A Hybrid Learning and Model-Based Optimization for HVAC Systems: A Real World Case Study

This paper discusses how to effectively integrate learning and model-based methods to optimize economic costs and operational efficiency of heating, ventilation, and air-conditioning (HVAC) systems. By leveraging learning-based methods, heuristics and manufacturing data of each unit in HVAC systems can be well approximated and integrated into the optimization framework. This paper provides an accurate and flexible modeling of an HVAC system to reach a highly economic and efficient daily operation schedule. To demonstrate the efficacy of proposed method, a real world public infrastructure is considered with detailed models and historical operational data. After combining data-driven models and physical models, the overall optimization problem formulation falls into the category of mixed-integer nonlinear optimization, and is further converted into a smooth nonlinear problem for easy-solving. Numerical results are compared to the existing energy consumption record, showing a substantial saving (50%) from the proposed method.