Use of an artificial neural network to predict the heat transfer of supercritical R134a in a horizontal internally ribbed tube

Abstract

The heat transfer of supercritical R134a in a horizontal internally ribbed tube was predicted by using a back propagation artificial neural network (ANN). The network was trained based on 4440 experimental data points. The effects of the network input parameters, data division method, training function, transfer function, number of hidden layers, and number of neurons on the prediction results were analyzed in detail, and a new empirical formula for determining the optimal number of neurons was proposed. The prediction results by the network were then compared with those of four traditional classical correlations. The results revealed that the mean absolute errors of the ANN for predicting Nutop and Nubottom were only 35.28% and 33.03%, respectively, of those of the traditional model. Furthermore, 99.02% of Nu could be predicted with deviations smaller than 30% by the ANN, whereas only 88.7% could be predicted by traditional correlations, indicating that the ANN has a higher prediction accuracy. The present study provides a useful reference for the application and optimization of ANNs for heat transfer prediction and the design of supercritical fluid heaters.