Iterative Route Planning for Modular Transportation Simulation

The TRANSIMS (TRansportation ANalysis and SIMulation System) project is a large scale transportation system project produced by Los Alamos National Laboratory for transportation planning. In TRANSIMS, all processes are represented on the microscopic level. These processes range from decisions of individuals about their daily activities all the way to signal operations and traffic movements. TRANSIMS consists of several modules, some of which are listed here: Route planner, which generates travel plans for each driver. Micro-simulation, which executes all plans simultaneously and in consequence computes the interaction between different travelers, leading e.g. to congestion. Feedback: The above modules are interdependent. For example, plans depend on congestion but congestion depends on plans. This is solved via an iterative method, where an initial plans set is slowly adapted until it is consistent with the resulting travel conditions. As part of the eventual goal of implementing the TRANSIMS software for all of Switzerland, we are running simulations on a test-case with the Switzerland transportation network. We use a similar simulation framework as found in TRANSIMS, but with our own, simpler versions of the three modules. We discuss the operation and interaction of these modules, and bring to light a combined flaw in our route planner and feedback modules. This flaw initially caused several unrealistic simulation results, such as freeways being avoided by vehicles in favor of lower-capacity roads. We illustrate several improvements made to the modeling logic of the modules in an effort to correct these problems, and compare simulation results from the various methods. We also discuss the results of our most substantial improvement, which is the addition of a database that gives each driver a “memory” of its past routes from earlier iterations, plus the performance of those routes. When a new plan-set is generated, each driver chooses a route from those in its memory, based on their relative performance. This solution appears to be very robust, because it does not depend on having a route planner that works perfectly all the time.