Study on artificial neural network prediction of heat transfer in vertically upward internally rifled tubes with supercritical water

Abstract

This paper presents an ANN model for heat transfer prediction of heat transfer in vertically upward internally rifled tubes with supercritical water. The model is trained using a dataset of 2071 experimental data points on supercritical heat transfer in internally rifled tubes. Dimensionless number of rib geometry associated with rib geometry are incorporated into the internally rifled tube structure, and optimization is performed on the input features and structure of the neural network. The obtained neural network model demonstrates accurate heat transfer predictions performance with overall test set R² = 0.9975 (coefficients of determination), RMSE = 0.2353 (root mean square error), and MRE = 1.65 % (mean relative error). The model has a prediction error range of 2 % for wall temperature, respectively. Furthermore, the neural network model exhibits superior accuracy in heat transfer predictions compared to several typical empirical correlations. This study provides a more extensive and accurate approach for predicting heat transfer in supercritical water.