Unsupervised Energy Disaggregation of Home Appliances

Energy management is a growing concern especially with the increasing growth of smart appliances within the home. Energy disaggregation is an ongoing challenge to discover the appliance usage by examining the energy output of a household or building. Unsupervised NILM presents the additional challenge of energy disaggregation without any reliance on training data. A key issue to address in Unsupervised NILM is the discovery of appliances without a priori information. In this paper we present a new approach based on Competitive Agglomeration (CA) which incorporates the good qualities of both hierarchical and partitional clustering. Our proposed energy disaggregation algorithm makes use of CA in order to discover appliances without prior information about the number of appliances. Validation with experimental data from the Reference Energy Disaggregation Dataset (REDD), and comparison with recent state of the art Unsupervised NILM indicates that our proposed algorithm is effective.