Nonintrusive Load Monitoring (NILM) Using a Deep Learning Model with a Transformer-Based Attention Mechanism and Temporal Pooling

Abstract

Nonintrusive load monitoring (NILM) is an important technique for energy management and conservation. In this paper, a deep learning model based on an attention mechanism, temporal pooling, residual connections, and transformers is proposed. This article presents a novel approach for NILM to accurately discern energy consumption patterns of individual household appliances. The proposed method entails a sequence of layers, including encoders, transformers, attention, temporal pooling, and residual connections, offering a comprehensive solution for NILM while effectively capturing appliance-specific energy usage in a household. The proposed model was evaluated using UK-DALE, REDD, and REFIT datasets in both seen and unseen cases. It shows that the proposed model in this paper performs better than other methods stated in other papers in terms of F1-score and total error of the results (in terms of SAE). This model achieved an F1-score equal to 92.96 as well as a total SAE equal to −0.036, which shows its effectiveness in accurately diagnosing and estimating the energy consumption of individual home appliances. The findings of this research show that the proposed model can be a tool for energy management in residential and commercial buildings.