Deep Learning-Based ON/OFF Detection for Enhanced Energy Disaggregation
Nidhal Balti, Baptiste Vrigneau, Pascal Scalart · 2024
Non-Intrusive Load Monitoring (NILM) attempts to break down the aggregate electrical consumption signal into the power consumption of each individual appliance, which is a classic example of blind source separation problems. This paper proposes a deep neural network that combines a classification subnetwork with a regression subnetwork for solving the NILM problem. In particular, we use a dual CNN-RNN architecture, in which the CNN is used to detect the appliances states and the RNN to predict the associated power value. Experiments conducted on a real-world dataset demonstrate that our model significantly outperforms state-of-the-art models while having good generalization capacity, achieving roughly 75% MAE gain and 60% RMSE gain to unseen appliances.