Electricity Disaggregation

Smart meters have drastically increased the temporal resolution of electrical load measurements available to both customers and utilities. Past research has shown promising results towards utilizing this information to break down measurements into the constituent loads within each customer’s facility without requiring the use of additional metering hardware. In this project we focus on a specific class of unsupervised algorithms based on deep learning techniques that can learn instantaneous power waveforms for individual devices as well as their activation patterns given sufficient data from a single meter.