A New Approach for Supervised Power Disaggregation by Using a Denoising Autoencoder and Recurrent LSTM Network

Abstract

Non-intrusive load monitoring (NILM) is a task of estimating the contribution of individual appliance to the overall power consumption by using a set of electrical signals measured by a smart meter. In this paper, we propose a comprehensive and extensible framework based on DNNs. We employ denoising autoencoder (dAE) to reconstruct the power signal of individual appliance from aggregated power consumption, and we use long short term memory (LSTM) network to make sure which appliance the power signal belongs to. We select 5 appliances to validate our method, and the results have shown the advantages of the proposed framework in some aspects compared to hidden Markov models (HMMs) and premier dAE.