Assessment of Non-Intrusive Load Monitoring as a Blind Source Separation Problem

Non-Intrusive Load Monitoring (NILM) allows consumers to monitor appliances' power consumption without installing appliance-level sensors. NILM has become popular with the rapid deployment of smart energy meters. Neural network-based disaggregation approaches have provided more promising results in NILM. However, most of them must have labeled energy data for model training. The availability of energy data is inadequate, and the data must be updated frequently for higher disaggregation performances. However, none of the existing datasets is periodically updated. Therefore, training data has become a limiting factor for commercializing disaggregation solutions. This study addresses the issue by proposing an Independent Component Analysis (ICA) based solution to perform energy disaggregation as a blind source separation problem for the first time in the NILM domain. The aggregate energy signal has been used to prepare the independent input signals with a novel signal preprocessing approach. The results revealed the effectiveness of ICA in NILM applications.