Blind non-intrusive appliance load monitoring using graph-based signal processing

With ongoing massive smart energy metering deployments, disaggregation of household's total energy consumption down to individual appliances using purely software tools, aka. non-intrusive appliance load monitoring (NALM), has generated increased interest. However, despite the fact that NALM was proposed over 30 years ago, there are still many open challenges. Indeed, the majority of approaches require training and are sensitive to appliance changes requiring regular re-training. In this paper, we tackle this challenge by proposing a "blind" NALM approach that does not require any training. The main idea is to build upon an emerging field of graph-based signal processing to perform adaptive thresholding, signal clustering and feature matching. Using two datasets of active power measurements with 1min and 8sec resolution, we demonstrate the effectiveness of the proposed method using a state-of-the-art NALM approaches as benchmarks.