Building thermal control: Hierarchical design from limited data using gray-box or black-box internal models for model predictive control
Résumé
This paper presents a hierarchical model predictive control (MPC) framework designed to accommodate the diversity of heating equipment in building energy systems. The proposed architecture consists of a supervisory MPC for power planning and tracking controllers for device-level regulation. The framework systematically compares grey-box system identification, using an equivalent Resistance-Capacitance (RC) model, against black-box Subspace State Space Identification (4SID), evaluating their performance with limited datasets. At the tracking level, Virtual Reference Feedback Tuning (VRFT) offers a model-free approach to equipment control, eliminating the need for a detailed Heating, Ventilation, Air Conditioning system model. The proposed approach is validated using a multi-zone residential test case from the Building Optimization Testing Framework. Results show that although both models achieve similar open-loop prediction accuracy, their performance diverges under closed-loop control. Under different weighting configurations, the proposed scheme using the RC model achieves approximately a 3-12% reduction in energy consumption while maintaining comparable or even lower levels of thermal discomfort compared to the blackbox 4SID model. The study concludes with a discussion of practical considerations and the potential for broader deployment.
| Origine | Fichiers éditeurs autorisés sur une archive ouverte |
|---|---|
| Licence |