Predicting the Effect of Cosolvents on CO₂ Solubility in Choline Chloride-Based DESs Using Artificial Neural Networks

Deep eutectic solvents (DESs) are increasingly recognized as environmentally friendly alternatives for CO₂ capture in a wide range of industrial processes. Their low vapor pressure, tunable composition, and potential biodegradability make them attractive candidates for gas absorption applications. However, a major limitation of them is their inherently high viscosity, which can restrict mass transfer and practical operability. To mitigate this drawback, cosolvents, such as water and amine-based compounds, are commonly added to reduce viscosity; however, their addition can significantly alter the physicochemical properties, including CO₂ solubility of the system, making reliable predictive models essential. In this study, an artificial neural network (ANN) model is developed to predict CO₂ solubility in choline chloride-based DES systems combined with different cosolvents, covering both binary and ternary mixtures. The focus on choline chloride-based DESs is motivated by their widespread use in CO₂ capture studies. The database comprises approximately 1200 experimental solubility data points over a broad range of compositions, temperatures, pressures, and cosolvent types. Molecular descriptors derived from COSMO-SAC σ-profiles are employed as ANN inputs, enabling a molecular-level description of the solvent systems. Several ANN architectures are evaluated to identify the most suitable network configuration. Statistical analyses are performed to assess model robustness and prevent overfitting. Additionally, the extrapolation capability of the ANN is tested by predicting CO₂ solubility in quaternary systems not included in the training dataset. Preliminary results indicate a broad applicability domain and reliable predictive performance, highlighting the potential of ANN-based models as open-source tools for screening and designing DES-based solvents for CO₂ capture.