First let us briefly describe the traditional approach to the problem. The traditional urban transportation planning process consists of four steps []:
).
Traditional assignment assumes that traffic streams are constant throughout the evaluation period. For that reason, people often concentrate on something like the ``A.M. peak'' or the ``P.M. peak'' period. When these time periods become too short, the theoretical foundation of these models is no longer valid. However, longer time periods (such as an hour) are not capable of generating dynamic short-term effects, such as queue build-up or congestion spreading during the onset of the rush hours.
TRANSIMS, in contrast, generates activities. That is, origin and destination information is always attached to individual people. Nevertheless, TRANSIMS also at a certain point has all the origin-destination relations given and needs to assign these on a network. One could even translate the TRANSIMS information to origin-destination matrices, although one would give away information, such as exact starting times of trips, and information from trip chaining, e.g. the effect that delays in the morning may lead to later departure times in the evening.
Iterated microsimulations of traffic provide an alternative for the assignment portion. Several groups (e.g. [,,,,,]) have used the iterative approach of routing-microsimulation-feedback of travel-times to obtain an assignment (route set) that is, within the accuracy permitted by any implicit or explicit stochasticity of the model, self-consistent. In this paper we outline part of the framework of TRANSIMS [,] as it was used for an extensive case-study of the Dallas / Fort Worth (Texas) street network, and for preliminary studies using data from the Portland (Oregon) metropolitan area.
For the ``Dallas'' study [], demand for travel was given by origin-destination matrices provided by the local transportation planning authority (NCTCOG). The study used a so-called focused road network, which means that for a 5 miles times 5 miles study area all streets were represented, whereas with further distance from the study area more and more of the less-important streets were left out. This network contained 9864 nodes and 24622 uni-directional links, of which 2276 nodes and 6124 links were in the study area.3
The Portland study is planned to be run on the complete road network of Portland/Oregon, including all local streets. This network contains about 200000 uni-directional links. We also use another network, with 20024 uni-directional links, which has been used by the local transportation planning authority (Portland Metro) for their traditional assignment studies. For the Portland study, demand generation will be done via activities generation, as intended by the TRANSIMS design. Results in this paper are based on very preliminary sets of home-to-work trips []. This should be of no consequence for the computational results presented in this paper.
The simulation set-up itself consists of two main applications: (a) a route planner based upon a fast implementation of the Dijkstra algorithm that uses time-dependent link travel-times to compute shortest routes, and (b) a traffic microsimulation that executes the routes generated by the planner and supplies a feedback file which is used in subsequent calls of the router.
In the following sections we will describe the two applications and their mutual dependency in more detail.