Quasi-dynamic traffic assignment: static traffic assignment with queueing and spillback

Historical traffic patterns can be used for the prediction of traffic flows, as input for macroscopic traffic models, for the imputation of missing or erroneous data and as a basis for traffic management scenarios. This paper investigates the determination of historical traffic patterns by means of Ward’s hierarchical clustering procedure. Days were clustered before and after a pre-classification into working days and non-working days, using two different definitions of a daily traffic profile. The results of the clustering after pre- classification are clearly better than before classification. Moreover, working days are easier to classify into distinctive, recurrent traffic patterns than non-working days. Finally, clustering on the basis of 15 minutes traffic flows resulted in a better classification of working days than the two-step clustering that used the total daily traffic flow, peak flows, peak times and ratios. The clustering on the basis of 15 minutes traffic flows resulted in a classification into five clusters that show distinct daily flow profiles and are representative for the days within the clusters. The day of the week and vacation periods are determinative for the cluster a working day is classified to.