ch

Bryan Raney
Dept.Computer Science, ETH Zürich, CH-8092 Zürich, Switzerland
Telephone: +41 1 632 0892
Fax: +41 1 632 1374
E-mail: raney@inf.ethz.ch

Nurhan Cetin
Dept.Computer Science, ETH Zürich, CH-8092 Zürich, Switzerland
Telephone: +41 1 632 0891
Fax: +41 1 632 1374
E-mail: cetin@inf.ethz.ch

Andreas Völlmy
Dept.Computer Science, ETH Zürich, CH-8092 Zürich, Switzerland
Telephone: +41 1 632 7212
Fax: +41 1 632 1620
E-mail: res@vis.ethz.ch

Milenko Vrtic
Dept.Civil, Environmental and Geomatics, ETH Zürich, CH-8092 Zürich, Switzerland
Telephone: +41 1 633 31 07
Fax: +41 1 633 10 57
E-mail: vrtic@ivt.baug.ethz.ch

Kay Axhausen
Dept.Civil, Environmental and Geomatics, ETH Zürich, CH-8092 Zürich, Switzerland
Telephone: +41 1 633 3943
Fax: +41 1 633 10 57
E-mail: axhausen@ivt.baug.ethz.ch

Kai Nagel
Dept.Computer Science, ETH Zürich, CH-8092 Zürich, Switzerland
Telephone: +41 1 632 2754
Fax: +41 1 632 1374
E-mail: nagel@inf.ethz.ch

Zusammenfassung:

In a multi-agent transportation simulation, each traveler is represented individually. Such a simulation consists of at least the following modules: (i) Activity generation. (ii) Modal and route choice. (iii) The traffic simulation itself. (iv) Learning and feedback. In order to find solutions which are consistent between the modules, a relaxation technique is used. This technique has similarities to day-to-day human learning.

Using advanced computational methods, in particular parallel computing, it is now possible to run such a system for large metropolitan areas with 10 million inhabitants or more. This paper reports on such a simulation system for of all of Switzerland. Our focus is on a computationally efficient implementation of the agent-based representation, which means that in fact each agent is represented with an individual set of plans as explained above. A database is used to store the agent's strategies, which are loaded into the simulation modules as required; the modules then feed back individual performance measures into the database. This approach allows that additional modules can be coupled easily, and without degrading computational performance.

The set-up was tested for Swiss morning peak traffic. Hourly demand matrices were taken from work with the VISUM assignment package and converted to our needs. Routes were assigned via feedback learning using the agent data base. In other words, the current implementation uses a car-only versions of the modules (ii), (iii), and (iv). Resulting flow volumes are compared to the VISUM assignment results, and to field data.