Representative Profiling of Prosumers with Local Distributed Energy Resources and Electric Vehicles Using Unsupervised Machine Learning

In this paper, the representative profiles of residential prosumers owning local distributed energy resources (L-DERs) and plug-in electric vehicles (PEV) at different levels of generation and demand are identified. The Pecan Street household dataset is used in this work with high-granularity data of one second and it includes the roof-top solar photovoltaic (PV), home battery energy storage system (HBESS) and PEV profiles. Because of the large variance in the data due to such high granularity and different levels of generation/demand in the residential profiles, this study presents a systematic approach to identify a comprehensive list of representative profiles. The machine learning techniques such as principal component analysis (PCA), and the unsupervised K-means clustering and K-nearest neighbor are employed to identify the representative profiles. The study included 123 residential homes, which includes a set of different combinations of PVs, PEVs and HBESS and the results have shown that they can be represented by only 17 representative profiles. This reduction in the number of representative profiles at such high-granularity will lead to significant advances in accelerating the distribution system time-series analysis studies in particular when considering the presence of prosumers with L-DERs and PEVs.