Machine Learning-Based Non-Intrusive Load Monitoring for Three-Phase Smart Energy Consumption Disaggregation

Abstract

Non-Intrusive Load Monitoring (NILM) has emerged as a pivotal technology for promoting energy efficiency and sustainability by disaggregating aggregate household or industrial energy consumption into appliance-level usage profiles. This capability enables consumers and utilities to gain deeper insights into consumption behavior, reduce wastage, and improve energy management practices. In this study, a machine learning-based NILM framework is presented, focusing on enhancing accuracy, adaptability, and scalability across three-phase energy systems, which are common in residential and industrial applications. The proposed approach employs the Cubic K-Nearest Neighbor (Cubic KNN) algorithm for effective classification and disaggregation of energy data. Unlike traditional NILM methods, the Cubic KNN model demonstrates robustness in handling challenges such as low sampling rates, overlapping appliance signatures, multi-state appliances, and dynamic operational profiles. The methodology integrates feature extraction, supervised machine learning, and rigorous validation against realworld energy data obtained from a Tenaga Nasional Berhad (TNB) household smart meter. Both online and offline energy monitoring and data logging systems were developed to ensure data reliability and continuity. Experimental evaluations show that the Cubic KNN classifier achieves high accuracy, exceeding 95% for most appliances, with negligible misclassification errors. Furthermore, comparative analysis against utility billing data confirms the reliability of the proposed system, with overall error rates below 6%. The findings underscore the potential of advanced machine learning algorithms in NILM applications, contributing toward the development of scalable, efficient, and accurate solutions. This research highlights the role of NILM in supporting smarter energy monitoring, conservation initiatives, and sustainable energy management strategies for future smart grid systems.