Theoretical investigations on the purification of petroleum using desulfurization process: Analysis and optimization of process

This paper introduces a comprehensive approach for predicting chemical concentrations (C) of sulfur compound in a two-dimensional space (x, y) by numerical solution of mass transfer equation and integration of machine learning. Data of training/validation are obtained from mass transfer modeling on adsorption separation utilizing porous adsorbent for petroleum desulfurization. Mass transfer modeling data were utilized for machine learning models which included three different regression models, namely Support Vector Machine (SVM), Decision Tree Regression (DT), and Multilayer Perceptron (MLP). The hyper-parameters of these base models were optimized using the Grey Wolf Optimization (GWO) algorithm. The ensemble models, denoted as ADA-DT, ADA-MLP, and ADA-SVM were assessed based on key performance metrics, including R2, MAE, RMSE, and MAPE. Results demonstrated the efficacy of the ensemble models in capturing complex relationships within the dataset. ADA-DT obtained an exceptional R2 score of 0.99031, highlighting its outstanding predictive accuracy. Similarly, ADA-MLP and ADA-SVM showed great accuracy, achieving R2 of 0.88272 and 0.96842, respectively. In this study, we uncover valuable insights into the application of ensemble methods and hybridized optimization methods for accurate and robust regression modeling in chemical concentration prediction scenarios applicable for petroleum engineering.