چكيده به لاتين
In the present study, through exploring recent studies in this field, experiments of CO¬2¬ absorption using amine-based (e.g., primary, secondary, tertiary, sterically hindered) solutions with promoters (such as functionalized amines, ionic liquids, and physical absorbents) are reviewed, and their performance is compared. Afterward, due to this study's goal, which is to utilize neural networks due to their accuracy and speed, a modeling method with neural networks (MLP and RBF) is used to predict CO¬2 ¬absorption in amine mixtures with promoters using experimental data from the literature. Also, MLP and RBF neural networks are used to predict CO2 absorption in the aforementioned solvents in a collected dataset. For this purpose, several related studies with similar experimental conditions were chosen, from which ¬1073 ¬collection of data was selected for the modeling. This dataset was further divided into training and test divisions, and numerous networks were created and explored regarding the number of layers, neurons, and type of activation functions. Temperature, mass percent concentration of components of solutions (excluding water), the apparent molecular weight of the solution, and partial pressure of CO2 were selected as inputs, and CO2 loading was assigned as the output of the ANN. MSE, AARD%, and R2 were the criteria for choosing the most suitable network. MSE, AARD%, and R2 for the selected MLP network were ¬0.000046, 0.67, ¬and ¬0.999, ¬respectively. In addition, these parameters for RBF were -3.61¬E-¬05, 0.47, ¬and ¬0.998, ¬respectively. It can be concluded that these networks have predicted the results of actual experiments with high accuracy. The best MLP network has two hidden layers with ¬20¬ and ¬10¬ neurons, respectively, utilizing L-M for training, tansig and purelin activation functions for hidden layers and output, respectively. Also, the best RBF network was created using a spread of ¬2.2 ¬with ¬700 ¬neurons (epochs) using Gaussian activation function.
