Machine learning accelerates high-voltage electrolyte discovery for lithium metal batteries

Appropriate electrolytes are essential for ensuring the performance stability of high-voltage lithium metal batteries. However, the complexity arising from multiple solvents and their relative ratios leads to significant challenges for the design of electrolytes. Herein, we propose a machine learning (ML) approach that links the microscopic properties of electrolytes with their macroscopic battery performance, thus enabling the discovery of high-performance electrolyte formulations to be accelerated. By designing chemical groups of electrolyte solvents as features and establishing a cycling stability evaluation metric for batteries, an ML model is created to predict the capacity retention of high-voltage lithium metal batteries. Through integrating this ML model with a heuristic optimization algorithm, a series of superior electrolytes are identified within a ternary solvent design space (over 29,000 possible electrolytes). Our model reveals that a specific proportion of the fluorinated ether diluent is critical for achieving superior capacity retention in fluorinated electrolytes. To validate the cycling stability of these electrolytes, we experimentally tested them in a Li||LiNi_0.5Mn_1.5O₄ coin cell configuration. All the cells containing the discovered electrolytes demonstrate outstanding capacity retention, consistent with our model's prediction trend. This work highlights the potential of ML approaches for the design and optimization of stable, high-performance battery electrolytes.