基于深度强化学习策略的空气源热泵除霜能效提升  

Efficiency improvement of air source heat pumps using deep reinforcement learning-based defrosting strategy

作　　者：张博 田煦嘉 ZHANG Bo;TIAN Xujia(Beijing North Star Company Ltd.,Beijing 100100,China)

机构地区：[1]北京北辰实业股份有限公司,北京100100

出　　处：《能源研究与信息》2025年第1期27-32,共6页Energy Research and Information

摘　　要：热泵除霜技术在空调等制冷系统中发挥着重要的作用,但传统的除霜方法存在能耗高、效率低、系统性能下降等问题。采用先进的强化学习技术,对热泵除霜过程进行了模型优化,采用Q-learning算法及epsilon-greedy策略学习一个最优的动作值函数,同时在探索和利用之间进行权衡。结果表明:基于Q-learning算法构建的模型能够准确预测热泵除霜过程中的参数变化,并可实现热泵系统更高的能源利用效率。研究可为热泵除霜技术的优化提供理论基础和实际指导。The defrosting technologies of heat pumps are pivotal in air conditioning and refrigeration systems.Nonetheless,conventional defrosting techniques are plagued by high energy consumption,limited efficiency,and reduced system performance.This research optimized the heat pump defrosting process model using advanced reinforcement learning techniques,emphasizing exploration-exploitation equilibrium and deriving an optimal action-value function via the Qlearning algorithm and the epsilon-greedy approach.Findings demonstrate that the Q-learning algorithm-based model accurately forecasts parameter variations in the heat pump defrosting process,improving energy utilization efficiency within the heat pump system.It offers both theoretical underpinning and practical insights for optimizing heat pump defrosting technologies.

关 键 词：热泵 除霜 强化学习 Q-learning算法 epsilon-greedy策略 能耗 效率 

分 类 号：TU83[建筑科学—供热、供燃气、通风及空调工程]