Edinburgh Napier University

  Given that residential sectors in both developed and developing nations contribute to a significant portion of electric energy consumption, addressing energy efficiency and conservation in this sector is envisioned to have a considerable effect on the levels of nationwide and global electric energy consumption. Various approaches have been utilized to address these challenges with a number of positive outcomes being realized through Load Monitoring and Non-Intrusive Load Monitoring (NILM) in particular. These positive outcomes have been attributed to the increase in energy awareness of homeowners. Due to limited resources in a residential environment, methods utilizing unsupervised learning together with NILM can provide valuable and practical solutions. Such solutions are of great importance to developing nations and low-income households as they lower the barrier for adoption by reducing the costs and effort required to monitor electric energy usage. In this paper we present a low-complexity unsupervised NILM algorithm which has practical applications for monitoring electric energy usage within homes. We make use of Entropy Index Constraints Competitive Agglomeration (EICCA) to automatically discover an optimal set of feature clusters, and invariant Active Power (P) features to detect appliance usage given aggregated household energy data which includes noise. We further present an approach that can be used to obtain Type II appliance models, which can provide valuable feedback to homeowners. The results of experimental validation indicate that our proposed work has comparable performance with recent work in unsupervised NILM including the state of the art with regards to energy disaggregation.