Minimally intrusive electrical load monitoring using distributed sensing

Lead University: Carnegie Mellon University
PI: Mario Berges, Civil and Environmental Engineering
Co-PIs: Anthony Rowe, Electrical and Computer Engineering

This project plans to evaluate the performance of a novel distributed electricity disaggregation algorithm developed at CMU that is especially well-suited to estimate the electricity consumption of individual loads in facilities that have a similar load composition. The current version of the algorithm, which has been described in an academic publication and a provisional US patent, achieves unprecedented disaggregation performance on the publicly available benchmark dataset BLUED. In short, the algorithm is a deep neural network that is designed to decompose the aggregate current measurements (corresponding to one full voltage cycle) into reoccurring and additive components (also corresponding to full cycles).

This neural network can be trained in an unsupervised fashion to learn the components and, given sufficiently detailed measurements of current and voltage, and sufficiently large volumes of data, the resulting components readily correspond to individual appliances. Thus, the algorithm's performance is directly tied to the quality and volume of data that it is trained on.

One of the innovations being pursued in this project is to leverage previous advances by the Co-PIs in the area of large-scale distributed sensing (as a result of previous PITA funding on a project called Sensor Andrew) to facilitate the collection of the data required for the increased performance of the disaggregation algorithm. In essence, instead of attempting to disaggregate one single building at a time, the idea would be to allow the network to learn the components from the measurements collected at many buildings that share a similar load make-up.

For this project we will first develop a robust hardware prototype of the technology that can be deployed at Giant Eagle GetGo locations and stream data to a remote server. We will then deploy, test and refine the technology with the help of our partner, Giant Eagle.