Supervised Machine Learning‐Based Classification of Li−S Battery Electrolytes

Abstract Machine learning (ML) approaches have the potential to create a paradigm shift in science, especially for multi‐variable problems at different levels. Modern battery R&D is an area intrinsically dependent on proper understanding of many different molecular level phenomena and processes alongside evaluation of application level performance: energy, power, efficiency, life‐length, etc. One very promising battery technology is Li−S batteries, but the polysulfide solubility in the electrolyte must be managed. Today, many different electrolyte compositions and concepts are evaluated, but often in a more or less trial‐and‐error fashion. Herein, we show how supervised ML can be applied to accurately classify different Li−S battery electrolytes a priori based on predicting polysulfide solubility. The developed framework is a combined density functional theory (DFT) and statistical mechanics (COSMO‐RS) based quantitative structure‐property relationship (QSPR) model which easily can be extended to other battery technologies and electrolyte properties.