A machine learning-assisted study on organic solvents in electrolytes for expanding the electrochemical stable window of zinc-ion batteries

Currently, inherent deficiencies of water-based electrolytes, such as a narrow electrochemical stable window (ESW), lead to low operating voltage and insufficient energy density of zinc-ion batteries (ZIBs). Incorporating organic electrolytes into ZIBs is an effective strategy for expanding the ESW but the exploration on introducing organic solvent into zinc electrolyte is still scarce. In this work, the ESWs of 307 organic solvents in ZIBs were investigated assisted by machine learning (ML) methods. Four ML models were employed to predict the oxidation potentials (OPs) of organic solvents for zinc electrolytes. Among them, Gradient Boosting Regression (GBR) and Gaussian Process Regression (GPR) exhibit exceptional performance and achieve remarkable prediction results. Specifically, GBR model displays a highest R2 score of 0.905, an absolute error of 0.258 and an absolute percentage error of 8.30% on test set. The effect of selected features on the prediction results was investigated and the features with significant impact on the prediction of OP were summarized. ESWs (OPs) of six non-aqueous zinc electrolytes using three distinct organic solvents were measured by experimental methods and there is a notable agreement between measured ESW (OP) and the solvent OP computed by Density Functional Theory and ML models in general. Furthermore, Zn//Zn symmetrical batteries assembled with these electrolytes demonstrate remarkable cycling stability, showcasing their potential applications in ZIBs. This work develops ML models that can efficiently predict a large number of organic solvent OP for ZIBs, and provides a useful guidance for developing advanced non-aqueous and hybrid zinc electrolytes.