Bayesian Calibration for Office-Building Heating and Cooling Energy Prediction Model

Abstract

Conventional building energy models (BEM) for heating and cooling energy-consumption prediction without calibration are not accurate, and the commonly used manual calibration method requires the high expertise of modelers. Bayesian calibration (BC) is a novel method with great potential in BEM, and there are many successful applications for unknown-parameters calibrating and retrofitting analysis. However, there is still a lack of study on prediction model calibration. There are two main challenges in developing a calibrated prediction model: (1) poor generalization ability; (2) lack of data availability. To tackle these challenges and create an energy prediction model for office buildings in Guangdong, China, this paper characterizes and validates the BC method to calibrate a quasi-dynamic BEM with a comprehensive database including geometry information for various office buildings. Then, a case study analyzes the effectiveness and performance of the calibrated prediction model. The results show that BC effectively and accurately calibrates quasi-dynamic BEM for prediction purposes. The calibrated model accuracy (monthly CV(RMSE) of 0.59% and hourly CV(RMSE) of 19.35%) meets the requirement of ASHRAE Guideline 14. With the calibrated prediction model, this paper provides a new way to improve the data quality and integrity of existing building energy databases and will further benefit usability.