Metal-solvent interaction contribution on voltage for metal ion battery: An interpretable machine learning approach

Abstract

Rechargeable metal ion batteries (MIBs) are of paramount significance for electrochemical energy storage, utilization, and transportation in modern civilization. Various electrode materials have been explored to improve the voltage of battery. However, the role of metal-solvent interaction energies in voltage determination is yet to be explored in MIBs. Here, we have considered a large number of metal-solvent combinations to predict the interaction energy using the machine learning (ML) techniques followed by anode half-cell voltage calculation. A total of 1584 metal-solvent systems consisting of six metals (Li, Na, Mg, Al, K, Ca) and 66 solvents have been considered for this work. The gradient boosting regression (GBR) has been found to be the best-fitted ML model for the prediction of interaction energy. Further, with increasing the solvent number around the metal center, the effect of voltage changes has been investigated systematically. Moreover, an interpretable ML algorithm (shapash) has been implemented for local and global feature analysis. Our results establish the relation between metal solvent interaction energy and voltage and also offers suitable solvents for different MIBs. It further establishes ML techniques as promising alternative for computationally demanding calculations as first screening tools for energy storage devices.