Customized Software Package for the Simultaneous Prediction of Multiple Fluidization Characteristic Parameters in Liquid–Solid Fluidized Beds

Abstract

In liquid−solid fluidized beds, there are many parameters that need to be determined for their optimization design. This study develops four machine learning models for the simultaneous prediction of the bed expansion ratio, voidage, and drag coefficient in liquid−solid fluidized beds. The Ridge regression model, K-nearest neighbor model, support vector regression, and XGBoost model are trained and tested based on the liquid−solid fluidization experimental data set. The methods of grid-search and nested cross-validation are employed for hyperparameter tuning. It is found that the XGBoost model is the best-performing model. The results of 5-fold cross-validation and Leave-One-Out confirm that there is no overfitting problem. The voidage, drag coefficient, and bed expansion ratio are all accurately predicted, and the average R 2 is 0.992, 0.976, and 0.974, respectively. Furthermore, the method of principal component analysis is employed to analyze the high-dimensional fluidization data, and the feature importance of each principal component is obtained. For convenience in use, a customized software package is developed by Python language. The software has a good graphical interface function and supports user-defined models, which can be trained using users' own data set and customized model parameters. The software is expected to benefit the preliminary design of fluidized beds.