Statistical Modeling of Daily Energy Consumption in Commercial Buildings Using Multiple Regression and Principal Component Analysis

Abstract

Statistical models of energy use in commercial buildings are being increasingly used not only for predicting retrofit savings but also for identifying improper operation of HVAC systems. The conventional approach involves using multiple regression analysis to identify these models. However, such models tend to suffer from physically unreasonable regression coefficients and instability due to the fact that the predictor variables (i.e., climatic parameters, building internal loads, etc.) are intercorrelated. A relatively new approach proposed to circumvent these drawbacks is principal component analysis. The objective of this paper is to evaluate the multivariate regression and the principal component analysis approaches, using measured whole-building energy use data from a large commercial building in central Texas. For the types of correlation strengths among the regressor variables present in our data, we find that there does not seem to be much justification in selecting the principal component analysis approach. A more careful and elaborate investigation using data sets which exhibit a wide range of multicollinearity strengths is required in order to ascertain when principal component analysis yields predictive models superior to those of a multiple regression approach.