Paper Conference

Proceedings of Building Simulation 2017: 15th Conference of IBPSA

     

Bayesian Calibration of Residential Building Clusters using a Single Geometric Building Representation

Martin Heine Kristensen1∗, Ruchi Choudhary2, Rasmus Høst Pedersen3, Steffen Petersen1
1Department of Engineering, Aarhus University, 8000 Aarhus C, DK
2Department of Engineering, University of Cambridge, Cambridge CB2 1PZ, UK
3AffaldVarme Aarhus, Bautavej 1, 8210 Aarhus V, DK∗Corresponding author (mhk@eng.au.dk)


DOI: https://doi.org/10.26868/25222708.2017.330
Abstract: For a homogeneous cluster of single-family dwellings, an archetype model incorporating simple scalable geometry and an hourly dynamic building energy model was set-up to represent its energy performance. Using metered annual energy use for a random sample of 450 buildings in the cluster, the archetype model was calibrated in a Bayesian regression framework using the floor area as common scale for regression of the physics-based input to the hourly dynamic energy model. In this process, posterior estimates of seven selected building parameters shared by buildings within the cluster were inferred. The calibrated archetype model was used to make predictions of annual building energy use with a normalized mean bias error (NMBE) of 2.3 % and a coefficient of variation of the root mean squared error (CVRMSE) of 26.5 %.
Pages: 1294 - 1303
Paper:
BS2017_330