Optimization of CO₂ absorption into MDEA-PZ-sulfolane hybrid solution using machine learning algorithms and RSM

This study presents the modeling and simulation of carbon dioxide (CO₂) absorption in hybrid amine solutions using machine learning algorithms and response surface methodology (RSM). The process was governed by adjustable input parameters, including pressure (0.50-7.76 bar), temperature (292.8-343.1 K), time (0-1680 s), and solvent concentrations (1-5 wt% for N-methyl diethanolamine (MDEA), 1-5 wt% for sulfolane, and 1-5 wt% for piperazine (PZ)). The primary objective was to leverage the synergistic effects of chemical and physical solvents for enhanced CO₂ absorption. Seven machine learning models-MLP, RBF, LightGBM, XGBoost, Random Forest, ExtraTrees, and Adaboost-were employed for accurate prediction and parametric analysis. Among these, MLP with 147 neurons and RBF with 560 neurons demonstrated superior performance, achieving R² values of 0.9982 and 0.9975, respectively. A comparative analysis with RSM confirmed that neural networks exhibited superior predictive accuracy and generalization. The findings revealed that CO₂ absorption capacity increased with rising pressure and time but decreased with higher temperatures and solvent concentrations. Additionally, optimization through a genetic algorithm (GA) was employed to identify the best input parameters for maximizing CO₂ loading, achieving optimal conditions with a CO₂ loading of 0.62, 0.55, and 0.50 for RSM, MLP + GA, and RBF + GA, respectively.