Model identification and prediction uncertainty of linear building energy use models with autocorrelated residuals

Autocorrelated residuals from regression models of building energy use present problems when attempting to estimate retrofit energy savings and the uncertainty of the savings. This paper discusses the causes of autocorrelation in energy use models and proposes methods to deal with autocorrelation. A hybrid of ordinary least squares (OLS) and autoregressive (AR) models is developed to accurately predict energy use and give realistic uncertainty estimates. A procedure for model selection is presented and tested on data from three commercial buildings participating in the Texas LoanSTAR program. In every case examined, the hybrid OLS-AR model provided an uncertainty estimate for energy use far more accurate than the OLS estimate.