Classification of historical mean speed patterns on a motorway for prediction of near-future traffic flow states.

This paper focuses on using historic traffic flow patterns for the prediction of near-future traffic states and travel time. Historic traffic flow patterns, especially mean speed fluctuation patterns, are one of the important factors for a prediction technique of near-future travel time of a long section on motorways. The authors use more than 1 year's travel flow fluctuation data collected by vehicle detectors on a long motorway section. In this study, some classification techniques were introduced to converge less dispersed mean speed patterns. It is pointed out that annual daily mean speed patterns clearly are different among motorway sections, and average coefficient of variations during congested flow conditions were about two to twenty times larger than free flow conditions. The results suggest that to increase the accuracy of travel time prediction on motorway sections, it is necessary to prepare many historic mean speed patterns.