基于可解释机器学习的超临界流体传热特性预测与分析  

Prediction and Analysis of Heat Transfer Characteristics of Supercritical Fluids Based on Interpretable Machine Learning

作　　者：李浩哲 宋美琪 刘晓晶[1,3] Li Haozhe;Song Meiqi;Liu Xiaojing(School of Mechanical Engineering,Shanghai Jiao Tong University,Shanghai,200240,China;College of Smart Energy,Shanghai Jiao Tong University,Shanghai,200240,China;Shanghai Numerical Nuclear Reactor Technology Integration Innovation Center,Shanghai,200240,China)

机构地区：[1]上海交通大学机械与动力工程学院,上海200240 [2]上海交通大学国家电投智慧能源创新学院,上海200240 [3]上海市数值反应堆技术融合创新中心,上海200240

出　　处：《核动力工程》2024年第6期63-74,共12页Nuclear Power Engineering

基　　金：国家自然科学基金青年科学基金项目(12305189)。

摘　　要：超临界流体在拟临界温度附近发生剧烈的物性变化,传热特性难以准确预测。本研究采用可解释机器学习的研究方法预测并分析超临界流体传热特性。使用粒子群优化算法(PSO)搜索反向传播神经网络(BPNN)模型最优超参数,建立了超临界流体传热预测模型,并与传统经验关联式进行了精度比较。使用SHAP可解释算法对BPNN模型进行了全局和局部解释,根据不同工况下特征重要度的变化发现超临界相关机理现象。结果显示,所建立的神经网络模型在测试集上的平均绝对百分比误差(MAPE)为1.4%,决定系数R^(2)为0.9992,与经验关联式相比,该模型具有更高的预测精度;对于垂直向上流动,浮升力效应在传热恶化工况中明显具有更高的特征重要度,是传热恶化行为发生的主要因素。因此,本研究建立的基于可解释机器学习的研究方法对进一步研究超临界流体传热特性具有一定的参考意义。The physical properties of supercritical fluids change drastically near the pseudo-critical temperature,making it challenging to accurately predict heat transfer characteristics.In this study,the method of interpretable machine learning was used to predict and analyze the heat transfer characteristics of supercritical fluids.The particle swarm optimization algorithm(PSO)was used to search for the optimal hyperparameters of the back propagation neural network(BPNN)model,the supercritical fluid heat transfer prediction model was established,and its accuracy was compared with the traditional empirical correlation.The global and local interpretation of the BPNN model was carried out by using the SHAP interpretable algorithm,and the supercritical correlation phenomenon and mechanism were found according to the change of feature importance under different conditions.The results show that the MAPE of the established neural network model on 2 the test set is 1.4%,and the coefficient of determination R is 0.9992,which has higher prediction accuracy compared with the empirical correlation formula.For vertical upward flow,buoyancy effect obviously has higher feature importance in heat transfer deterioration condition,which is the main factor of heat transfer deterioration behavior.Therefore,the research method based on interpretable machine learning established in this study has certain reference significance for further study of the heat transfer characteristics of supercritical fluids.

关 键 词：超临界流体 可解释机器学习 传热预测 

分 类 号：TK124[动力工程及工程热物理—工程热物理] TL331[动力工程及工程热物理—热能工程]