Data-driven online control for real-time optimal economic dispatch and temperature regulation in district heating systems
A new AI control framework eliminates the need for forecasts, achieving stable near-optimal operation for industrial energy systems.
Researchers Xinyi Yi and Ioannis Lestas have published a novel AI framework for optimizing the complex, real-time operation of District Heating Systems (DHS). The core innovation is a Data-Enabled Policy Optimization (DeePO)-based controller that embeds steady-state economic optimality conditions directly into the system's temperature dynamics. This design allows the closed-loop system to autonomously converge to the most cost-effective operating point, eliminating the traditional and often unreliable dependency on disturbance forecasts and precise nominal models. The researchers further enhanced the controller by incorporating the Adaptive Moment Estimation (ADAM) optimization algorithm to improve its closed-loop performance and learning stability.
The paper establishes formal mathematical guarantees for the system's convergence and performance, providing a solid theoretical foundation. The practical efficacy of the framework was validated through simulations on a real-world industrial-park DHS located in Northern China. The results demonstrated that the AI controller achieves stable, near-optimal economic dispatch and temperature regulation. Crucially, it exhibited strong empirical robustness against both static and time-varying model inaccuracies under practical disturbance conditions, a common failure point for conventional control strategies. This represents a significant step toward more resilient and autonomous infrastructure management.
- The DeePO framework embeds economic goals into system control, removing reliance on error-prone forecasts and models.
- It incorporates the ADAM optimization algorithm to enhance learning and closed-loop performance with proven convergence guarantees.
- Simulations on a Northern China industrial DHS confirmed stable near-optimal operation and robustness to real-world model mismatches.
Why It Matters
This enables more resilient, cost-effective, and autonomous management of critical energy infrastructure without perfect predictive data.