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 three-dimensional 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.
关键词
云计算/机器学习/计算流动动力学/气流组织温度场
Key words
Cloud computing/Machine learning calculation of flow dynamics/Airflow organization/Temperature field
维普
万方数据