Prediction of discharge performance of hard carbon materials for high-performance sodium-ion batteries based on machine learning

Abstract

The development of high-performance hard carbon anodes is crucial for enhancing the performance of sodium-ion batteries (SIBs). Traditional density functional theory calculations and experimental methods, which are time-consuming and costly, are inadequate for large-scale data analysis and the exploration of complex relationships. In contrast, machine learning methods are capable of efficiently processing vast amounts of data and uncovering the underlying relationships between material properties. Based on this, a hard carbon structure-performance database comprising 503 sets of data was constructed based on previously reported experimental data. A variety of machine learning models, including random forest, gradient boosting decision tree (GBDT), and extreme gradient boosting (XGB), were employed in conjunction with cross-validation and data augmentation techniques to predict and analyze the discharge performance of hard carbon materials. In this work, the concept of a “cyclic factor” was innovatively defined through the comparison and fitting of multiple datasets. This factor is designed to more accurately quantify the cycling stability of hard carbon materials. The results showed that the R2 values of the best-fitting models for different performance indicators on the test set ranged from 0.670 to 0.847, with root-mean-square errors (RMSE) below 8%, indicating good generalization performance. Furthermore, through feature importance analysis, the key structural characteristics affecting discharge performance were effectively identified. Combined with single-factor partial correlation analysis and two-factor interaction analysis, the influence trends of these characteristics were further explored in depth. This study provides scientific guidance for the structural design and performance optimization of hard carbon materials.