ABSTRACT
Recently, high-voltage lithium-ion batteries (LIBs) have gained a lot of attraction due to the potential application in electric vehicles. However, conventional carbonate solvent-based electrolytes are less stable against high-voltage cathodes. In this chapter, machine learning to design new electrolytes for high-voltage LIBs has been discussed. Different metrics for molecular selection, training, and predicting models are considered to find out an appropriate method. The two important parameters for characterizing electrolytes, namely electron affinity (EA) and ionization energy (IE), are chosen. Our results show that the SMILES (one-hot encoding) format combined with a recurrent neural network (RNN) to predict EA/IE has better prediction performance compared to other neural network (NN) models. Furthermore, the generative model was applied to carry out inverse design. The machine learning techniques lead to a new way to create brand-new functional molecules for electrolytes without doing complicated calculations and costly experiments. The generative model could serve as a first step design toward the further investigation into other necessary properties of electrolytes for high-performance LIBs.
