Data-driven Representation of Heterogeneous Electric Vehicle Charging Behavior

Understanding electric vehicle (EV) charging behavior is important for power system planning and operation. However, EV charging behavior is highly heterogeneous, which makes direct modeling difficult and computationally expensive. In this paper, we propose a data-driven framework to construct a scalable representation of EV users’ charging behavior. We first map individual mobility-charging trajectories into a feature space that summarizes key behavioral characteristics. Then we extract behavioral archetypes by clustering mobility and charging features. Finally, we approximate the full population using a set of representative users, where the approximation error is quantified as the Wasserstein distance between the empirical distribution of the original population and that induced by the representative users. A case study in the San Francisco Bay Area shows that the proposed approach can approximate the population while preserving its statistical properties using only about 0.3% of users. This framework enables scalable modeling of EV charging behavior for downstream grid applications.