基于机器学习方法的三维气流组织温度场重构

Reconstruction of the Three-Dimensional Temperature Field of Air Ditribution Based on the Machine Learning Method

从局部离散流场数据点重构全局流场信息对数据中心机房的节能节碳具备重要的研究意义。基于局部散点对数据中心进行气流组织温度场快速重构,可有效降低数据中心总能耗。主要通过发展一种基于多特征输入的三维U形神经网络架构(U-net),利用布设在数据中心机房的温度传感器数据值进行气流组织温度场重构。并研究了在不同学习率与数据集大小的设置下,该机器学习模型的训练预测能力。对比结果表明:不同学习率对于模型训练的结果有较大影响,应通过预实验选取最佳学习率。在同等条件下,应优先选取较大的数据集进行训练,便于提取高维物理特征。

It is of important research significance for energy conservation and carbon conservation in the computer rooms of the data center to reconstruct global flow field information from local discrete flow field data points,and the rapid reconstruction of the temperature field of air distribution in the data center based on local scatter points can effectively reduce the total energy consumption of the data center.This article mainly develops a threedimensional U-net neural network architecture based on multi-feature input,which utilizes the data value of tem⁃perature sensors deployed in the computer rooms of the data center to reconstruct the temperature field of air distri⁃bution,and studies the training prediction ability of the machine learning model under the setting of different learn⁃ing rates and dataset sizes.The comparison results indicate that different learning rates have great impacts on the training results of the model,and the optimal learning rate should be selected through pre-experiments,and that under the same conditions,priority should be given to selecting a larger dataset for training,in order to extract high-dimensional physical features.