Machine learning to predict electrochemical performance of biomass carbon electrodes in lithium/sodium ion batteries

Batteries, as one of the research directions for high-value utilization of biomass, often employ biomass-derived carbon as an active electrode material. However, the complex non-linear relationship between different biomass types, carbonization conditions, battery assembly conditions and testing conditions during the preparation process leads to the extensive experiments required to continuously explore the electrochemical performance of the electrodes, which impedes the rapid development of high-performance biomass-based electrodes. Therefore, in this study, based on nine machine learning models, nine types of input features were selected to predict the first cycle discharge capacity (Capacity-1), initial Coulombic efficiency (ICE), and discharge capacity after certain cycles (Capacity-x) of lithium/sodium ion batteries. The correlation of different input features was analyzed using the Spearman correlation coefficient. The feature importance and Shapley additive explanation analysis were utilized to elaborate the contribution of input features to the model prediction results. The results show that the gradient boosting regression model after hyper-parameter optimization is suitable for predicting Capacity-1 and Capacity-x, with R2 values of 0.93 and 0.90, respectively. The extreme gradient boosting model is suitable for predicting ICE, with an R2 value of 0.90. Carbonization temperature, doping conditions, and electrode components became the main influencing parameters for the three output features. Finally, the accuracy of the three models was verified by experiments. This study breaks through the traditional material research model and establishes a prediction model for the whole chain of chemical composition-microstructure-material properties of biomass electrodes, serving as a reference for the development of biomass-derived carbon electrodes.