In European countries, space heating is the main energy consumer in the residential sector, contributing significantly to CO2 emissions that badly impact both the environment and public health. Green energy presents a promising solution to tackle this issue; however, space heating solutions must be flexible to accommodate the variable availability of this energy source. In this thesis, we develop advanced heating solutions that efficiently adapt to the inherent variability of green energy sources. We suggest intelligent model predictive controllers (MPCs) as a sophisticated tool for developing advanced control strategies to optimize energy efficiency in domestic heating systems. We construct ordinary differential equations (ODEs) to capture the thermal dynamics of the buildings. We then utilize historical sensor data and the derived ODEs in grey-box models to estimate concealed heat transfer coefficients among various energy variables. We utilize the estimated model and implement a reinforcement-learning based online MPC framework in Uppaal Stratego tool. We develop several advanced control strategies tailored to identify potential energy-saving opportunities within this framework. These strategies encompass diverse aspects of occupants' preferences, including fixed set-points, set-point ranges, and adjustments for low temperatures during occupants' absence or sleep. Leveraging the power of reinforcement-learning, these strategies integrate future energy prices and weather forecasts to learn near-optimal decisions. Additionally, we propose dynamic and adaptable buffer tank and mixing loop models to broaden the applicability of our solution. The proposed methods and approaches collectively constitute a comprehensive tool chain designed to control domestic heating systems effectively. We validate our tool-chain by applying it to sensor data from different domestic heating environments, including an experimental single-family house and a real multi-apartment building. The experimental analysis demonstrates substantial reductions in energy consumption costs while maintaining user comfort levels. These results confirm the seamless integration of our methods with real data, thus validating their accuracy.

All interested parties are welcome. After the defense the department will be hosting a small reception in cluster 1.