The use of machine-learning algorithms in optimizing the energy efficiency of HVAC systems has been widely studied in recent years. Previous research has focused mainly on data-driven model predictive controls and reinforcement learning. Both approaches require a large amount of online interactive data; therefore, they are not efficient and stable enough for large-scale practical applications. In this paper, a Bayesian optimization framework for HVAC control has been proposed to achieve near-optimal control performance while also maintaining high efficiency and stability, which would allow it to be implemented in a large number of projects to obtain large-scale benefits. The proposed framework includes the following: (1) a method for modeling HVAC control problems as contexture Bayesian optimization problems and a technology for automatically constructing Bayesian optimization samples, which are based on time series raw trending data; (2) a Gaussian process regression surrogate model for the objective function of optimization; (3) a Bayesian optimization control loop, optimized for the characteristics of HVAC system controls, including an additional exploration trick based on noise estimation and a mechanism to ensure constraint satisfaction. The performance of the proposed framework was evaluated by using a simulation system, which was calibrated by using trending data from a real data center. The results of our study showed that the proposed approach achieved more than a 10% increase in energy-efficiency savings within a few weeks of optimization time compared with the original building automation control.

Reinforcement learning (RL) is being gradually applied in the control of heating, ventilation and air-conditioning (HVAC) systems to learn the optimal control sequences for energy savings. However, due to the “trial and error” issue, the output sequences of RL may cause potential operational safety issues when RL is applied in real systems. To solve those problems, an RL algorithm with dual safety policies for energy savings in HVAC systems is proposed. In the proposed dual safety policies, the implicit safety policy is a part of the RL model, which integrates safety into the optimization target of RL, by adding penalties in reward for actions that exceed the safety constraints. In explicit safety policy, an online safety classifier is built to filter the actions outputted by RL; thus, only those actions that are classified as safe and have the highest benefits will be finally selected. In this way, the safety of controlled HVAC systems running with proposed RL algorithms can be effectively satisfied while reducing the energy consumptions. To verify the proposed algorithm, we implemented the control algorithm in a real existing commercial building. After a certain period of self-studying, the energy consumption of HVAC had been reduced by more than 15.02% compared to the proportional–integral–derivative (PID) control. Meanwhile, compared to the independent application of the RL algorithm without safety policy, the proportion of indoor temperature not meeting the demand is reduced by 25.06%.