Traffic Prediction in Metropolitan Freeways

The successful wide scale deployment of Advanced Traveler Information Systems (ATIS) and Advanced Traffic Management Systems (ATMS) relies significantly on the capability to perform accurate short-term predictions of traffic parameters over the entire road network. Most of previous studies and applications of traffic prediction have developed statistical or heuristic methods based on historical or realtime traffic data collected by fixed sensor networks. Meanwhile the effectiveness and wide applicability of such methods is strongly affected by the high cost of deploying and maintaining fixed sensor networks able to provide large spatial coverage. Recently, mobile sensors or probe vehicles appeared as a complementary solution to fixed sensors in order to increase coverage areas and prediction accuracy without requiring expensive infrastructure investments. The focus of this paper is to present findings of a study on short-term traffic prediction based on data collected by a probe vehicle system operating on the Italian national scale and consisting of a large fleet of privately-owned vehicles. Particularly the paper proposes and compares two models, respectively based on Artificial Neural Networks and Pattern-Matching techniques, designed to on-line perform short-term (15 to 60 minutes) predictions of link travel speeds. The Rome Ring Road (GRA - Grande Raccordo Anulare) is used to assess the feasibility of the two models. Testing results showed that the proposed approaches for short-term predictions are very promising and effective.