Contribuitions and developments on nonintrusive load monitoring

Abstract

Energy efficiency is a key subject in our present world agenda, not only because of greenhouse gas emissions, which contribute to global warming, but also because of possible supply interruptions. In Brazil, energy wastage in the residential market is estimated to be around 15%. Previous studies have indicated that the most savings were achieved with specific appliance, electricity consumption feedback, which caused behavioral changes and encouraged consumers to pursue energy conservation. Nonintrusive Load Monitoring (NILM) is a relatively new term. It aims to disaggregate global consumption at an appliance level, using only a single point of measurement. Various methods have been suggested to infer when appliances are turned on and off, using the analysis of current and voltage aggregated waveforms. Within this context, we aim to provide a methodology for NILM to determine which sets of electrical features and feature extraction rates, obtained from aggregated household data, are essential to preserve equivalent levels of accuracy; thus reducing the amount of data that needs to be transferred to, and stored on, cloud servers. As an addendum to this thesis, a Brazilian appliance dataset, sampled from real appliances, was developed for future NILM developments and research. Beyond that, a low-cost NILM smart meter was developed to encourage consumers to change their habits to more sustainable methods.