Machine Learning Properties of Electrolyte Additives: A Focus on Redox Potentials
Yun Zhang, Xiaojie Xu
Abstract
Electrolyte additives for lithium-ion battery (LIB), commonly categorized into anode additives, cathode additives, redox shuttle additives, and fire retardants, can improve properties of electrolytes and provide protection of electrodes and battery operations. Redox potentials are major properties that influence the performance and applications of the additives. In this study, we develop Gaussian process regression models to predict redox potentials of electrolyte additives for LIBs from molecular structural features of electrolyte additives. The models are simple and fast to implement, produce predictions with low root mean squared errors, and thus might be considered as efficient alternatives to the DFT approach for estimations of redox potentials. The Gaussian process regression models also provide statistical correlations between the molecule structure and redox potential.