Coordination Architecture Shapes Continuous Demand Response Outcomes in Building Districts
Researchers find hybrid AI cuts tracking error by 45% while keeping comfort violations low.
Deep Dive
Researchers compared four coordination architectures for a 25-building residential district demand response. The hybrid MPC-SAC controller achieved the best balance: 4.8% NMBE tracking bias, 16.8% comfort exceedance, and lowest spatial variability. Centralized MPC tracked well (8.8% NMBE) but caused 24.8% comfort exceedance and spatial imbalance. Decentralized RL distributed control evenly but failed to sustain accurate tracking.
Key Points
- Hybrid MPC-SAC architecture achieved lowest tracking bias at 4.8% NMBE and moderate comfort violations of 16.8%.
- Centralized MPC tracked well (8.8% NMBE) but caused 24.8% comfort exceedance and concentrated discomfort on specific buildings.
- Decentralized RL distributed control actions evenly but failed to sustain accurate load profile tracking.
Why It Matters
For smart grid deployment, choosing the right coordination architecture can cut energy waste and improve occupant comfort at scale.