Prediction of R1234yf flow boiling behavior in horizontal, vertical, and inclined tubes using machine learning techniques

Abstract

In the present study, the utilization of machine learning algorithms (MLAs) is proposed for the prediction of the heat transfer coefficient and pressure drop in horizontal, vertical, and inclined tubes during flow boiling of R1234yf. A total of 339 experimental data points sourced from the literature are employed to develop and train four methods of MLAs, including the multi-layer perceptron (MLP) neural network, support vector regression (SVR), random forest, and adaptive boosting (AdaBoost). Inclination angle, mass velocity, vapor quality, and heat flux are used as input variables, while the corresponding heat transfer coefficient and pressure drop are considered as the output variables. According to the results obtained in the prediction of the heat transfer coefficient, AdaBoost model performs the best with the mean absolute percentage error (MAPE) of 5.73 % and correlation coefficient (R) of 0.979 on the test dataset. In the case of pressure drop prediction, MLP neural network shows the best performance with MAPE of 7.62 % and R of 0.990. In addition, the remarkable effect of using machine learning methods in improving prediction accuracy is demonstrated by comparing the results of MLAs with the predictions derived from some widely recognized empirical correlations.