<![CDATA[Accurate monitoring of appliance-level energy consumption plays a pivotal role in advancing smart grid operations and residential energy usage optimization. Non-Intrusive Load Monitoring (NILM) offers a non-invasive means to infer individual device usage from aggregated household electricity measurements, eliminating the need for dedicated sensors on each appliance. This study implements Gradient Boosting, specifically LightGBM, for multi-label appliance classification within NILM systems utilizing the public ECO dataset from a selected residential unit. Five essential household appliances: TV, lamp, kettle, fridge, and freezer were chosen, with three months of data used for training and one month for testing to ensure temporal consistency and generalization. Feature extraction was performed on 60-second windows of aggregated smart meter data, capturing statistical characteristics to enhance model learning. The proposed method demonstrated robust accuracy in appliance classification, with results of 93.66% for the fridge, 92.63% for the freezer, 99.60% for the kettle, 99.37% for the lamp, and 96.21% for the TV, demonstrating the effectiveness of Gradient Boosting for multi-label appliance detection within NILM systems using real-world data. This implementation contributes to the development of scalable, accurate NILM frameworks suitable for integration within smart grid and energy management applications.]]>
Copyrights © 2025
