Event and feature based electrical load disaggregation using graph signal processing

Electrical load disaggregation continues to attract new explorations due to its challenging nature as well as utility. When the loads to be separated are characterized by suitable features, there is a possibility to solve the problem by utilizing the techniques from the emerging area of Graph Signal Processing (GSP). In this paper, we propose a three-staged approach comprising of (i) Event Detection and Clustering (ii) Event Pairing and Feature Extraction and (iii) Load Classification, each of them being pivoted on GSP. For load classification in particular, a robust spectral clustering strategy is appropriately adopted using joint spectrum computed from different features. The efficacy of this novel combination is demonstrated through the results obtained on both public data sets and the simulated active power signals.