Development of hybrid CNN-LSTM for non-intrusive load monitoring

Abstract

 Non-Intrusive Load Monitoring (NILM) is a technique used to distinguish the energy consumption of individual electrical devices from aggregated energy consumption data without requiring additional sensors on each device. This technology plays a crucial role in efficient energy management, reducing energy costs, and supporting the development of smart buildings. This research focuses on developing a hybrid deep learning network to enhance NILM efficiency by combining convolutional neural networks with long short-term memory networks. This combination enables the analysis of complex electrical power signals, improving the accuracy of device classification, reducing prediction errors, and enhancing learning efficiency from diverse data. The proposed method is trained using electrical appliance interaction data in three configurations: 2-appliance, 3-appliance, and 4-appliance interactions. Experimental results demonstrate training accuracies of 98.59%, 98.59%, and 93.09%, respectively, while the highest testing accuracies are 98.59%, 95.61%, and 92.94%. These results highlight the potential for further advancements in NILM technology, enabling more efficient energy monitoring systems and promoting sustainable energy use in the future.