Machine learning and prediction study on heat transfer of supercritical CO2 in pseudo-critical zone

The utilization of supercritical carbon dioxide(S-CO2) as a working fluid in energy conversion systems has gained widespread recognition as an efficient and environmentally friendly option. However, accurately predicting the heat transfer process is still challenging due to the significant variation of thermophysical properties within the pseudo-critical zone. The accurate prediction of the S-CO2 heat transfer process is of utmost importance for the design of heat exchangers and the safe operation of the system. Aiming at the current problems of high experimental cost and long numerical simulation time, machine learning is adopted in this paper to predict the heat transfer characteristics of S-CO2 in this temperature region. In this paper, the heat transfer process of S-CO2 flowing upward in a circular tube under heating conditions is taken as the research object, and a total of 11,032 sets of experimental data samples in the open literature are collected. Four machine learning models, namely, Random Forest (RF), Extreme Gradient Boosting (XGBoost), Support Vector Machines (SVR), and Artificial Neural Networks (ANN) are trained by taking mass flow rate, wall heat flux, pressure, fluid enthalpy, and tube diameter as the input parameters, and wall temperature as the output parameter. The prediction performance of the four machine learning models and the heat transfer correlations were compared. The results show that all four machine learning models have excellent prediction performance, and the ANN model provides the best prediction performance, with an R2 of 0.995 on new data. XGboost and ANN can accurately predict the heat transfer deterioration when the fluid temperature (Tb) approaches the pseudo-critical temperature (Tpc) or over Tpc, yet the accuracy decreases in the region of Tb < Tpc, suggesting that the prediction error is mainly originated in this region. Compared with the existing heat transfer correlations, the prediction accuracy of the ANN model obtained from the training in this paper is higher. The present study further elucidated the feasibility and accuracy of utilizing an ANN model for predicting the S-CO2 heat transfer process. A trained ANN model is a useful tool that can be directly applied to system design and heat exchanger design.