A model of speed profiles for urban road networks using G-means clustering

In this paper, we present a method for computing speed profiles by processing GPS data collected by vehicles in an urban area. The vehicles were tracked during a five year period on the road network of the capital city of Croatia (Zagreb). Traffic congestions in Croatia appear almost exclusively in urban areas, therefore Zagreb was chosen for this study as it is by far the largest city. The profiles for the roads were computed for each day of the week, where each day was segmented into five minute intervals. As there are no congestions during the night, the free flow speed for the roads was determined by averaging vehicle speeds which were recorded during that time. The speed profiles were clustered using G-means, a variant of the k-means clustering algorithm, to reduce storage space and to categorize roads observed in future research based on their profile. By applying this algorithm we reduced the number of total speed profiles by more than 90%. The profiles were developed as part of the SORDITO project, with the goal of developing algorithms for vehicle route optimizations.