Multi-Level Bayesian Modelling of Building Energy Consumption

Abstract

This paper describes an approach to Multi-Level Bayesian modelling of building energy consumption. A total of six model parameterizations were tested. They address two challenges of predictive models. These are precision and interpretability. The model parametrizations are based on a database of more than 360 thousand building records spread across 96 cities in the United States. The most accurate model had 4-Levels and a MAPE of 13.1%. For demonstration purposes, this model was used to predict the Energy Use Intensity of commercial and residential buildings in the United States until mid-century. The paper described the benefits of the Multi-Level Bayesian approach and lay recommendations for future work.