Non-Intrusive Load Monitoring Using Machine Learning Accelerator Hardware for Smart Meters

Abstract

Residential electricity customers consume a significant amount of energy due to the extensive use of inefficient appliances. In order to save energy and reduce the electricity bills of the customers, load monitoring provides such customers with information to make informative cost-effective decisions. Non-Intrusive Load Monitoring (NILM) determines which appliances are on at the electrical input to a residence. Machine Learning (ML) based methods of NILM offer flexibility at the cost of computational complexity. This paper investigates addressing the problem of the computational bottleneck using a novel ML-based acceleration hardware. In this work, ML is used to develop a NILM algorithm, which is then tested on a publicly available dataset named the Reference Energy Disaggregation Dataset (REDD). Subsequently, a physical system modelling an end-to-end smart metering solution is designed and tested. The results show a significant decrease in the time and energy required to run the ML algorithms and most importantly, the successful real-time operation of NILM algorithms embedded in a smart meter. By using the newly developed ML acceleration hardware and ML-based algorithms, NILM can be embedded into next-generation Smart Meters.