Rapid Screening Strategy of 2D Materials for Li-Ion Battery (LIB) Electrode Based on Deep Neural Networks (DNN)

Abstract

The development of electrode materials is crucial for achieving an optimal performance in secondary ion batteries. Previous research has accumulated a substantial amount of data on electrode materials, creating varied data sets that include information on ion species, voltage, and other relevant characteristics. In this study, we processed the latest data and employed a deep neural network (DNN) machine learning (ML) platform to construct a regression model. The model relies on easily accessible input information, such as the initial structure, and utilizes high-quality data to validate its reliability. The two-dimensional material data set containing only the material structure is taken as the target set to predict the average discharge voltage (U_av), according to which more than 2500 potential electrode materials are selected. From this pool, we rigorously selected a subset of anode materials for detailed density functional theory (DFT) calculations. These materials exhibit promising elemental compositions and have not been previously investigated as electrode materials. The results of DFT calculations confirmed the reliability of the ML model’s predictions, demonstrating that the combination of ML and DFT calculations can effectively screen data sets lacking expensive DFT-calculated data. This strategy can significantly reduce computational costs by predicting specific performance metrics and conducting preliminary screenings.