基于深度强化学习的多区域通风系统风量控制方法研究  

Research on the Air Flow Control Method of Multi-zone Ventilation System Based on Deep Reinforcement Learning

作　　者：李春晓 崔璨 黎明[1] LI Chunxiao;CUI Can;LI Ming(College of Engineering,Ocean University of China,Qingdao 266100,China)

机构地区：[1]中国海洋大学工程学院,山东青岛266100

出　　处：《控制工程》2025年第2期265-272,共8页Control Engineering of China

基　　金：山东省自然科学青年基金资助项目(ZR2021QE291);山东省优秀青年科学基金项目(海外)(2022HWYQ-064)。

摘　　要：通风系统在实际运行时,其风量负荷动态变化,且各区域风量之间存在强耦合,导致其风量调节的难度高、调节时间长。针对此问题,提出一种基于深度确定性策略梯度的多区域通风系统风量控制方法,实现对各区域风量的快速、准确控制。此外,提出一种动态目标训练机制,有效提高强化学习训练的效率。最后,建立通风系统仿真环境,验证所提控制方法的性能。仿真结果表明,所提控制方法能够快速实现对风量的准确控制,能够抵抗一定程度的系统噪声,并且适用于不同拓扑结构的通风系统。The demand flow is dynamically changing during the operation of the ventilation system,and there is strong coupling between the air flow of each zone.This poses challenges for the control of air flow and can cause long regulation time.An air flow control method based on deep deterministic policy gradient(DDPG)is proposed to realize fast and accurate control of air flow in each zone.In addition,a dynamic target training mechanism is proposed,which effectively improves the training efficiency.Finally,a ventilation system simulation model is built to verify the performance of the proposed control method.The simulation results show that the proposed method can achieve fast and accurate control of air flow.Besides,the results also show that the proposed method is not susceptible to system noise,and can be applied to other ventilation systems of different topologies.

关 键 词：风量控制方法 深度强化学习 深度确定性策略梯度 动态目标 多区域通风系统 

分 类 号：TP39[自动化与计算机技术—计算机应用技术]