In this work, 2099 experimental data of binary systems composed of CO₂ and ionic liquids are studied to predict solubility using a multilayer perceptron. The dataset includes 33 different types of ionic liquids over a wide range of temperatures, pressures, and solubilities. The main objective of this work is to propose a procedure for the prediction of CO₂ solubility in ionic liquids by establishing four stages to determine the model parameters: (1) selection of the learning algorithm, (2) optimization of the first hidden layer, (3) optimization of the second hidden layer, and (4) selection of the input combination. In this study, a bound is set on the number of model parameters: the number of model parameters must be less than the amount of predicted data. Eight different learning algorithms with (4,m,n,1)-type hidden two-layer architectures (m = 2, 4, …, 10 and n = 2, 3, …, 10) are studied, and the artificial neural network is trained with three input combinations with three combinations of thermodynamic variables such as temperature (T), pressure (P), critical temperature (Tc), critical pressure, the critical compressibility factor (Zc), and the acentric factor (ω). The results show that the 4-6-8-1 architecture with the input combination T-P-Tc-Pc and the Levenberg–Marquard learning algorithm is a very acceptable and simple model (95 parameters) with the best prediction and a maximum absolute deviation close to 10%.

The solubilities of 498 datasets of N₂O and ionic liquid systems were predicted using a multilayer perceptron. The data used to train the artificial neural network was subjected to the Gibbs–Duhem test to analyze their thermodynamic consistency. The Peng–Robinson cubic equation of state, combined with the Kwak–Mansoori mixing rule, was used as the thermodynamic model to implement the test. The analysis indicated that 71.9% of the data were declared thermodynamically inconsistent. The ability of artificial neural networks (ANNs) to predict the solubility of these systems using experimental datasets that do not satisfy the thermodynamic consistency criteria based on the Gibbs–Duhem equation was studied. The multilayer perceptron model achieved an average absolute deviation of 1.81% and a maximum individual deviation of 7.56%. These results highlight the potential of ANNs as robust predictive tools even when the available data do not fully satisfy thermodynamic consistency criteria.