Energy Efficiency Optimization and Preventive Maintenance of HVAC Systems Using Machine Learning

Aleksey Bratukhin, Christian Stippel, David Sengl, Markus Kobelrausch, Thilo Sauter

Abstract: HVAC systems are characterized by rigid and complex internal behaviour that limits the options for efficiency optimization. KI4HVACs project aims to apply machine learning to improve office building HVAC operation and maintenance efficiency. Reinforcement learning is used for energy use optimization and supervised learning for preventive maintenance. The models are wrapped into an MPC based and a decision tree algorithm for preventive maintenance. The system uses the TRNSYS18 simulation environment that matches an actual office building with multiple zones.