We have implemented a computational feedback between demand generation and traffic simulation in a real world setting in Portland/Oregon. This was done via a double relaxation loop: an inner loop for relaxation of the route assignment with fixed demand, and an outer loop for relaxation of the demand. Typically, about 70 runs of the traffic micro-simulation are necessary for one relaxed result. We have used data from Portland/Oregon.
For simplicity, we have concentrated on assigning workplaces to workers (whose home locations were given). The challenge was to perform this workplace assignment self-consistently such that the resulting trip times correspond to the trip time distribution given via census data.
Our results demonstrate that with current computational technology and simple models, it is possible to do such studies while retaining microscopic resolution throughout the whole computation. Microscopic resolution here means that each of the about 500000 travelers and each vehicle are represented individually in each step of the method. Our simulations were run on single CPU workstations; one relaxation series typically took about four days of computer time.
Because of the many simplifications, we did not expect our results to be a good model of reality. Nevertheless, in order to provide a benchmark we compared our results to real world morning peak volume counts from the Portland/Oregon area, and we included into the comparison results of an older study by Portland Metro using different methods. These results are summarized in Fig. 36.4. It is encouraging that one gets so close with so relatively little investment in terms of input data. In fact, input data consists of nothing more but the EMME/2 street network information, some population characteristics from the census (home locations of workers; overall trip time distribution for home-to-work trips; overall trip starting time distribution), and the locations of workplaces. The methodology uses a relaxation algorithm of workplace assignment, a fastest-path routing, and a queuing micro-simulation. Our study demonstrates that such a microscopic approach is both computationally and methodologically feasible even on modest computing hardware.