A Machine Learning Approach for NILM based on Superimposed Current Profiles

This research focuses on identifying a new implementation of a machine learning approach for Non-intrusive load monitoring (NILM). We mathematically superimpose current profiles of individual appliances and compare against the actual combinational current profiles. This simple yet effective method is tested on combinations of 6 household devices in a typical low voltage residential installation and the high accuracy of correct identification confirms the proposed method is feasible. The proposed method eases the burden of the training phase which is considered as an inherent limitation of all supervised deep learning NILM models. We deploy the method on a Raspberry Pi 3 providing a solution to increase the scalability of NILM.