摘要
强化学习方法在近年来被逐渐尝试应用于工程控制之中,并且展示出其强大的控制能力和潜力。强化学习算法虽然具有较好的控制性能,但是其控制策略的收敛需要建立在一段时间后的训练上,而这段时间在工程应用上可能会造成一些浪费。为了解决这一问题,提出了基于先验知识的DQN对暖通空调(HVAC)系统中的冷却水系统进行控制,将先验知识引入DQN算法的前期训练中,以减少训练时间,加速收敛,节约成本。基于先验知识的DQN算法不仅能够对系统进行有效的控制,并且能够比DQN更早地实现节能。
Reinforcement learning method has been gradually applied to engineering control in recent years,and has shown its strong control ability and potential.Although reinforcement learning algorithms have good control performance,the convergence of its control strategy needs to be established after a period of training,which may cause some unnecessary cost waste in engineering application.In order to solve this problem,this paper proposes DQN based on priori knowledge to control the cooling water system in HVAC system,and introduces priori knowledge into the early training of DQN algorithm,so as to reduce training time,accelerate convergence and save cost.DQN algorithm based on priori knowledge can not only effectively control the system,but also save energy earlier than DQN.
出处
《工业控制计算机》
2023年第3期32-33,36,共3页
Industrial Control Computer
关键词
强化学习
先验知识
加速收敛
暖通空调
reinforcement learning
priori knowledge
accelerate convergence
HVAC