Data-driven modeling of heat pumps and thermal storage units for MPC

Abstract

Heat pumps can play a crucial role in the European energy strategy 2050, which aims to achieve net-zero greenhouse gas emissions. When coupled with thermal energy storage and integrated with advanced control strategies, heat pump operation can be optimized to reduce carbon footprint and respond to the needs of system operators. However, to scale in a multitude of buildings, the transferability of the modeling into heterogeneous systems is crucial. In this paper, two different interpretable linear models, a hybrid (grey-box) and a fully data-driven (black-box) model are investigated. Specifically, two regression-based identification methods (SINDYc and DMDc) are used for dynamic models and the LASSO regression is used for static models. The transferability of the approach is evaluated using two real-world facilities with heterogeneous sizing and configuration. The results show a similar simulation performance for both cases with a maximum normalized RMSE of 0.41 and 0.60, respectively. This confirms the transferability of the approach that is necessary for large-scale implementation.