An automatic learning framework for smart residential communities

Author(s): Helia Zandia, Michael Starkea, Jeffrey Munka, Teja Kurugantia, James Leveretteb, Jens Gregorc
a Oak Ridge National Laboratory, Oak Ridge, TN, USA
b Southern Company, Birmingham, AL, USA
c University of Tennessee, Knoxville, Knoxville, TN, USA
International Journal of Smart Grid and Clean Energy, vol. 9, no. 3, May 2020: pp. 485-494
ISSN: 2315-4462 (Print)
ISSN: 2373-3594 (Online)
Digital Object Identifier: 10.12720/sgce.9.3.485-494

Abstract: Predictability has been foundational to matching supply and demand in the day-to-day operation of the electric power system. Demand predictability is eroding because of the increased use of renewable energy resources and more sophisticated loads, such as electric vehicles and smart appliances. In this paper, an automatic software framework is described which can be used for load forecasting in smart communities. A time-varying clustering-based Markov chain approach is used to predict the energy consumption of residential buildings in a smart community. The training data is based on 1-minute meter data of occupied homes over one month. The data points are first clustered based on the energy consumption and the time of the day. Then, the original data is converted using the Centroids of the clusters. A time-varying Markov chain is subsequently trained to model the energy consumption behavior of residents for each home using the transformed data. The trained model is shown to successfully predict load in 5-minute intervals over a 24 hours period.

Keywords: Smart grid, internet of things, cyber-physical system, Markov chain, load forecast, demand response, clustering
Full Paper.pdf
 

Copyright © 2020 International Journal of Smart Grid and Clean Energy, All Rights Reserved