Machine learning assisted rapid screening of multi-element doped solid-state electrolyte materials

With the rapid development of electric vehicles and energy storage systems, solid-state batteries have attracted extensive attention due to their high safety, energy density, and long cycle life. The solid-state electrolyte is the core component, but developing high-performance and stable materials remains a major challenge. This study focuses on Li 3 PS 4 and applies a multi-element co-doping strategy (Si, Ge, Sn at the P site). Combined with machine learning, we efficiently identify thermodynamically stable candidates and narrow the search scope. Our framework incorporates physically informed feature engineering, systematic hyperparameter tuning and stacking ensemble (XGBoost, LightGBM, RandomForest), SHAP analysis, uncertainty quantification (90% prediction intervals with actual coverages of 93.2% and 90.2%), and active learning simulation. Beyond the specific predictions, this work establishes a generalizable screening paradigm that integrates interpretable ML prediction, softBV-based kinetic ranking, and DFT-validated property assessment, enabling holistic evaluation of multi-component doped electrolytes. Using these methods, we rapidly screen several promising candidates with superior ion transport properties. The model achieves test R 2 = 0.879 for formation energy and 0.967 for energy above hull.