Evaluating heat pump-enabled building flexibility potential via deep reinforcement learning and trade-off analysis
Electricity grids increasingly face volatility due to supply-demand imbalances and congestion. To manage this, grid operators seek flexibility not only from power plants and batteries but also from underutilized resources, such as the thermal inertia of residential buildings equipped with heat pumps. This study explores how these systems can contribute to grid stability while balancing consumer objectives like comfort, cost savings, and environmental impact reduction. We propose a Deep Reinforcement Learning (DRL) approach to model and optimize heating control strategies. An agent learns to determine optimal heating actions and temperature setpoints based on different reward structures. These include comfortoriented control (maintaining indoor temperature), cost minimization, and reduction of environmental impact. We extend this to include flexibility maximization, where the agent adapts its behavior to grid flexibility requirements. The approach is validated through a simulation environment. Results demonstrate the agent’s ability to analyze and negotiate trade-offs between competing objectives and respond dynamically to grid flexibility requirements. This work serves as a demonstration of embedding trade-off analysis in evaluating and activating flexibility.