Study on prediction model of liquid hold up based on random forest algorithm.

• A dataset containing 2755 sets of experimental data is established. • Based on the data set, a random forest machine learning model is established. • Based on data statistics, the application scope of the model is analyzed. • The model's accuracy is verified by comparing the conventional model with the experimental data. The liquid holdup prediction model is a mainly empirical formula derived from researchers' experimental data. These models have poor generalization due to the limitations of the experimental condition. In this study, the liquid holdup machine-learning model is developed based on a random forest algorithm using 2755 groups of experimental data in various conditions. Through the detailed hyper parametric model optimization, and combined grid-search, the model's optimal parameters are obtained to prevent over-fitting. The determination coefficient of the model on the training set and the test set is high and roughly the same, indicating that the model has high accuracy and has not been over-fitted. The model is validated with published independent data and compared with other widely used empirical models. The results show that the model in this study has higher generalization and can accurately predict the liquid holdup under the various conditions of pipe diameters, inclination angles, gas–liquid velocity, temperatures, and pressures within the applicable scope of the model. [ABSTRACT FROM AUTHOR]