Deep Reinforcement Learning-Based Cost-Optimized Control for Energy Management in Facilities with Geothermal Storage

This paper presents an approach to optimizing the energy cost of heating a building during the cold season using geothermal energy. A building model is developed using the IDA-ICE software and calibrated with real building data for accuracy. The optimization problem is addressed with a deep reinforcement learning (DRL) agent, which minimizes overall energy costs by interacting with the building while considering real-time electricity price and weather forecasts. The proposed method adapts to dynamic conditions, ensuring efficient thermal management and occupant comfort. The results demonstrate the effectiveness of the DRL-based controller in reducing energy costs compared to the currently used rule-based control strategy while maintaining optimal comfort levels. This work contributes to the advancement of cost-efficient heating strategies for sustainable building energy management.