Disaggregation of Superposed Time-Series: Device Detection in Aggregated Load Profiles

Abstract

With increasing complexity in modern power grids, behind-the-meter load disaggregation has become essential for accurate optimization and prediction in energy communities. This paper proposes a deterministic approach for time-series disaggregation, based on image processing techniques. Therefore, large annotated training datasets or high-frequency measurements are not required, thus significantly reducing manual data labeling efforts. The method is evaluated using aggregated smart-meter data, where it successfully identifies fridge and gas-boiler events and reconstructs their corresponding device-level load profiles.