In general, we want to generate ``realistic traffic'' via computer simulation. Thus, our ultimate research goal is to have a model which, when applied to today's situation, will yield today's traffic, and when applied to a hypothetical scenario, will yield a meaningful prediction. In our actual implementations, however, we (as everybody else) make simplifications. We are, however, not interested in optimal solutions of the simplified problems; our interest is how close to reality we can get with our simplified models and computational procedures.
We envisage that such a realistic computer simulation will be a combination of population generation, activities generation, routes assignment, and traffic micro-simulation, coupled via feedback iterations. So what is done in the following is to pick (simple) versions of these modules, embed them into feedback iterations, and try this on real world input data. The research question was twofold: (1) What are the computational issues? (2) How close to reality (or not) does one get with simple assumptions?
The question of the necessary degree of realism in each of these modules is an open problem which will need further research. That question is not treated in this paper. We do not claim that the degree of realism (or not) chosen in any of the modules used for our investigation is the correct degree of realism in order to obtain meaningful results. In particular, we expect that more sophisticated demand generation techniques (e.g. Bowman, 1998; Doherty and Axhausen, 1998; Arentze et al, 1998) will lead to more realistic results. We do expect, however, that a systematic inclusion of transportation network impedance, as demonstrated in our study, will contribute to better and more robust models.
The problem for this paper is how to assign workplace locations to workers via using computer simulation. It is known from data where people live, and it is also known where they work, but one has to match these two sets of data. The problem is similar to the trip distribution step in the four step process. In the work described here, this is done via some strongly simplified assumptions. One of these simplifications is to only look at traffic resulting from people driving from home to work. By this one neglects, for example: delivery trucks, people returning from night shifts, travelers using alternative modes of transportation, etc. There is also much more complexity in the afternoon peak than in the morning peak. Again, our investigation is a demonstration of a computational procedure, not an attempt to obtain the most possible realistic results for a certain field problem.
Having said that, let us describe our scenario. Our scenario area is
Portland in Oregon. Our input data are: (a) a description of the
Portland transportation network; (b) a synthetic population based on
Portland demographic data; (c) a list of workplaces including location
and size; (d) the distribution 
of actually encountered
trip times 
from home to work by the Portland population; and (e) a
distribution of starting times. The problem for this study was to
match workers (who have home locations) and workplaces such that the
resulting traffic yields trip times which, when aggregated, match the census trip times.36.1