Name: Iago Cupeiro Figueroa


KU Leuven

Promotor / Supervisor

Lieve Helsen

Samenvatting van het onderzoek / Summary of Research

Introduction / Objective
Hybrid GEOTABS is a holistic building concept that has a savings potential of 20-70%. Model Predictive Control strategies are desired to handle the increased control complexity that can hamper these envisaged savings. However, the ground dynamics time constant in the geothermal system is too large to be captured by the typical prediction horizons of MPC. This PhD thesis assesses whether incorporating the short- and long-term dynamic behavior of the geothermal system in MPC is important

Research Methodology
To this end, a set of control-oriented borefield models, each of them aiming at different dynamic time scales, is developed. The models are used in a simulation environment with MPC to answer the aforementioned question. A methodology to obtain accurate predictions of the borefield models in real systems using feedback information from the measurements is also evaluated. As a last step, the developed methodologies are tested in a real application case to demonstrate their robustness and flexibility.

Afbeelding 1

Results & Conclusions
The developed set of borefield models showed great accuracy for different ranges of complexity.
The results in the short-term dynamic window show that the formulation of the ground source heat pump model substantially affect the control performance, whereas the borefield model improves the reliable operation of the system.
In the long-term dynamic window, a methodology to account for the long-term ground dynamics in MPC is developed and evaluated. It is demonstrated that there is potential in hybrid geothermal systems beyond typical MPC horizons.

Afbeelding 2

State estimators were used as a feedback system, showing increased accuracy compared to raw estimations of the borefield model states. The use of Kalman Filters is recommended for the fast-dynamic states whereas Moving Horizon Estimator is suggested for the long-term dynamic states.
The developed methodologies still worked in a virtual real demonstrator case, showing their flexibility, robustness and scalability towards practical application.

Afbeelding 3