Recovery of Brine Resources Through Crown-Passivated Graphene, Silicene, and Boron Nitride Nanosheets Based on Machine-Learning Structural Predictions

The rising global demand for brine resources necessitates the exploration of alternative sources to complement existing natural sources. It is imperative to explore innovative approaches, such as emerging machine learning-aided tools, to ensure sustainable and secure brine resources. We proposed a kernel support vector regression (k-SVR), and Gaussian process regression (GPR) based on several feature engineering selectivity approaches that were employed for modeling adsorption energy (kcal/mol). For this purpose, two different scenarios of crown-embedded 2D materials using first-principles density functional theory simulations were obtained. Subsequently, ensemble machine learning (ML) was employed to improve the accuracy of prediction skills of the 2D materials. The data for the successful creation of ion transmission channels utilizing 9-crown-3 (distance within cavity O 9 –O 6 = 3.105 Å, O 3 –O 6 = 2.934 Å, O 3 –O 9 = 2.961 Å) and 12-crown-4 (distance within cavity O 3 –O 9 = 4.538 Å, O 6 –O 12 = 3.223 Å) molecules on graphene, hexagonal boron nitride, and silicene nanosheets were used in this study. The predictive results proved that GPR-C1 with a numerical comparison of RMSE = 0.096, NSE = 0.9610 in the training phase and RMSE = 0.6630, NSE = 0.911 in the testing phase outperformed the other model combinations. The study also proposed federated learning for reliable modeling and recovery of complex and poor selectivity of brine resources. The present study utilizes ML algorithms to provide insights into brine resource recovery, which contributes to multiple sustainable development goals, addressing environmental, economic, and social dimensions of sustainable development.