Rapid Prediction Methods And Applications Of Thermal Hydraulic Parameters In Boiling Heat Transfer Based On Machine Learning

Boiling of water is very common in our daily life and it is also an efficient heat transfer mechanism,and boiling is used in various high energy density engineering systems for heat management and energy conversion processes.From the cooling of power electronics to power plants and especially in the heat exchangers of nuclear reactors,boiling heat exchange is everywhere.The bubbles during boiling are disorganised,the boiling parameters are difficult to measure and the boiling crisis can easily occur during boiling,which can lead to instantaneous burn-out of the heat exchange equipment and cause safety accidents.For engineering applications,boiling heat transfer related parameters are usually predicted by empirical correlation or semiempirical models,which are time consuming and error prone.With the rapid development of artificial intelligence,machine learning has been gradually applied to boiling heat transfer,but it suffers from unstable prediction results,lack of extrapolation capability and poor interpretability.Compared to linear heat transfer problem,the researches on the rapid acquisition of boiling bubble dynamics information,the rapid prediction of boiling parameters,and the prediction of critical heat flux(Critical Heat Flux,CHF)density in boiling crisis are still extremely challenging.This paper explores the application of machine learning algorithms to the above three key feature problems in boiling heat transfer,and improves machine learning methods applied to different thermal hydraulic parameter predictions.The main research contents and results of this paper are as follows.(1)For the study of boiling bubble dynamics,the images are processed by different feature extraction methods,and the flow patterns are identified by different individual machine learning methods.By comparing the prediction effect of different machine learning methods and applying integrated learning algorithms,the prediction performance of individual learner and integrated learning is compared.A mixture of different machine learning models is integrated and optimized by using integrated learning ideas,and the results of the optimized model shows that both prediction efficiency and accuracy are taken into account at the same time.(2)For the rapid prediction of boiling parameters,the deep feed-forward neural network is used to predict both the heat transfer coefficient and the gas content rate.And different machine learning models are compared according to the mean square error(MSE),mean absolute error(MAE)and R2 goodness of fit.The results show that the deep feed-forward neural network predicts the boiling parameters best.The extrapolation capability of the model to predict the heat transfer coefficient is verified,and the variation of the heat transfer coefficient with physical characteristics is discussed by using the trained model.(3)For the prediction problem of boiling crisis CHF,based on physical parameters and physical cloud maps in narrow rectangular channel boiling CHF simulation,the hybrid framework of empirical correlation and machine learning model is adopted to predict CHF.This hybrid approach exploits the empirical understanding of physics and uses machine learning to capture deeper information between the actual target and empirical knowledge,making the results physically interpretable.Comparing the single machine learning approach with the hybrid framework,the results show that the hybrid framework combined with empirical correlations has better prediction performance and more stable prediction results.This paper investigates the application of machine learning to the rapid prediction of boiling bubble dynamics,boiling parameters and boiling crisis CHF,while considering the prediction speed and accuracy,providing some reference value for industry.