ESTIMATING ROUTE TRAVEL TIME VARIABILITY FROM LINK DATA BY MEANS OF CLUSTERING

Accurate route travel time estimation is today one of the most challenging problems in traffic theory. This research proposes a novel method for the estimation of route travel time distributions, based on historical link travel time observations. Central in the development of this framework is the distinction between (cheap) off-line storage and computations and (expensive) on-line computations. For that it is important to minimize the on-line computational effort of calculating a route travel time histogram. The key elements in the method are correlations in link travel time fluctuations and a clustering algorithm. Tests on the Belgian road network show that 1) the clustering method is on-line computationally efficient while keeping the off-line storage under control, 2) that it can accurately estimate route travel time distributions and 3) that it can be applied successfully for route reliability estimation.