Several groups are developing simulations which can microscopically simulate whole metropolitan areas in faster than real time (e.g. DYNAMIT, 2000; MITSIM, 2000; Mahmassani et al, 1995 (DYNASMART); Rickert, 1998 (PAMINA); Gawron, 1998 (LEGO); Rakha and Van Aerde, 1996 (INTEGRATION); Esser, 1998 (OLSIM) ). By ``microscopic'' we mean that each traveller is individually resolved. Thus, if one can generate detailed travel plans for each individual, these simulations can execute these plans, while recording for example where conflicts in the form of congestion delay the plans.
In consequence, it is only a question of time until it will be easy to couple such models with models of travel demand generation, as has been demanded for many years (e.g. Axhausen, 1990). Such a coupling will probably include a modal-choice-and-routing module (``router''), and it will do systematic feedback iterations between all the modules. That is, the results of the micro-simulation will be fed back into the router again and again until some relaxation with respect to route choice is obtained, and then the result will be fed back into the activities generation module, which will generate new activities which now take into account the slower speeds in the network caused by congestion.
In this paper, an early implementation of such a computational feedback of the microsimulation into the activities module is demonstrated. In fact, practitioners have often done some version of such a feedback, by adjusting origin-destination matrices in order to move the volume counts of the assignment model closer to reality. There are also computational procedures with respect to assignment models (e.g. Metaxatos et al, 1995). What will be done here is use such a computational procedure in connection with an explicit traffic microsimulation. We will however simplify in several ways: Cars will be used as the only mode, travel from home to work will be the only demand, and the traffic micro-simulation is rather simplified. The simulation will be iteratively adjusted towards the census trip time distribution. This is an early step, and we expect much progress in the near future. In particular, we expect that transportation microsimulation, where each traveller is individually resolved, will lend itself much better to integration with activity-based demand generation than the aggregating technique of traditional assignment does. Although the focus of our work was the compuration integration of dynamic traffic assignment with demand generation, we will compare our results with existing volume counts in the Portland/Oregon area.
The structure of the paper is as follows: In Sec. 36.2, the problem is stated, followed by a description of our approach with respect to demand generation and feedback (Sec. 36.3). After a discussion of related work (Sec. 36.4), the paper moves on to our actual study (Sec. 36.5) and its results (Sec. 36.6). The paper is concluded by a discussion and a summary.