Untangling Degradation Chemistries of Lithium-Sulfur Batteries Through Interpretable Hybrid Machine Learning.

The development of emerging rechargeable batteries is often hindered by limited chemical understanding composing of entangled patterns in an enormous space. Herein, we propose an interpretable hybrid machine learning framework to untangle intractable degradation chemistries of conversion-type batteries. Rather than being a black box, this framework not only demonstrates an ability to accurately forecast lithium-sulfur batteries (with a test mean absolute error of 8.9 % for the end-of-life prediction) but also generate useful physical understandings that illuminate future battery design and optimization. The framework also enables the discovery of a previously unknown performance indicator, the ratio of electrolyte amount to high-voltage-region capacity at the first discharge, for lithium-sulfur batteries complying practical merits. The present data-driven approach is readily applicable to other energy storage systems due to its versatility and flexibility in modules and inputs.