On performance evaluation and machine learning approaches in non-intrusive load monitoring

Abstract

Non-Intrusive Load Monitoring (NILM) is a set of techniques to gain deep insights into workflows inside buildings based on data provided by smart meters. In this way, the combined consumption needs only to be monitored at a single, central point in the household, providing advantages such as reduced costs for metering equipment. Over the years, a plethora of load monitoring algorithms has been proposed comprising approaches based on Hidden Markov Models (HMM), algorithms based on combinatorial optimisation, and more recently, approaches based on machine learning. However, reproducibility, comparability, and performance evaluation remain open research issues since there is no standardised way researchers evaluate their approaches and report performance. In this paper, the author points out open research issues of performance evaluation in NILM, presents a short survey of deep learning approaches for NILM, and formulates research questions related to open issues in NILM. An outline of future work is given including applied methodology and expected findings.