This project aims to develop a novel and scalable building energy modelling and optimal control framework by using modern AI techniques to optimize campus building HVAC operations and transform campus buildings into grid-interactive smart buildings.
Research themes: Energy and GHG emissions
This project aims to develop a novel and scalable building energy modelling and optimal control framework by using modern AI techniques to optimize campus building HVAC operations.
We combine state-of-the-art AI techniques: deep reinforcement learning (RL) and Bayesian machine learning (ML), as well as a physics-based building simulation tool. First, we will develop a Bayesian physics-informed ML module to develop a virtual building model with which deep RL will be trained. Second, we will develop a deep RL algorithm that learns the optimal control policy with the developed model. We will demonstrate the framework through a series of simulation studies and experimental evaluation with a campus building.
Previous solutions for HVAC operations required either labour-intensive engineering effort or a long period of data collection for each individual building; these have prevented their adoption. The developed framework will enable the development of optimal HVAC controllers for each individual building without extensive engineering effort and an immense amount of data. We expect that the controllers developed with the proposed framework will save 10-30% of the target building’s HVAC energy consumption and reduce the peak demand while satisfying occupants.