USING SYNTHETIC DATA TO EVALUATE MULTIPLE-REGRESSION AND PRINCIPAL COMPONENT ANALYSES FOR STATISTICAL MODELING OF DAILY BUILDING ENERGY-CONSUMPTION

Multiple regression modeling of monitored building energy use data is often faulted as a reliable means of predicting energy use on the grou nds that multicollinearity between the regressor variables can lead bo th to improper interpretation of the relative importance of the variou s physical regressor parameters and to a model with unstable regressor coefficients. Principal component analysis (PCA) has the potential to overcome such drawbacks. While a few case studies have already attemp ted to apply this technique to building energy data, the objectives of this study were to make a broader evaluation of PCA and multiple regr ession analysis (MRA) and to establish guidelines under which one appr oach is preferable to the other. Four geographic locations in the US w ith different climatic conditions were selected and synthetic data seq uences representative of daily energy use in large institutional build ings were generated in each location using a linear model with outdoor temperature, outdoor specific humidity and solar radiation as the thr ee regression variables. MRA and PCA approaches were then applied to t hese data sets and their relative performances were compared. Conditio ns under which PCA seems to perform better than MRA were identified an d preliminary recommendations on the use of either modeling approach f ormulated.