Nonintrusive Load Monitoring Algorithm Using SOM-AdaDBSCAN

In this paper, a nonintrusive load monitoring algorithm based on the SOM-AdaDBSCAN algorithm is investigated to address the problems that traditional event based non-intrusive load decomposition cannot accurately estimate load fluctuations, and the state-based algorithm is difficult to balance accuracy and generalization ability, as well as the parameters are empirically selected. First, an event detection algorithm based on edge sharpening and singular value decomposition is proposed by merging continuously changing load edges to address the phe-nomenon of event omission and false detection when the load event lasts for a long time. Second, Self-Organizing Map (SOM) clustering algorithm is proposed for the problem that traditional DBSCAN is difficult to handle multiple density clusters and slow in computation for first-step clustering. Then, the secondary clustering of load events is done by using AdaDBSCAN. Finally, we compare the SOM-AdaDBSCAN algorithm with K-means and DBSCAN and other algorithms to verify the effectiveness of SOM-AdaDBSCAN and improve the accuracy and speed of nonintrusive load monitoring.