Paper Conference

Proceedings of Building Simulation 2021: 17th Conference of IBPSA


Data-driven calibration of joint building and HVAC dynamic models using scalable Bayesian optimization

Ankush Chakrabarty 1, Emilio Maddalena 2, Hongtao Qiao 1, Christopher Laughman 1
1 Mitsubishi Electric Research Labs, United States of America
2 École polytechnique fédérale de Lausanne, Switzerland

Abstract: High-fidelity simulation models of coupled building and HVAC dynamics are typically expressed using grey-box simulators. These simulators comprise systems of differential algebraic equations, typically informed by physics, that need to be calibrated to realworld data. In this paper, we describe a framework for calibrating simulation model parameters using scalable Bayesian optimization (BO). BO uses probabilistic learners to approximate objective functions and leverages statistical information to efficiently explore and exploit the parameter space. This approach typically requires fewer simulations than Monte-Carlo methods or population-based algorithms, without relying on gradient information. To render the Bayesian framework tractable in high-dimensional parameter spaces, we also provide a scalable BO framework employing sparse Gaussian process regression. We demonstrate that our proposed approach can simultaneously calibrate 17 parameters (including radiative emissivities, heat transfer coefficients, and thickness of walls/floors) of a Modelica model of joint building and HVAC dynamics with >85% accuracy, and 13 of those 17 with > 90% accuracy.
Keywords: Bayesian optimization, machine learning, Modelica, parameter calibration, Gaussian processes
Pages: 1498 - 1505