Increasing car mobility has lead to an increasing demand for traffic information. This contribution deals with information about travel times. When car drivers are provided with this type of information, the travel times should ideally be the times that they will encounter. As a result travel times must first be measured, and further be predicted on a short-term basis. Since dual induction loop detectors yield spot measurements of flow and speed, travel times produced by data from these sensors can only be estimated, not actually measured. The performance of five algorithms to estimate travel times was assessed using a data set with actually measured travel times. These were collected through license plate recognition. Subsequently two estimation methods that are currently being used as if they produced predictions, i.e. Static Travel Time Estimations (STTE) and Dynamic Travel Time Estimations (DTTE), and a new travel time prediction method using an Artificial Neural Network (ANN) were applied on the A13 motorway from The Hague to Rotterdam and their performance were compared. Results show that the ANN method significantly outperformed DTTE, which in turn outperformed STTE.
Abstract