Deep learning prediction of electrode voltage for metal-ion batteries

Abstract

Voltage is a critical factor in estimating the energy density of metal-ion batteries. However, the current exploration of electrode materials with ideal voltage through trial and error is time-consuming and costly. In this study, we develop a deep learning model incorporating residual networks and multi-head attention mechanisms to accurately predict the voltages of electrode materials. With a comprehensive training and testing, the developed model can achieve a mean absolute error (MAE) of 0.04 V, representing the highest accuracy reported to date for voltage prediction. Notably, even if the training datasets for metal-ion systems such as K-ion and Y-ion batteries are limited, the model still maintains an impressive accuracy. Furthermore, the voltages of 26 potential high-voltage electrode materials can be correctly predicted, demonstrating the robustness and applicability of the model. This work not only highlights the potential for high-precision predictions under data-scarce conditions but also provides a promising pathway for screening advanced electrode materials in advanced metal-ion batteries.