Paper Conference

Proceedings of Building Simulation 2023: 18th Conference of IBPSA

   

A virtual testbed for robust and reproducible calibration of building energy simulation models

Sicheng Zhan 1, Ankush Chakrabarty 2, Christopher Laughman 2, Adrian Chong 1
1 National University of Singapore, Singapore
2 Mitsubishi Electric Research Laboratories, Cambridge, MA, United States


DOI: https://doi.org/10.26868/25222708.2023.1482
Abstract: A reliable building energy simulation model is critical for improving building energy performance. While many auto-calibration approaches have been proposed, robust and reproducible BES model calibration remains a challenge due to the lack of a universal evaluation approach and benchmarking framework. Hence, we establish a virtual testbed based on the DOE prototype buildings to systematically evaluate calibration results. The Modelica-based testbed enables customized dataset generation and provides the model discrepancy between the calibrated models and the calibration target, which is the key to emulating realistic calibration tasks. We identify three categories of typical pitfalls in BES model calibration and demonstrate them using the virtual testbed. Lastly, a hierarchical model evaluation framework is designed upon the testbed for further calibration studies. This study investigates model calibration for buildings from a new perspective and facilitates further research with a standardized framework.
Keywords: building energy simulation, model calibration, synthetic dataset, optimization
Pages: 2339 - 2346