Data-driven electrical load disaggregation using graph signal processing

Graph Signal Processing is a very active area of research and is being increasingly used for different applications in different domains like vision, sensor networks etc. It deals with processing signals defined on the vertices of a graph which captures the relational dependencies of the signal. Recently, graph signal processing has been applied to source separation problem, in particular, electrical load disaggregation where, given smart meter measurements it is required to ascertain the contribution of multiple loads which could have resulted in those measurements. Inspired by these works, an attempt is made to adapt the results for real-life situations. Apart from incorporating pre-processing and post-processing stages, the disaggregation algorithm is appropriately modified to arrive at useful results for both load identification and consumption estimation. Both are important as disaggregation results, since the consumption information facilitates load prognostics/diagnostics and helps in giving advisory feedback to the consumers. Apart from outlining the algorithms, which work for different sampling rates like 10 seconds and 1 minute, some of the promising results obtained during the extensive simulation studies are presented in the paper.