Attention-Based Multitask Probabilistic Network for Nonintrusive Appliance Load Monitoring

Monitoring individual appliances’ operating state and energy consumption in a building enables significant energy-saving opportunities. These days, smart meters perform this task nonintrusively using sophisticated signal processing, machine-learning, and/or deep-learning (DL) approaches. To this end, this article proposes a novel multitask DL model that uses readily available low-frequency energy data from the smart meter for simultaneous appliance state detection (SD) and energy disaggregation (ED). The model creatively adopts and customizes the famous transformer model from the field of language modeling for the above task. Furthermore, the model output is produced as a mixture of probability density functions to handle uncertainties. The model performance is evaluated using the publicly available REFIT and UKDALE datasets. The test results indicate the proposed model’s superiority, generalizability, and transferability compared to other state-of-the-art models.