Study on artificial neural network-based prediction of thermal characteristics of supercritical CO2 in vertical channels

Abstract

The aim of this paper is to achieve an accurate prediction of heat transfer deterioration and Nu of supercritical CO2 upward flowing in vertical channels using the artificial neural network method. Firstly, based on extensive experiments of round tubes and the square annular channel, a database with 11,589 data in 431 sets of experimental conditions was established. Then ANN models were trained with the training set which only contains the round tube data, and the optimal model structures were determined. The effects of input features on the prediction performance, including the experimental parameters, the dimensionless parameters, and the thermophysical property ratios, were compared, and the results showed that the model with all features in the input parameters performed best. The test set contains the experimental data of the round tubes and the square annular channel, the prediction results of the models on the test set showed that the model for HTD only has high precision on the round tube data, while the model for Nu has precise prediction performance both on the data of round tubes and the square annular channel. Simultaneously, the model for Nu shows a higher prediction performance than the correlations.