Automated determination of optimal component design for a binary solvent for absorption-based acid gas removal

Abstract

Abstract. Natural gases containing impurities, namely carbon dioxide (CO2), heavy hydrocarbons, hydrogen sulfide (H2S), and water vapour, need treatment for removing acidic gases (CO2 and H2S) to reduce corrosion and enhance the heat capacity of the gas. This gas is commercially known as "sour", and typically, sour gas is any gas that contains significant levels of hydrogen sulfide. The presence of carbon dioxide can affect natural gas quality, which can also lead to CO2 freezing issues; hence reliable techniques for reducing CO2 and H2S from natural gases is necessary. New blends of amines show CO2 and H2S uptake capacity comparable to traditional MEA benchmark solutions. This work aimed to create different regression models using open-source software and estimate the best fit model for a given amine solvent. For this purpose, data were obtained from simulation using Aspen HYSYS V12.1 for MDEA (40-45 wt.%), MDEA +PZ (42-50wt.% + 0-2.5wt.%), DEA (21-26wt.%). Regression models for different amine solvent blends were developed and validated. The study showed that the XGB Regression model was best suited for the MDEA solution, while MDEA + PZ and DEA were best suited for multiple linear regression. The data is generated using simulation from ASPEN HYSYS and models were created in python correlating the simulation-generated values with the model results. These models showed low MSE, RMSE and high R2 values for the tried solvents.