The purpose of this paper and this study is to demonstrate that for multi-module transportation simulations, not only is the functionality of the single modules important, but also how they interact. In particular, an agent-based implementation of the interfaces between the modules is capable of correcting for artifacts in the modules. An agent-based representation means that travelers are considered as agents, which have a memory of different strategies and their respective performances. In general, they chose the strategy with the best performance, but from time to time re-try one of the other strategies just to check if its performance is still unchanged. Also from time to time, new strategies are generated and added to the pool.
In this particular example, we apply this approach to route feedback for dynamic traffic assignment. The problem was that the router uses aggregated feedback information from the micro-simulation, and that this aggregation with most plausible algorithms lead to artifacts in the resulting traffic. Specifically, the router under-estimated long distance travel times, leading to the fact that the router assumed the existence of congestion for later parts of the trip while in fact the congestion was long gone. This resulted in travelers using the side roads where the freeway would have been much better. The use of the agent data base solves this problem without any changes in the router. That is, even when the router consistently generates faulty plans, the agent database approach will compensate for this as long as at least some of the routes are plausible. The approach was implemented using MySQL as a data base, and AWK as scripting languages. The agent database was then applied to a simulation of the morning rush hour of ``all of Switzerland''. The results of this are reported elsewhere (45).