare the user-equilibrium model (UET) to a baseline static traffic assignment (STA) model. Results are presented for the San Francisco Bay Area which accounts for 19M trips/day and an urban road network of 1M links and is validated against multiple data sources. In order to select the gradient descent step size, we use a line search using Newton's method with parallelized cost function evaluations and compare it to the method of successive averages (MSA). Using the parallelized line search provides a 49 percent reduction in total execution time due to a reduction in the number of gradient descent iterations required for convergence. The full day simulation using results of 96 optimization steps over 15 minute intervals runs in \textasciitilde6 minutes utilizing 1,024 compute cores on the NERSC Cori computer, representing a speedup of over 34x versus serial execution. To our knowledge, this compute time is significantly lower than any other traffic assignment solutions for a problem of this scale.