Comparative analysis of heat transfer prediction for falling film evaporation on the horizontal tube based on machine learning methods

Abstract

Precise prediction of heat transfer for falling film evaporation is necessary for the horizontal tube falling film evaporator design. In this study, models to predict the heat transfer are compared, which are developed by traditional Least Square method (LSM) and Machine Learning (ML) methods, including Support Vector Regression (SVR), Artificial Neural Network (ANN) and Random Forest Regression (RFR). The training data is obtained by a series of experiments, containing 510 experimental data of fresh water and seawater of salinities from 30 to 60 g/kg. The experiments were conducted with the test tube diameter of 19, 25.4 and 38 mm and with the heat flux of 7.7–12 kW/m2. All ML models behave better performances than the LSM model and the SVR model is suggested as the best model to predict the heat transfer, considering the prediction accuracy and generalization ability comprehensively. SHapley Additive exPlanations (SHAP) is used to explain the relationships between Nusselt number and input variables, which also helps to resolve controversial issues about the effects of spray density and tube diameter on the heat transfer. The importance of variables is ranked, in which the tube pitch and seawater salinity are regarded as the most influential variables. Besides, the limitations and performance of models are analyzed in terms of both dataset and algorithms.