Statistical modeling of the building energy balance variable for screening of metered energy use in large commercial buildings

Abstract

Abstract Whole building energy data are the fundamental metric of building energy performance and used for evaluation of investments in energy efficiency, utility billing, and calibration of building energy simulations. Metered energy use data contain anomalies and bias, and those must be flagged and investigated to reach useful conclusions from the data. The energy balance load E BL representing the aggregate building thermal load is a variable calculated from the metered energy use, and the constancy of E BL as a function of influential variables such as outside air temperature can be used to check the validity of metered whole building energy use. This paper proposes regression models to statistically identify the building-specific daily E BL pattern without prior knowledge of building systems and operations. The proposed models are designed based on simplified load calculation principles, so that the regression parameters have physical significance. The models were applied to the E BL data for 56 buildings on the Texas A&M University campus to examine the applicability. The mean CV-RMSEs of the four-parameter change-point (4P-CP) models, multiple linear regression (MLR) models, and the MLR models incorporating AR(1) error structure ranged from 6.9% to 10.4%. Overall, the MLR models with the outside air temperature and humidity load variables presented the most balanced performance taking account of the availability of variables and the ease of estimation and parameter interpretation.