Predictions of Urban Volumes in Single Time Series

Congestion is increasing in many urban areas. This has led to a growing awareness of the importance of accurate traffic-flow predictions. In this paper, we introduce a prediction scheme that is based on an extensive study of volume patterns that were collected at about 20 urban intersections in the city of Almelo, The Netherlands. The scheme can be used for both short- and long-term predictions. It consists of 1) baseline predictions for a given preselected day, 2) predictions for the next 24 h, and 3) short- term predictions with horizons smaller than 80 min. We show that the predictions significantly improve when we adopt some straightforward assumptions about the correlations between and the noise levels within volumes. We conclude that 24-h predictions are much more accurate than baseline predictions and that errors in short- term predictions are even negligibly small during working days. We used a heuristic approach to optimize the model. As a consequence, our model is quite simple so that it can easily be used for practical applications.