Research
| Title： | Process optimization and decoupling multi-parameter interaction in ion adsorption rare earth ore leaching via a machine learning approach |
|---|---|
| First author： | Liu, Jingxin; Lyu, Huafei; Cheng, Can; Peng, Yuesong; Guo, Xiudeng; Yan, Kang |
| Journal： | JOURNAL OF CLEANER PRODUCTION |
| Years： | 2025 |
| DOI： | 10.1016/j.jclepro.2025.146560 |
| Abstract： | The extraction of ion adsorption rare earth ore is crucial for the advancement of modern technologies, however, the leaching process is often resource-intensive, with significant environmental impacts. Traditional methods for process optimization are time-consuming and require extensive trial-and-error experimentation, which can be costly and inefficient. In this study, extensive historical data were collected from literature, and a dataset consisted of 2365 sets of data involving 14 input variables and 1 output variable was constructed after careful selection. Subsequently, 6 machine learning algorithms were trained, tuned, and compared. The optimized gradient boosting regressor model with RMSE of 7.874 % and R2 of 0.9063 for the test set was deemed robust and reliable. Leaching time was identified as the most influential factor on the extraction efficiency of rare earth oxides, followed by the concentration of lixiviant I, leaching method, pH value, and concentration of lixiviant II. Furthermore, partial dependence plots quantitatively decoupled the interaction between multiple parameters and proposed optimization strategies, overcoming the limitations of conventional single-variable experimental approaches. The development of this machine learning framework represents a methodological advancement in data-driven approaches for resource extraction and industrial process optimization. |
