Knowledge-driven eutectic electrolyte design for Zn-ion batteries

Eutectic electrolytes are considered as promising candidates for Zn-ion batteries (ZIBs) due to the wide liquid range, broad electrochemical window, high safety, high tunability, and low cost. However, current design of eutectic electrolytes primarily relies on trial-and-error methods, which are time-consuming and costly due to their complex structure and compositions. To address these challenges, we propose an ionic conductivity prediction method of eutectic electrolytes through data-driven machine learning. To improve model interpretability, we construct domain knowledge-driven features based on the physicochemical properties of electrolyte components. By integrating domain knowledge, the Categorical Boosting (CatBoost) model achieves high accurate predictions of ionic conductivity. Moreover, feature contribution quantified by the Shapley Additive exPlanations (SHAP) identifies donor number of anion and solvent as well as dipole moment of H2O as most significant descriptors affecting ionic conductivity. This understanding offers a fundamental design principle for developing high-ionic-conductivity eutectic electrolytes. The eutectic electrolytes designed with our machine learningassisted method are tested in Zn||Zn symmetrical cells, providing the experimental evidence for high ionic conductivity and stability. This work provides a new guidance for designing high-safety and high-performance eutectic electrolytes for ZIBs that can operate efficiently over a broad temperature range.