Analysis of the Coupled Variations of Holdup and Droplet Diameter Based on Machine Learning Methods

Abstract

Due to limited research on holdup-droplet diameter coupled variations and poor predictive performance of machine learning (ML) models on small datasets, this study integrates LASSO, support vector regression (SVR), and random forest regression (RFR) to develop a parameter analysis algorithm. The algorithm features generalization for small cross-system datasets and parameter physical significance analysis capability. Applying it, we explored the combined effects of 14 experimental operating parameters on holdup and Sauter mean diameter, with the model's prediction error for their coupled variations ≤10%. It provides a new perspective for understanding hydrodynamic characteristics in pulsed extraction columns, guides key process equipment design and operational optimization, and lays a theoretical foundation for engineering applications of related ML methods.