The optimal use of HVAC (Heating, ventilation, and air conditioning) resources in buildings for demand side management is complicated by the emergence of geothermal energy, district heating, HVAC participation on primary frequency reserve markets as well as using the buildings thermal mass as a heat storage. An emerging body of literature on multi-energy virtual power plants is addressing these trends partly. Further, this literature models buildings at a coarse granularity, so indoor temperature, air quality and deployment related issues require further research. In this project, a reinforcement learning agent is used to optimize building energy efficiencies, electricity bills and district heating bills under the uncertainties related to occupancy, energy prices and weather. A digital twin of a building complex is developed as a high-fidelity environment for training the agent.

2022

2026