New insights into lithium recovery from unconventional water resources

ABSTRACT The recovery of lithium from unconventional water resources is a complex physicochemical process influenced by numerous non-linear variables, making the optimization of adsorption materials through traditional experimental methods time-consuming and costly. This study aims to develop a robust data-driven framework to accurately predict the lithium adsorption capacity of lithium-ion sieve (LIS) materials by leveraging historical experimental data. To achieve this, a comprehensive dataset comprising 855 data points was constructed from peer-reviewed literature, incorporating sixteen input features representing water chemistry, material properties, and operational conditions. Fifteen distinct machine learning algorithms, ranging from linear regressors to advanced ensemble methods and deep neural networks, were developed, hyperparameter-tuned, and rigorously evaluated using statistical metrics such as the Coefficient of Determination (R 2 ) and Mean Relative Deviation (MRD%). The results demonstrated a stark contrast in performance, where linear models failed to capture the underlying process mechanisms (R 2 <0.26), while tree-based ensemble methods exhibited superior predictive capabilities. The Gradient Boosting regressor emerged as the optimal model, achieving a testing R 2 of 0.9896 and a low MRD% of 14.2%, with minimal overfitting compared to single Decision Trees or XGBoost. Furthermore, SHAP analysis revealed that initial lithium concentration is the primary driver of adsorption efficiency, followed by contact time and temperature, while competing ions showed negligible interference. In conclusion, this study validates that gradient-boosted machine learning models can effectively replace trial-and-error experimentation, providing a precise and scalable tool for designing high-efficiency lithium extraction processes.

Hali tarjima qilinmagan