Analysis of Heat Transfer at Supercritical Conditions With Explainable Machine Learning Model

Abstract Explainable machine learning algorithm will be applied to provide accurate heat transfer prediction and information mining. Dimensionality analysis will be carried out at first in order to generate reasonable input features (dimensionless parameters) to the neural network. The supercritical heat transfer dataset with water as coolant is used for neural network training, in which 80% of the data set is randomly partitioned as the training set, while the remaining 20% constitutes the test set. Z-Score standardization is applied to preprocess the data on both the training and test sets. Following this, the neural network will be built to predict the Nusselt number with the particle swarm optimization algorithm using the supercritical heat transfer dataset and selected dimensionless parameters. Finally, the mechanism of heat transfer deterioration will be analyzed by applying the explainable algorithm for global and local model explanations. The study investigates the phenomenon of supercritical heat transfer excitation based on the changes in feature importance under different operating conditions.