Deep Reinforcement Learning Based HVAC Optimization Control Algorithm Application

To address the shortcomings of Heating, Ventilation, and Air Conditioning systems (HVAC) with low energy efficiency, this paper introduces the application and effectiveness of deep reinforcement learning in HVAC systems. Deep reinforcement learning mainly includes model-based algorithms and model-free algorithms. Model-based algorithms require a large amount of system and environment knowledge, which is usually difficult to obtain, while model-free algorithms do not need this knowledge to model the environment, so model-free deep reinforcement learning is of high research value. The applications of model-free deep reinforcement learning algorithms in the HVAC domain are reviewed and divided into three categories: value function-based methods, policy gradient-based methods, and actor-critic-based methods, and the three categories are also described in detail. Finally, the shortcomings of current deep reinforcement learning in HVAC control systems are summarized, and the future application directions are prospected.