This paper presents a traffic forecasting model that consists of a dynamic Origin-Destination (OD) estimation module and a dynamic assignment module. Both modules are designed to utilize data obtained from the Dutch Motorway Traffic Management (MTM) system. The aim of the work presented is to develop models that are valid from a theoretical point of view, while paying attention to feasibility, both technical and practical. Data was collected on a continuous basis during three weeks at 141 locations on the Amsterdam beltway. The aim is to test the combined OD estimation/dynamic assignment model. As an alternative to the developed traffic prediction system, historical averages were compiled from all observed from all observed data. Comparisons between true, predicted, and averaged data show that a lot of effort must be invested in dynamic forecasting procedures in order to get equally good or better predictions compared to those obtained from historical averages. This is explained by the fact that predictions obtained from historical data are not hindered by mis-specification of OD network and network characteristics. Severe incidents, however, were best reproduced with the dynamic forecasting procedure. The results suggest that the ideal forecasting procedure is a combination of both approaches.
Abstract