A Reinforced Learning (RL) algorithm for optimal control of HVAC: Comparison with a standard PI control for the supermarket DOE reference building

Efficient control of Heating, Ventilation and Air Conditioning systems can lead to great reduction in energy consumption. This can be achieved by new data-driven control algorithms based on Reinforcement Learning (RL). In this work Dynamic Simulation is coupled with a model-free RL algorithm to study its performance in terms of energy saving and thermal comfort in a realistic scenario. Two models are derived from the DOE Supermarket Reference Building for two climate locations. The simulations performed show a reduction between 5.4% and 9.4% in primary energy consumption for the two locations, guaranteeing the same thermal comfort of state-ofthe-art controls.