Modeling and Classification of EV Charging Profiles Utilizing Topological Data Analysis

The increasing amount of electric vehicles (EVs) has brought environmental benefits yet challenges to the power grid distribution system. This paper proposes a novel modeling and classification for a distribution system in the local region based on a single-house total energy demand pattern to classify houses with or without EVs. To conduct the classification, multi-dimension time series data is constructed to better represent user patterns based on the single house demand profile based on the first difference and simple moving average (SMA). To preserve all features of the high dimension dataset, the Topolog-ical Data Analysis (TDA) is adopted. The persistence diagram (PD), which represents the characteristics of a single house's demand, is constructed by constructing topology space for each house's multi-dimension demand profile. The persistence image is adopted for the final classification task based on obtained PD. The classification model is obtained by applying logistic regression on the persistence images of houses with and without EVs. Numerical results are presented using the pecan street data.