ANN model with feature selection to predict turbulent heat transfer characteristics of supercritical fluids: Take CO2 and H2O as examples

Abstract

Heat transfer deterioration of supercritical fluids is difficult to predict accurately due to the numerous influencing factors. Traditional correlations failed to fit the relationships between many parameters over a wide range of experimental conditions. To this end, we developed a general artificial neural network model for simultaneously predicting the Tw and Nu of supercritical H2O and CO2, which is the first model currently available for multi-fluid prediction. To maintain the compactness of the model feature space, redundant features are eliminated by feature selection. This model was trained with circle tube data but still maintained high accuracy on the non-circle tube data, demonstrating its strong generalization and independence of channel shape. High accuracy, applicability, and convenience over a wide range of applications are embraced in the ANN model, which lays a solid foundation for engineering application of the ANN model. The generalizability of the ANN model could be further improved by incorporating other fluid data.