摘要
数据中心制冷系统因外界环境以及IT负荷的实时变化,无法及时更新系统参数设置,且因空调系统架构复杂、设备种类繁多、设备性能不一等原因无法准确设置最优参数,从而使其无法达到系统运行最佳状态,存在一定节能降耗潜力。为进一步提升数据中心制冷系统运行安全及能效,本文提出了一种基于多层粒子群算法的能耗优化调度方法,对空调制冷系统的神经网络黑盒模型进行智能寻优,并通过切片及分层寻优方法解决神经网络模型过拟合问题,避免伪最优解的直接输出,实现在确保数据中心制冷系统安全运行基础上,寻找到能效最低的最优运行状态。
The cooling system in data centers cannot promptly update its system parameters due to real-time variations in external environmental conditions and IT loads.Additionally,the complexity of the air conditioning system architecture,the diversity of equipment types,and varying equipment performance make it challenging to accurately set optimal parameters,preventing the system from achieving its best operational state.Consequently,there is potential for energy savings and efficiency improvements.To enhance both the operational safety and energy efficiency of data center cooling systems,this paper proposes an energy consumption optimization scheduling method based on the multi-layer particle swarm optimization algorithm.This approach conducts intelligent optimization of a neural network black-box model for the air conditioning cooling system and addresses overfting issues through slicing and hierarchical optimization methods.This ensures that the system avoids directly outputting pseudo-optimal solutions and achieves the most energy-eficient operational state while maintaining safe operation of the data center cooling system.
作者
孙晓杰
刘昊儒
余韵滢
邹凯凯
曹国水
SUN Xiao-jie;LIU Hao-ru;YU Yun-ying;ZOU Kai-kai;CAO Guo-shui(China Mobile Group Zhejiang Co.,Ltd.,Hangzhou 310016,China)
出处
《电信工程技术与标准化》
2024年第S01期242-246,共5页
Telecom Engineering Technics and Standardization
关键词
数据中心
制冷能耗
优化调度
粒子群算法
data center
refrigeration energy consumption
optimization scheduling
particle swarm optimization
