Effect of sour gas composition on amine-based gas sweetening: A hybrid simulation-machine learning model

Natural gas, a primary fuel source for numerous industries and households, varies in composition due to factors such as petrophysical rock characteristics, origin, and thermodynamic conditions, which influence the quality of sweetened gas. This study examines the often-overlooked impact of raw natural gas composition on the quality of sweetened gas using a hybrid approach, focusing on acidity and thermodynamic features like exergy, higher heating value (HHV), and required make-ups. An Amine-Based gas sweetening unit (MDEA) gas sweetening unit was simulated to analyze process performance under varying input gas compositions and operating conditions. TabNet, ANN, Random Forest, and Genetic Algorithms (GA) were employed to develop predictive models capable of accurately analyzing trace-level parameters in the ppm range. Key input variables included methane, ethane, propane, butane, nitrogen, H 2 S , and C O 2 , alongside operational parameters such as regenerator tower reflux ratio, converter temperature and pressure, and inlet flow intensity. Simulation results demonstrated high predictive accuracy for real process outputs. GA achieved superior accuracy with R 2 > 99% for all targets and RMSE values of 9.78 e − 9 for H 2 S , 2.02 e − 6 for C O 2 , 2.92 e − 6 for amine, 69.75 for water, and 2.20 and 6.50 for HHV and exergy, respectively. Data correlation analysis revealed that sweetening performance is highly influenced by gas composition. Increased concentrations of acid gases (H 2 S, CO 2 ) lead to higher amine make-up requirements, while H 2 S reduces water make-up needs. Heavier hydrocarbons improve HHV, whereas methane decreases it, and nitrogen shows minimal impact.