Prediction of CO2 solubility in blended amine solutions using machine learning based on structural encoding

Key Words:CARBON-DIOXIDE; AQUEOUS MIXTURES; N-METHYLDIETHANOLAMINE; EQUILIBRIUM SOLUBILITY; ACID GASES; DIETHANOLAMINE; ABSORPTION; CAPTURE; MONOETHANOLAMINE; PIPERAZINE

Abstract:In this study, the prediction of CO2 solubility in blended amine is investigated using machine learning (ML) techniques. A novel feature engineering method of blended amine structure encoding (BSE) was developed and combined with four models: eXtreme Gradient Boosting (XGBoost), Random Forest (RF), Convolutional Neural Network (CNN), and Deep Neural Network (DNN). The newly developed BSE method demonstrated higher regression prediction performance compared to traditional blended amines critical property descriptors. The BSE-XGBoost model achieved an R2 of 0.984 and an MSE of 0.001, showing the best performance in predicting CO2 solubility in blended amine. Additionally, a genetic algorithm (GA) is employed to identify the optimal blending ratio for maximum CO2 solubility under specific conditions. This work highlights the potential of combining ML techniques with innovative feature engineering and optimization methods to enhance the accuracy and efficiency of CO2 solubility predictions in blended amines, providing valuable insights for industrial applications and future research in CO2 capture.

Volume:316

Issue:

Translation or Not:no