An integrated development framework for data-driven model predictive control of thermal zones in buildings

Area of Science:

• Building energy systems
• Control theory
• Renewable energy integration

Background:

• Growing renewable energy integration requires enhanced building energy flexibility.
• Data-driven model predictive control (DMPC) is a key technology for building energy management.
• Existing simulation tools may lack the flexibility to test advanced DMPC strategies.

Purpose of the Study:

• Introduce a comprehensive simulation environment for developing and evaluating DMPC solutions for buildings.
• Provide a validated framework for testing advanced control strategies, including those for Thermally Activated Building Structures (TABS) and solar shading.
• Facilitate research into decentralized control approaches for complex building systems.

Main Methods:

• Developed a customizable zone model for multi-room configurations.
• Integrated DMPC algorithms with adaptable state-space models and reinforcement learning.
• Supported various optimization architectures (centralized, decentralized).
• Validated the environment using over a year of data from a living-lab office building.

Main Results:

• Thermal zone models achieved high accuracy (mean absolute errors < 0.5 °C).
• The simulation environment accurately captured dynamics of TABS and solar shading.
• Demonstrated robust performance in simulating complex building systems.

Conclusions:

• The validated simulation environment effectively supports the development of DMPC for buildings.
• The framework is suitable for evaluating advanced control strategies and decentralized approaches.
• Enables researchers to address challenges in complex building energy management and grid integration.