This paper introduces two forecasting methods for building energy consumption data that are recorded from smart meters in high resolution. For utility companies, it is important to reliably forecast the aggregate consumption profile to determine energy supply for the next day and prevent any crisis. The proposed methods involve forecasting individual load on the basis of their measurement history and weather data without using complicated models of building system. The first method is most efficient for a very short-term prediction, such as the prediction period of one hour, and uses a simple adaptive time-series model. For a longer-term prediction, a nonparametric Gaussian process has been applied to forecast the load profiles and their uncertainty bounds to predict a day-ahead. These methods are computationally simple and adaptive and thus suitable for analyzing a large set of data whose pattern changes over the time. These forecasting methods are applied to several sets of building energy consumption data for lighting and heating-ventilation-air-conditioning (HVAC) systems collected from a campus building at Stanford University. The measurements are collected every minute, and corresponding weather data are provided hourly. The results show that the proposed algorithms can predict those energy consumption data with high accuracy.

Date Published: 19 April 2013
PDF: 8 pages
Proc. SPIE 8692, Sensors and Smart Structures Technologies for Civil, Mechanical, and Aerospace Systems 2013, 86920T (19 April 2013); doi: 10.1117/12.2009894

Published in SPIE Proceedings Vol. 8692:
Sensors and Smart Structures Technologies for Civil, Mechanical, and Aerospace Systems 2013
Jerome Peter Lynch; Chung-Bang Yun; Kon-Well Wang, Editor(s)