A two-step process for clustering electric vehicle trajectories

Abstract

The aim of this work is the identification of typical patterns in urban mobility on the basis of real data and advanced data mining algorithms. To achieve this goal a trajectory pattern recognition system has been developed. This system encompasses two steps: the first one is a fast K-means clustering to group the trajectories according to their start and destination coordinates and the second one is the classification over a Self-Organizing Map of the trajectories grouped before. Previously to this second step, the system standardizes the trajectories to equal length criterion using Dynamic Time Warping. This work also includes the results of testing the system on a real database of fifty trajectories of trucks.