TRANSFORMERS FOR NON-INTRUSIVE LOAD MONITORING PROBLEM

Abstract

With the current environmental issues, improving electrical management for households becomes more and more important. One solution consists in moni-toring the electrical consumption of the different appliances present in an house or apartment, without using direct sensors on appliances. This problem is called Non-Intrusive Load Monitoring (NILM). By having access to the way appliances are used and how they consume energy, it is possible to provide to the users information that can help improve the management of electrical appliances and thus, save energy. In this paper, we focus on Deep Learning methods, and show that simple structure of Transformers with an adapted loss function outperform CNN for the NILM problem.