Analysis of CO2 solubility in ionic liquids as promising absorbents using response surface methodology and machine learning

Abstract

This study explored CO₂ solubility in ionic liquids using Response Surface Methodology (RSM) and Artificial Neural Networks (ANNs), specifically Multilayer Perceptron (MLP) and Radial Basis Function (RBF) networks, to model and optimize absorption processes. In our study, we analyzed several ionic liquids (ILs), including [bmim][PF6], [P(5) mpyrr][Tf2N], [mp(3)pip][FSI], [mp(3)pyrr][FSI], [N1223][FSI], [bmmim][Tf2N], and [P4441][Tf2N], chosen for their unique properties such as high thermal stability and ionic conductivity. Experimental data analysis identified mass, viscosity, pressure, molar concentration, and surface tension as key influencing parameters. RSM demonstrated excellent fit with an R² of 0.9999, while ANNs exhibited superior predictive accuracy, with R² values approaching unity. The MLP network, employing a two-layer training activation function, achieved a minimum Mean Squared Error (MSE) of 0.001082 for test data. The RBF network with 26 neurons and a spread of 2 reached a minimum MSE of 0.0011252. 3D response surface analyses of MLP, RBF, and RSM revealed intricate parameter interdependencies. Increased ionic liquid mass enhances CO₂ absorption by expanding the interaction space and providing more binding sites. Elevated pressure significantly increases solubility by compressing the gas phase and driving more CO₂ molecules into the liquid. Higher viscosity impedes CO₂ movement within the liquid, while lower viscosity facilitates faster diffusion. A higher molar concentration of CO₂ in the gas phase increases the driving force for absorption, leading to a greater influx of CO₂ into the ionic liquid. While RSM surfaces exhibited rigid, polynomial-based trends, the smoother MLP plots effectively captured complex nonlinearities, highlighting ANNs' superior predictive capabilities for optimizing CO₂ capture systems in ionic liquids.