Hierarchical energy profile characterization of electric vehicle charging stations integrated with photovoltaic systems based on clustering techniques

Abstract

Secondary and primary substations of networks with electric vehicle (EV) chargers and photovoltaics (PVs) experience net loads characterized by uncertainty. Accurate characterization of EV, PV and net load energy profiles is necessary to plan new installations and to develop forecasting methodologies. This paper provides a novel contribution to the energy profile characterization of EVs and PVs, exploiting clustering techniques in a hierarchical framework to eventually characterize the overall net load profiles. In the proposal, the lower levels of the hierarchy identify clusters of EV load and PV generation profiles at individual installations, alternatively using one clustering technique among DBSCAN, Gaussian mixture models (GMMs), K-means algorithm (KMA), and spectral clustering (SC). The intermediate levels of the hierarchy reconstruct the overall EV load and PV generation profiles through a proposed frequentist combination of the lower-level profiles. The upper level of the hierarchy characterizes the overall net load through a novel approach based on the quantile convolution of the intermediate-level EV and PV profiles. Real EV load and PV generation data are used to evaluate the performance of the presented hierarchical methodology, with relative fitting improvements between 1% and 8% (compared to a two-level hierarchical benchmark) and between 16% and 29% (compared to a direct, non-hierarchical benchmark).