Non-intrusive load classification for energy management of electrical appliances using convolutional long-short term memory

Abstract

Non-intrusive load classification (NILC) is a crucial technique in energy management, helping to reduce unnecessary energy consumption and support the development of smart buildings. However, accurately classifying devices with similar characteristics and handling the complexity of electrical signals remain significant challenges. This research presents a deep learning-based NILC designed to efficiently extract key features from energy data. The convolutional LSTM combines a residual block (RB) and squeeze-and-excitation (SE) layers within a convolutional neural network (CNN) to enhance feature extraction while minimizing information loss. It consists of three main convolutional blocks, each incorporating SE layers to improve feature attention, along with long-short-term memory (LSTM) to capture sequential dependencies, leading to improved classification accuracy. The proposed model is trained on datasets containing 2, 3, and 4-electrical appliance operation scenarios, with feature data transformed into kurtograms to enhance signal characteristics. The training results achieved peak accuracy scores of 98.08%, 99.96%, and 99.75% and precision scores of 99.96%, 95.65%, and 97.10% for the respective scenarios. These results highlight the effectiveness of NILM in optimizing household energy usage, marking a significant step toward developing advanced technologies that reduce energy costs, promote sustainable energy consumption, and enhance energy management in future smart homes.