Machine learning based thermodynamic modelling of acid gas absorption in aqueous MDEA and aqueous Piperazine

Acid gas chemical absorption with aqueous amine solvents is an important industrial technology for gas processing and CO2 capture. Vapor-Liquid Equilibrium (VLE) reflects the efficiency of solvents and is essential to model the thermodynamics of the absorption and solvent regeneration process. In this study, several machine learning (ML) approaches were used to develop VLE models for acid gas absorption in aqueous methyldiethanolamine (MDEA) and piperazine (Pz), namely CO2-MDEA-H2O, H2S-MDEA-H2O, CO2-H2S-MDEA-H2O and CO2-Pz-H2O systems. New experimental data are presented for the CO2-MDEA-H2O and H2S-MDEA-H2O ternary systems, and they are used to compare the accuracy of the ML models to an earlier reported activity coefficient (e-NRTL)-based thermodynamic model. For the quaternary system CO2-H2S-MDEA-H2O, the ML models and the physical model are compared using experimental data from literature, because the physical model of the quaternary system is only trained on the ternary experimental systems. The results indicate that the optimal ML approach is not systematically more accurate than the physics-based model.