Deep Learning Based Non-Intrusive Load Monitoring for a Three-Phase System

Abstract

Non-Intrusive Load Monitoring (NILM) is a method to determine the power consumption of individual appliances from the overall power consumption measured by a single measurement device, which is usually the main meter. Increase in the adoption of smart meters has facilitated large scale implementation of NILM, which can provide information about individual loads to the utilities and consumers. This will lead to significant energy savings as well as better demand-side management. Researchers have proposed several methods and have successfully implemented NILM for residential sectors that have a single-phase supply. However, NILM has not been successfully implemented for industrial and commercial buildings that have a three-phase supply, due to several challenges. These buildings consume significant amount of power and implementing NILM to these buildings has the potential to yield substantial benefits. In this paper, we propose a novel deep learning-based approach to address some of the key challenges in implementing NILM for buildings that have a three-phase supply. Our approach introduces an ensemble learning technique that does not require training of multiple neural network models, which reduces the computational requirements and makes it economically feasible. The model was tested on a three-phase system that consists of both three- phase loads and single-phase loads. The results show significant improvement in load disaggregation compared to the existing methods and indicate its applicability.