Exploratory Analysis and Load Forecasting with High Resolution Building Energy Data

DOI: 10.5176/2301-394X_ACE14.102

Authors: Dr. W. Tian, Dr. R. Choudhary and Jean Specklin

Abstract:

Hourly or half-hourly energy data for buildings are becoming increasingly available. This paper explores techniques to make use of high-resolution metered energy data from buildings for the purposes of exploratory analysis and load forecasting. Exploratory analysis aims for interpreting the energy usage of the building, and five methods can be used including scatter plots, box plots, cluster analysis, seasonal decomposition, and spectral analysis. Scatter and box plots are two simple methods, but can provide valuable information by focusing on multivariable visualization. Cluster analysis can group original energy data into several clusters to show the similar characteristics of energy consumption in every cluster. Decomposition and spectral analysis can detect the trend and the hidden periodicity of building energy data, respectively. For load forecasting, the combination of multiple linear regression and cluster analysis can be used to create a simple and accurate model in order to detect anomalies, evaluate load shifting, and/or estimate energy savings.

Keywords: Energy Data, Building Energy Forecasting, Energy Metering