Travel time prediction for long trajectories on motorways is necessary to provide travelers with information needed to make reliable choices in route and modality. To get proper long-range predictions it is expected that very good short-range predictions on which the long-range prediction can be based are necessary. For the development of a short-range travel time predictor artificial neural networks have been chosen as the key technology, because of their usefulness in approximating non-linear relations. To get superior results from this technique, it is important to transform the available field measurement into variables that account for the trajectory travel time. To be able to find the right variables out of the infinite number of possibilities it is necessary to use traffic science and experience. In contrast to other studies the main focus of the development has been on the search for these variables instead of the type of the neural net. Finally a short-range travel time predictor has been developed and tested on real-world data, which has proven the chosen approach to be a promising one. With the algorithms used in modern traffic control centers in the Netherlands a reliability of about 85% for motorway traffic for a trajectory of about 15 km is achievable. With neural networks reliability up to 97% is feasible.
Samenvatting