Deriving traffic demand patterns from historical data

Abstract

The development and decreased cost of technology and communications have brought about a huge increase in the availability of traffic data. With every passing day, traffic management centers must deal with an increased amount of detailed data. Once the real time use of these data is complete, they must be stored for long periods of time. In this long term context, the vast amount of raw data is meaningless, which is a clear example of data asphyxiation. Traffic management centers must aggregate and synthesize the data in order to extract the maximum knowledge from them. Pattern classification is a way to deal with this issue. Traditionally, traffic demand patterns have been easily constructed using ad hoc methods, where “experience” is their main attribute. These procedures lack the required rigor to support current needs in terms of planning and operational management. The present paper proposes a method to systematically derive traffic demand patterns from historical data. The method is based on the cluster analysis technique, and allows the inclusion of preexistent knowledge, which eases the interpretation and practical use of the results. The proposed pattern classification procedure is applied to five years of hourly traffic volumes on a Spanish highway. The obtained results prove the validity and utility of the method to accurately summarize the seasonal and daily characteristics of traffic demand.