CFD与机器学习相结合的某数据机房冷通道热环境模拟
Simulation of cold aisle thermal environment in a data room based on CFD and machine learning
摘要：
为了降低数据机房模拟的成本,快速地预测机房的热环境,本文以常州市某大型数据中心内一个实际运行中的机房为研究对象,首先采用多孔介质方法模拟服务器内部的流动与传热过程,并建立了一个CFD简化模型,然后将实测数据与CFD模拟结果对比,验证其有效性。在此基础上,通过改变机房工况参数,建立并比较了反向传播(BP)神经网络、随机森林及极限梯度提升树(XGBoost)3种机器学习模型在不同情形下的预测性能。结果表明:BP神经网络模型具有良好的泛化能力,其预测温度时,绝对误差在0.56 ℃以内,决定系数R2在0.966以上;预测速度时,绝对误差在0.01 m/s以内,R2为0.999。
Abstract：
In order to reduce the cost of data room simulation and quickly predict the thermal environment of the data room, this paper takes an actually operating data room in a large data center in Changzhou as the research object. Firstly, the porous media method is used to simulate the flow and heat transfer process inside the server, and a simplified CFD model is established. Then, the measured data is compared with the CFD simulation results to verify its effectiveness. On this basis, by changing the working condition parameters of the data room, three machine learning models, namely back propagation (BP) neural network, random forest, and extreme gradient boosting tree (XGBoost), are established and their prediction performances under different situations are compared. The results show that the BP neural network model has good generalization ability. When predicting temperature, the absolute error is within 0.56 ℃, and the coefficient of determination R2is above 0.966. When predicting speed, the absolute error is within 0.01 m/s, and R2is 0.999.
Keywords:data center; machine learning; computational fluid dynamics(CFD); prediction model; cold aisle; thermal environment