Evaluation of CO ₂ Absorption Performance in Multiamine Blended Solvents: Combining Experimental and Machine Learning Approaches

This study investigated the CO2 absorption performance of four single-component amine solutions: monoethanolamine (MEA), diethanolamine (DEA), 2-amino-2-methyl-1-propanol (AMP), and N-methyldiethanolamine (MDEA), as well as blended solutions combining MEA with other amines at varying ratios, experimentally. Results illustrate that significant differences could be observed in absorption rates and capacities among the tested amines. MEA exhibits the fastest absorption rate, DEA achieves the highest absorption capacity, and AMP shows the greatest desorption capacity. Compared to single-component solutions, the MEA-MDEA blended solution shows the most notable performance improvement, with both maximum absorption rate and capacity significantly surpassing those of pure MDEA. Furthermore, six machine learning models─Feedforward Neural Network (FNN), Multilayer Perceptron (MLP), Long Short-Term Memory Network (LSTM), Random Forest (RF), Gradient Boosted Tree (GBT), and Support Vector Regression (SVR)─were employed for predicting CO2 absorption rates in both single and blended amine systems. The RF model demonstrated superior predictive accuracy, achieving a coefficient of determination (R2) of 0.99, root-mean-square error (RMSE) of 0.039, and mean absolute error (MAE) of 0.02. FNN and MLP models also performed well for blended amine solutions, while LSTM and SVR exhibited comparatively lower prediction efficacy.