Modeling convective heat transfer of supercritical carbon dioxide using an artificial neural network

Supercritical carbon dioxide (sCO(2)) is an ideal working fluid for an energy conversion system, which can be used in a wide variety of power-generation applications with increased efficiency. However, abrupt variations in its thermo-physical properties in pseudo-critical region make it difficult to accurately predict the heat transfer characteristics of sCO(2) using traditional analyzing methods, which to some extent hinders the development of sCO(2) power conversion technology. Therefore, it is critical to explore feasible approaches to accurately predict the characteristics of sCO(2) heat transfer. In this study, the performance of representative empirical correlations of sCO(2) heat transfer has been assessed with a databank collected from previous publications. The results of the assessment indicate that the existing correlations are not effective enough to describe sCO(2) heat transfer, especially when it is in the pseudo-critical region. An artificial neural network (ANN) is therefore proposed to model sCO(2) heat transfer with experimental datasets. The ANN model shows a great learning ability and satisfactory generalization performance with a correlation coefficient of 0.99 and a mean absolute percent error of 0.97% in the test dataset. The results show that the proposed ANN model is a more effective and efficient method to predict the heat transfer characteristics of sCO(2) than empirical correlations. The feasibility of the ANN model in the prediction of heat transfer with significant buoyancy force or flow acceleration are also tested and discussed.