Predictive Two-Level Energy Management for the Energetic Optimization of Multi-Family Houses and Districts

In Germany, about 23% of the total final energy is consumed for the heat supply (space heating and warm water) of residential buildings (as of 2022) (UBA 2024a, 2024b). Approximately three million older medium-sized multi-family houses with 3 to 12 residential units are responsible for a significant portion of CO₂ emissions in the building sector. To achieve climate goals, the number of renovations would need to increase from the current approximately 4.1 million to 13-16 million buildings by 2045. The dynOpt-San project (BMWK, 2024) supports the achievement of these goals by developing standardized renovation concepts and efficiently integrating innovative photovoltaic-thermal systems in combination with phase-change material storages (PVT-PCM systems). Additionally, a self-learning energy management system with integrated operational monitoring is being developed to optimize and monitor the operation of multi-family houses and districts.
In this contribution, we showcase a prototypical version of such energy management system as well as cloud-based monitoring tools, and we present initial results and lessons learned from first real demonstrator buildings. The predictive energy management relies on a two-level architecture to coordinate energy flows at both the building and district levels with minimal effort. To model the energy system components, we utilize the open-source python framework oemof to formulate mixed-integer linear problems. To facilitate predictive optimization, we incorporate information about future electricity prices, weather forecasts, as well as energy consumption forecasts on residential level, generated with machine learning approaches. The objectives of the energy management system include reducing costs and CO₂ emissions, achieving an optimal self-consumption rate within the buildings, and promoting grid-friendly behavior of the district.