Clustering of Usage Profiles for Electric Vehicle Behaviour Analysis

Abstract

Accurately predicting the behaviour of electric vehicles is going to be imperative for network operators. In order for vehicles to participate in either smart charging schemes or providing grid services, their availability and charge requirements must be forecasted. Their relative novelty means that data concerning electric vehicles is scarce and biased, however we have been collecting data on conventional vehicles for many years. This paper uses cluster analysis of travel survey data from the UK to identify typical conventional vehicle usage profiles. To this end, we determine the feature vector, introduce an appropriate distance metric, and choose a number of clusters. Five clusters are identified, and their suitability for electrification is discussed. A smaller data set of electric vehicles is then used to compare the current electric fleet behaviour with the conventional one.