This report presents a review of methods for forecasting the future traffic situation up to 30 minutes hence. These methods are: (1) adaptive forecasting; (2) neural networks; (3) origin destination matrices; and (4) simulation models. A number of variables were used to study these methods. It appears that traffic intensity, travel times and queue length are relevant variables, and that speed and traffic composition are variables of minor importance. It is concluded that the four methods have some potential making short term predictions. The less attractive method however is origin destination matrices: the method is complex and has to be used in combination with a quite sophisticated simulation model. Good results can be obtained for road sections such as the one from Utrecht to Amsterdam, by using a combination of different techniques. Relevant research which has applied these techniques in this way has already been carried out.
Abstract