Single-layer Economic MPC for Real-Time Optimal Energy Operation in Buildings

• Autores: José Antonio Borja Conde
• Directores de la Tesis: Filiberto Fele (dir. tes.), Daniel Limón Marruedo (dir. tes.)
• Lectura: En la Universidad de Sevilla ( España ) en 2026
• Idioma: inglés
• Número de páginas: 236
• Enlaces
  • Tesis en acceso abierto en: Idus
• Resumen

  • Buildings are responsible for a large share of electricity demand, with Heating, Ventilation and Air Conditioning (HVAC) systems being among the main contributors. This thesis develops a family of Quadratic-Program (QP)-based Economic Model Predictive Control (EMPC) schemes that preserve the computational efficiency of QPs while directly targeting economic performance and demand response objectives.

    The thesis starts by addressing a key enabler for their successful deployment of such controllers: the availability of accurate, reliable energy models of buildings. We focus on the creation of high-fidelity Digital Twins of building dynamics, providing a versatile basis for MPC-oriented surrogate models and broader energy-management tasks such as validation of controllers in a virtual testbench or system design. The proposed modeling approach couples the fidelity of white-box simulation environments with purposeful simplifications that reduce parameter requirements and remove reliance on detailed architectural metadata. Leveraging historical operational data, it offers a practical path to develop Digital Twins for existing buildings.

    After addressing the modeling aspects, we implement a QP-based EMPC adapted for multi-chiller HVAC systems. This formulation is based on a single-layer strategy that replaces the economic cost with its first-order Taylor expansion, thereby casting both the economic optimization and the control into a single QP at each sampling instant. Experiments on a realistic multi-chiller plant show improved economic operation compared with conventional set-point or tracking MPC, while revealing a key necessary insight: the employed formulations are solved at each sampling instant as steady-state problems, whereas optimizing dynamic trajectories is required to fully exploit system conditions and enable demand response.

    To address this, we propose a Periodic QP-based EMPC tailored to cyclic operation (e.g., daily prices and loads). By exploiting periodicity, the controller can anticipate and generate economically efficient trajectories under changing periodic operating conditions, while preserving the recursive feasibility, asymptotic stability, and the computational efficiency of the original formulation. Its effectiveness is demonstrated in a representative off-grid building case and in a realistic HVAC case study from the BOPTEST framework.

    We then enhance this approach through a line-search, making the controller design fully independent of the particular form of the economic cost function while also increasing the convergence speed. The new (QP) controller exhibits adaptivity to changes in the economic function, while preserving the control-theoretic guarantees.

    Next, we extend this framework to address non-convex economic criteria. The proposed controller achieves offset-free asymptotic optimality while preserving the computational complexity of a QP. The approach is based on an offline-trained quadratic surrogate of the original EMPC problem, complemented with a gradient-based refinement that is then applied online to ensure convergence to the optimum.

    Finally, we robustify the QP-based EMPC schemes by introducing a (QP) configuration-constrained tube formulation for robust periodic economic operation. The method handles additive disturbances and multiplicative parameter variations, enforcing constraints via robust forward-invariant tubes and an implicit terminal region that streamlines the design. A flexible tube parameterization broadens feasibility compared with rigid-tube alternatives.

    Across these contributions, the thesis demonstrates that gradient-informed QP formulations provide a unifying, computationally lightweight path to periodic demand response capability, robustness, and—even for non-convex objectives—economic optimality, making EMPC viable for real-time deployment in commercial platforms for building energy management.