Prediction of supercritical CO2 flow and heat transfer behaviors in zigzag-type printed circuit heat exchangers by improved POD-GABP reduced order model

Printed circuit heat exchangers (PCHE) are widely applied and investigated in supercritical carbon dioxide (SCO2) power cycles. However, it is computationally expensive to obtain the physical filed in PCHEs by numerical simulations due to the high dimensionality, strong nonlinearity and the strong coupling of the flow channel structure and thermophysical properties of the working fluids. To improve the computational efficiency of the local physical field, with the preservation of acceptable accuracy. This paper proposes a nonintrusive reduced order model that combines Proper orthogonal decomposition (POD) and Genetic algorithm-optimized BP neural network (GABP) to predict the flow and heat transfer performance of S-CO2 in PCHE channels. In the offline stage, the POD is performed on high-dimensional physical fields data for dimension reduction and feature capture and a GABP model was used to approximate the coefficients of the reduced model. In the online stage, physical fields under new working conditions can be quickly predicted using the established model. Fifty groups of test samples were used to assess the performance of this framework. The results show that it achieves good accuracy with maximum L2 norm error of 2.16 %. At the same time, the degree of freedom was reduced by 104 and model efficiency was accelerated by nearly three orders of magnitude compared with traditional CFD simulation. Besides, the maximum prediction errors of the calculated heat transfer coefficients heat transfer and fraction coefficients based on the forecasted physical field and for the hh, hc,fh and fc are 8.1 %, 15.7 %,10.5 % and 14.4 %, respectively. The present research may provide new and more effective approach to model flow and heat transfer of S-CO2 in PCHEs under varying working conditions.