Development of machine learning coupled one-dimensional simulation method for propane condensing flow and heat transfer in mini-channel

Abstract

LNG vaporizer with propane as the intermediate medium has unique advantages, of which the PCHE is the core heat transfer equipment. In this study, a one-dimensional simulation method was established, to achieve rapid and accurate calculation of propane condensing flow in mini-channel of PCHE. Firstly, the CFD study of propane condensing was conducted. The evolution of local h and $\Delta$ p in the channel was analyzed by micro segment method, and a database was established with multiple parameters. Then, the highest precision artificial neural network hyperparameter model was determined through comparative analysis of six machine learning algorithms. The optimal machine learning model achieved the highest R² values of 0.9887, 0.9947, and 0.9617, with the lowest root mean square error and mean absolute error. Finally, a one-dimensional simulation method is established based on the optimal machine learning model and an innovatively established vapor-liquid phase velocity ratio model. Comparing with computational fluid dynamics numerical simulation, the one-dimensional simulation only has an error of 2 %, while can improve the computational efficiency by tens of thousands of times. The new method of this research can provide a basis for the fast calculation of PCHE in LNG evaporator, and provide a new direction for the study of condensing flow in mini-channel.