Development and optimization of lithium-ion sieves through machine learning in complex brine systems

Lithium ion sieves and their composite materials show great potential in lithium extraction due to their high selectivity and adsorption capacity. However, their performance varies significantly under different conditions due to complex ion interactions and environmental factors, making process optimization challenging with traditional experimental methods. A comprehensive machine learning approach was developed using 12 key material and environmental parameters to predict lithium-ion sieves adsorption performance. Among nine machine learning methods evaluated, the optimized XGBoost model demonstrated superior predictive capability (R² = 0.9736 for training, R² = 0.8916 for testing) in handling the inherent nonlinearity of adsorption systems. SHapley Additive exPlanations analysis revealed the dominant influence of environmental parameters on adsorption behavior, providing mechanistic insights into the adsorption process. The reliability of the model was validated through 56 independent experiments with three different composite adsorbents (R² = 0.7696). A Python-based prediction system was developed for rapid performance assessment. This data-driven strategy establishes a rational framework for optimizing process parameters and developing materials, thereby advancing the industrialization of sustainable lithium extraction technologies.