ED Sciences Physiques et de l'Ingénieur
Innovative Self-Optimizing Control of Building Microgrids Exploiting Hydrogen Multiple Services Potential
by Fahad Ali SARWAR (ESTIA-Recherche)
The defense will take place at 9h30 - Amphi 300 Technopole Izarbel 90 Allée Fauste d'Elhuyar, 64210 Bidart, France
in front of the jury composed of
- Ionel VECHIU - Professeur - ESTIA - Directeur de these
- Furong LI - Professeure - University of Bath - Rapporteur
- Seddik BACHA - Professeur - G2E Lab Grenoble - Rapporteur
- Barry HAYES - Associate Professor - University College Cork - Examinateur
- Tapia Otaegui GERARDO - Professeur - universidad del pais vasco - Examinateur
The growing energy demand and the increasing emphasis on decarbonization are driving the necessity to integrate renewable energy sources into microgrids and buildings. Due to the intermittent nature of these renewable energy sources, energy storage systems have become critical for achieving this decarbonization and decreasing dependence on larger grids, particularly in the context of self-consumption in buildings and net-zero energy microgrids. Conventional batteries have been widely used for energy storage applications; however, they are unsuitable for long-term energy storage. To address this limitation, hydrogen has been proposed as an alternative energy source for long-term storage in microgrids, offering the ability to store energy over extended periods with minimal losses. Adding such a hybridized energy storage mechanism in microgrids increases their complexity and operation and requires the development of an integrated energy management system. Accordingly, this thesis provides a complete framework for developing and designing an integrated energy management system that can maximize local energy consumption in microgrids and buildings. In addition to self-consumption, maximizing hydrogen production has been a key aspect of this thesis. To obtain optimal self-consumption, the proposed energy management system also focuses on optimally managing the microgrid components while ensuring long lifetimes for a wide range of equipment including electrolyzers and fuel cells. Consequently, to develop an integrated energy management system with these key objectives, a strategy based on reinforcement learning is proposed for its development and control. This reinforcement learning-based methodology employs a weighted-average reward mechanism to determine the optimal operational setpoints of the microgrid, ensuring maximized outcomes. The proposed approach effectively balances multiple objectives, including optimizing self-consumption, hydrogen production, and equipment longevity, to deliver a holistic and efficient energy management solution. Based on this optimal control of the microgrid, the proposed technique offers a highly flexible solution capable of integrating and computing any additional parameters required for optimization, which makes the framework presented in this thesis a fully versatile tool for the integration of artificial intelligence in energy microgrids. The implementation of the proposed control strategy for microgrid operation highlights the effectiveness of reinforcement learning based energy management in optimizing self-consumption, achieving an increase of up to 17%. Moreover, a reduction in the average number of operating cycles of the microgrid equipment and energy storage system is also achieved, making the overall operation of the system substantially more efficient. Accordingly, based on the wide range of analyzed scenarios, the proposed strategy can lead to a reduced number of cycles in the electrolyzer of up to 15 %. In summary, the reinforcement learning based energy management system framework not only improves self-consumption and minimizes equipment wear but also enables efficient hydrogen integration into microgrids. These results emphasize the potential of advanced energy management strategy in achieving decarbonization targets while strengthening energy resilience and system performance.