Screening Vanadium-Based Electrode Materials for Lithium-Ion Batteries Based on Machine Learning and First-Principles

This study investigates the discovery of vanadium-based electrode materials for lithium-ion batteries using a combination of machine learning and first-principles calculations.The study extracted data for 4694 vanadium-based materials from the Materials Project database and calculated their maximum theoretical capacities using pymatgen.Correlation analysis revealed that material density has the highest correlation with theoretical capacity (Comprehensive score is 0.905), indicating that density is a key factor affecting the performance of vanadium-based materials.Based on this finding, three machine learning models (Deep Neural Network -DNN, Random Forest -RF, and Support Vector Regression -SVR) were constructed, with hyperparameters optimized using a genetic algorithm.The results showed that the DNN model performed best, achieving an R 2 value of 0.771.SHAP analysis further confirmed the significant contribution of density to the model predictions.Based on model predictions, the top 0.5% of high-density materials were selected from the original dataset, ultimately identifying 26 potential vanadium-based electrode materials.Among these, V6O13, MgV5O12, and V2MoO8 exhibited theoretical capacities exceeding 650 mAh/g, with open-circuit voltages of 2.56, 0.64, and 0.49 V, respectively.First-principles calculations indicated that V6O13 possesses metallic properties with high electron density of states, which is beneficial for enhancing conductivity; MgV5O12 and V2MoO8 exhibit metallic and semi-metallic characteristics, respectively.Adsorption energy calculations showed that V6O13 has the lowest adsorption energy, indicating its strongest lithium-ion adsorption capability.Based on open-circuit voltage calculations, V6O13 is suitable as a cathode material, while MgV5O12 and V2MoO8 are more appropriate as anode materials.This study successfully identified high-performance vanadium-based electrode materials through the integration of machine learning and first-principles calculations, significantly shortening the experimental process.It provides new insights into the design of lithium-ion battery electrode materials and demonstrates the enormous potential of machine learning in materials science.