Exploring the specific heat capacity of aqueous blends of K₂CO₃- PZ-MEA in CO₂ capture using ANN and RSM models

In this work, both artificial neural network (ANN) and response surface methodology (RSM) approaches were employed to estimate the specific heat capacity (C_P) of aqueous mixtures containing potassium carbonate (K₂CO₃), piperazine (PZ), and monoethanolamine (MEA) used in CO₂ capture processes. The multilayer perceptron (MLP) model was employed using the trainlm and trainbr functions for C_P prediction. The optimal model, trained using the trainlm algorithm with a hidden layer architecture consisting of 15 neurons in the first layer, 15 in the second, and 10 in the third, reached its best performance after 79 epochs, achieving a remarkably low mean squared error (MSE) of 0.0001 and a high R² value of 0.9993. In the radial basis function (RBF) approach, the best prediction was obtained with a spread value of 3 and 150 neurons, resulting in mean squared error of 0.0040 and R² value of 0.9858. A comparison of MSE and R² values for RSM and ANN indicates that the results from ANN, especially the MLP method, are more accurate than those from RSM. A total of 527 experimental data points, collected from various sources, were utilized to comprehensively predict C_P for different aqueous solution systems containing MEA, PZ, and K₂CO₃. The temperature range considered was from 313.15 to 393.15 K. Additionally, to improve prediction accuracy, physical properties of the solutions, such as density and molecular weight, along with temperature, were selected as input parameters. The results can effectively contribute to optimizing CO₂ absorption processes and designing aqueous-based absorption systems.