Disaggregation of EV Charging Profiles via Spatio-Temporal Graph Convolutional Networks

The proliferation of electric vehicles (EVs) accelerates the pace of transportation electrification. Meanwhile, the unprecedented nature of EV charging events within community brings challenges to the demand response in power systems. To support smart EV charging strategies aimed at enhancing system reliability, it is essential to identify charging patterns, integrate advanced load management techniques, and ensure seamless communication between community and the power grid. To disaggregate community level EV charging profile which is needed for better grid management, this paper proposes a novel bi-level Spatio-Temporal Graph Convolutional Network (ST-GCN) based framework for non-intrusive EV charging profile disaggregation at community level. Within the proposed framework, first level identify the charging event of each single household by utilizing inter-household dependencies. Second level disaggregate the EV charging profile by reconstructing the load profile with STGCN architecture. The proposed disaggregation framework is validated with real-world dataset from Pecan Street.