Energy disaggregation using variational autoencoders

Non-intrusive load monitoring (NILM) is a technique that uses single sensor to measure the total power consumption of building. Using an energy disaggregation method, individual appliances can be estimated from aggregate measurement. Recent algorithms have significantly improved performance NILM systems. However, generalization capability these methods different houses as well multi-state are still major challenges. In this paper we address issues and propose approach based on variational autoencoders framework. The probabilistic encoder makes efficient model for encoding information relevant reconstruction target appliance consumption. particular, proposed accurately generates more complex profiles, thus improving signal appliances. Moreover, its regularized latent space improves capabilities across houses. compared state-of-the-art approaches UK-DALE REFIT datasets, yields competitive results. mean absolute error reduces by 18% average all state-of-the-art. F1-Score increases than 11%, showing improvements detection in