A growth curve-based Bayesian hierarchical model for multi-building energy use data analysis

Abstract Data-driven models, often called inverse models, such as change point models and artificial neural networks, are frequently used to predict the energy use of single buildings, and to a lesser extent, used to predict the energy use of multiple buildings. This study proposes a statistical estimation method to polarize energy use patterns across a group of residential buildings, which is then used to determine the net energy use pattern across this group of buildings. A monomolecular growth curve is employed to approximate the energy uses of multiple buildings, then a Bayesian hierarchical model, which enables the combining of energy use simultaneously across the targeted buildings. The proposed method is applied to whole-home energy use of 91 homes over 25 weeks cooling period in Austin, Texas. Posterior mean and confidence regions of estimators are obtained to check the significance of predictor variables. Estimation results support the ability to test to determine which factor(s) are statistically significant in predicting multi-building energy use.