基于机器学习方法的微通道热沉性能预测研究  

Investigation on the Performance Prediction of Microchannel Heat Sink Based on Machine Learning Approach

作　　者：杨敏 刘源斌 于新刚[1] 苗建印[1] 车邦祥 马静 曹炳阳[2] YANG Min;LIU Yuanbin;YU Xingang;MIAO Jianyin;CHE Bangxiang;MA Jing;CAO Bingyang(Beijing Key Laboratory of Space Thermal Control Technology,Beijing Institute of Spacecraft System Engineering,China Academy of Space Technology,Beijing 100094,China;Key Laboratory for Thermal Science and Power Engineering of Ministry of Education,School of Aerospace Engineering,Tsinghua University,Beijing 100084,China)

机构地区：[1]北京空间飞行器总体设计部,空间热控技术北京市重点实验室,北京100094 [2]清华大学航天航空学院,热科学与动力工程教育部重点实验室,北京100084

出　　处：《工程热物理学报》2023年第6期1704-1708,共5页Journal of Engineering Thermophysics

基　　金：国家自然科学基金(No.51825601,No.51676108,No.51621062)。

摘　　要：本文提出使用机器学习方法快速准确地预测歧管-二次流混合结构微通道热沉的泵功率和总热阻。将混合结构微通道热沉的结构特征参数进行了无量纲化,利用计算流体动力学的方法获得数据集。测试了不同机器学习算法在混合结构热沉性能预测任务上的表现。结果表明在数据集有限的情况下,随机森林算法能准确地学习到无量纲结构参数与泵功率和总热阻之间的映射关系。本文研究结果将有助于微通道热沉的优化设计。A machine learning method was proposed to rapidly and accurately predict the pumping power and the total thermal resistance of a hybrid microchannel heat sink using manifold arrangement and secondary channels.A nondimensionalization of the structure parameters for the hybrid heat sink was performed,and the data set was obtained by computational fuid dynamic method.The performance of different machine learning algorithms in predicting the heat sink performance was tested.The results indicate that the random forest(RF)is able to accurately learn the mapping relationship between those dimensionless structural parameters and the pumping power or the total thermal resistance with limited data samples.Our study will be helpful for design optimization of microchannel heat sink.

关 键 词：微通道热沉 性能预测 机器学习 随机森林 

分 类 号：TK121[动力工程及工程热物理—工程热物理]