HVAC systems account for at least 40% of the energy consumption of general office buildings. Therefore, reducing the energy use of HVAC systems is indispensable. HVAC systems used in public office buildings mostly adopt a central air conditioning system. To save energy in the central air conditioning system, high-efficiency heat source machines have been adopted, and the inverter control of pumps has been generally introduced. As a further measure to save energy, an optimal control has been proposed. Model-based control has been mainly studied. However, the target HVAC system must be appropriately equipped with sensors for model-based control to use previous operation data. To solve this problem, we proposed a model-free optimal control method using Bayesian optimization. We verified its effectiveness in one-on-one (one cooling tower and one chiller) HVAC systems. As a result, the energy-saving ratio when compared to the rated specification control is 10.38% for our proposed method and 11.34% for the model-based approach, which shows the equivalent performance. In addition, the training results indicate that the optimal set values can be automatically determined in 4 weeks as the training proceeds under rated specification control with Bayesian optimization.

暖通空调系统至少占一般办公楼能耗的 40%。因此，减少暖通空调系统的能源使用是必不可少的。公共写字楼使用的暖通空调系统大多采用中央空调系统。中央空调系统为节能，采用高效热源机，普遍引入水泵变频控制。作为进一步的节能措施，已经提出了优化控制。主要研究了基于模型的控制。然而，目标 HVAC 系统必须适当配备传感器，用于基于模型的控制，以使用以前的操作数据。为了解决这个问题，我们提出了一种使用贝叶斯优化的无模型最优控制方法。我们在一对一（一个冷却塔和一个冷却器）HVAC 系统中验证了它的有效性。因此，与额定规格控制相比，我们提出的方法的节能率为 10.38%，基于模型的方法为 11.34%，这表明性能相当。此外，训练结果表明，随着训练在贝叶斯优化的额定规格控制下进行，可以在 4 周内自动确定最优设定值。