Abstract
In today’s world, the functioning of modern societies is fundamentally dependent on uninterrupted access to energy, powering everything from essential services such as heating, cooking, and healthcare to broader systems like communication, transportation, and digital infrastructure. As climate change accelerates and the geopolitical landscape becomes increasingly unstable, energy systems face mounting challenges, including supply disruptions, demand growth, and increasing trend in extreme weather events. These pressures expose the vulnerabilities of existing infrastructure and underscore the urgent need for more resilient, flexible, and adaptive energy solutions in urban areas. In this context, improving energy performance and resilience in buildings, where a significant share of energy is consumed, has become a key strategy.
This doctoral research explores the application of Collective Intelligence (CI) in decentralized energy management systems for buildings, aiming to improve energy efficiency, flexibility, and resilience under climate uncertainty and grid instability. The proposed system architecture features two hierarchical levels: edge nodes representing individual thermal zones or buildings, and cluster nodes coordinating groups of edge nodes. Flexibility signals, generated at the grid level and adapted locally, drive system coordination while enabling autonomous, localized decision-making.
Each edge node is equipped with a Smart Control System (SCS) comprising three modular components: engine, functions, and interface. The engine employs AI-based decision-making algorithms. Reinforcement Learning (RL) is used in this research to enable real-time decisionmaking based on dynamic environmental feedback. The functions module includes a signal interpreter, reward function, and action generator, which together tailor responses to both user preferences and system conditions. The interface module manages secure communication and interoperability with external devices.
The architecture was tested virtually in two Swedish residential buildings and seven Norwegian public buildings, and experimentally validated in a living lab at G2ELab in France. The virtual testing used high-resolution building energy models calibrated against measured energy use and indoor temperatures. In the experimental setup, the SCS was integrated into the building management system using the existing infrastructure of the living lab. This real-world deployment included building occupants and addressed challenges related to communication across multiple systems.
Results show energy use reductions of up to 40%, cost reductions of 30%, and peak power curtailment by 25% in the virtual test environment. Integration with renewable energy sources increased self-consumption to 100%, while storage systems raised grid autonomy to over 80% during midday periods. Despite extreme weather conditions, indoor thermal comfort was largely maintained, with discomfort thresholds exceeded in less than 1% of timesteps. The real-world deployment demonstrated an 18% reduction in overall energy use compared to the predicted baseline, with heating and cooling demands decreasing by 5% and 32%, respectively. Furthermore, peak power demand was reduced by up to 15% for heating and as much as 50% for cooling. The introduction of demand flexibility improved the system’s adaptability and lessened the sensitivity of energy use to outdoor temperature fluctuations. Positive effects of energy flexibility on energy consumption were observed in approximately 70% of winter and 80% of summer timesteps.
Overall, the research highlights the potential and benefits of CI-driven, scalable, and decentralized control systems to enhance building energy performance while maintaining comfort, supporting grid stability, and adapting to future energy and climate challenges.