基于机器学习的气液相变换热反演及应用  

Gas-liquid Transformation Thermal Inversion Based on Machine Learning and its Application

作　　者：杨博皓 焦炜 毕景良[3] 王雷 陆规[2] YANG Bohao;JIAO Wei;BI Jingliang;WANG Lei;LU Gui(School of Mathematics and Physics,North China Electric Power University,Beijing 102206,China;School of Energy Power and Mechanical Engineering,North China Electric Power University,Beijing 102206,China;CNNC Key Laboratory on Nuclear Reactor Thermal Hydraulics Technology,Nuclear Power Institute of China,Chengdu 610213,China)

机构地区：[1]华北电力大学数理学院,北京102206 [2]华北电力大学能源动力与机械工程学院,北京102206 [3]中国核动力研究设计院,成都610213

出　　处：《工程热物理学报》2023年第12期3341-3347,共7页Journal of Engineering Thermophysics

基　　金：国家自然科学基金资助项目(No.52076074,No.12005217)。

摘　　要：相比于线性的热传导问题,气泡信息的快速获取,换热系数的预测、沸腾现象的非线性预测的问题研究目前仍极具挑战。本文使用卷积神经网络、深度前馈神经网络、经验关联式结合随机森林的数据驱动方法,进行气泡动力学行为的快速获取、相变换热系数的快速反演及沸腾特殊现象的预测。结果表明,数据驱动的机器学习方法能够快速准确地预测相变换热过程中的问题,气泡信息获取准确率达到96%,相变换热系数反演及特殊现象预测拟合优度均高于0.95。本研究可为相变换热反演问题提供新的解决思路。Compared with the linear heat conduction problem,the rapid acquisition of bubble information,the prediction of heat transfer coefficient,and the nonlinear prediction of boiling phenomenon are still very challenging.In this paper,the data-driven methods of ensemble learning,deep feedforward neural network,empirical correlation combined with random forest are used to quickly obtain bubble dynamic behavior,quickly invert the thermal coefficient of phase transformation,and predict the special phenomenon of boiling.The findings suggest that the data-driven machine learning method can quickly and accurately predict the problems in the process of phase-change heat transfer.The accuracy rate of acquiring bubble information reaches 96%,and the goodness of fit of phase transformation thermal coefficient inversion and prediction of special phenomenon is higher than 0.95.This study can provide a new solution for the inversion of phase-change heat transfer.

关 键 词：相变换热 机器学习 气泡动力学 热工水力参数 

分 类 号：TK121[动力工程及工程热物理—工程热物理]