A Study on the Application of Machine Learning in Predicting the Performance of Porous Graphite Lithium-Ion Batteries

Abstract

In this study, a dataset consisting of 350 experimental entries from the literature is compiled, and four machine learning models are developed to identify critical factors and predict battery performance. ANOVA is employed for feature selection, which controls variables and improves the speed and accuracy of the process. The Random Forest (RF) model exhibits the best performance in predicting battery retention rate, achieving a root mean square error (RMSE) of 6.1279 and a coefficient of determination (R²) of 0.82 on the test set. Feature importance analysis identifies the La/Lc ratio, specific surface area (SBET), volume of electrolyte (Vtotal), average particle size (Savg), ID/IG ratio, and lattice strain (La) as the six most significant features. The La/Lc ratio is introduced as a critical structural parameter; an increased La/Lc ratio and SBET are positively correlated with initial charge-discharge capacity and cycling stability. Data analysis reveals that when the La/Lc ratio exceeds 1.5, the retention rate improves by an average of 15%, and the charge transfer impedance decreases by 20%, significantly enhancing overall battery performance. The synergistic effects of different features indicates that an optimal Vtotal and a larger Savg significantly increases capacity and retention rate.

Keywords: electrode parameters, Graphite, Explainable Machine Learning, Lithium-ion batteries, Feature Importance