数据中心高架地板湍流模型系数修正与优化研究  

Data center elevated floor model and turbulence model evaluation

作　　者：张一鸣 朱兵[1] 殷佳辉 黄金森 Zhang Yiming;Zhu Bing;Ying Jiahui;Huang Jingsen(College of Electrical Engineering,Guizhou University,Guiyang 550025,China)

机构地区：[1]贵州大学电气工程学院,贵阳550025

出　　处：《低温与超导》2023年第9期85-90,96,共7页Cryogenics and Superconductivity

基　　金：贵州省科技支撑计划项目(2017YFB0902100)资助。

摘　　要：随着5G时代的兴起,数据中心的热负荷上升。为提高数据中心温度分布预测精度,开发更加高效的数据中心热管理系统,本研究采用机器学习与全局寻优方法修正了Standard k-ε湍流模型的参数。通过模拟不同参数组合,构建数据集,使用神经网络进行训练的方法减小了近似物理模型(湍流模型)误差,提高了其精度。修正后的Standard k-ε湍流模型在机柜前后侧的平均均方根误差为0.946,与之前相比降低了29.47%,使得模型的准确性明显提高,为数据中心动态控制系统提供数据驱动,实现精确控制,提高节能率。The thermal load of data centres rises with the rise of 5G era.To improve the thermal management efficiency of data centres,the parameters of the Standard k-turbulence model were corrected by using machine learning and global optimality seeking methods.And the approximate physical model(turbulence model)error was reduced and its accuracy was improved by simulating different parameter combinations,constructing data sets and using neural networks for training.The average root mean square error of the corrected Standard k-turbulence model at the front and rear sides of the cabinet is 0.946,which is 29.47%lower than before,resulting in a significant improvement in the accuracy of the model,providing data-driven dynamic control system of the data centre for precise control and increasing energy savings.

关 键 词：湍流模型 系数修正 机器学习 计算流体力学 

分 类 号：TK221[动力工程及工程热物理—动力机械及工程]