Feature-based identification of distributed energy resources in buildings from aggregated smart meter data

Distributed energy resources (DERs), including electric vehicles, heat pumps and rooftop solar photovolatic systems, are becoming increasingly common-place in Europe. On the one hand, these developments drastically alter the net load, making it imperative for distribution system operators to be cognizant of their proliferation. On the other, they can provide cost-effective energy flexibility to mitigate the worst side effects of distributed generation and demand electrification. However, most of these new installations are not recorded in a centralized database, making planning and operation tricky. The simultaneous proliferation of smart meters is also inadequate to track this in and of itself since there is little visibility on behind the meter loads. In this paper we utilize a detailed corpus of data from 1,300 residential buildings in Flanders, Belgium to create strategies which can automatically detect different DERs in these settings. Using two interpretable feature extraction methodologies (domain informed and domain agnostic), we show strong predictive performance (macro-F1 score exceeding 0.85) in accurately categorizing buildings based on the DERs present behind the meter, without having this information up front. Such a strategy can help distribution grid operators (DSOs) better identify the proliferation of DERs in their control region without the investment and privacy issues related to sub-metering.