Prediction of CO2 adsorption performance of biochar based on machine learning

Machine learning shows great potential in high-dimensional data processing and complex problem analysis, and is a promising approach for biochar adsorption CO₂ modeling research. In this study, we innovatively predicted the CO₂ adsorption by biochar from the activation conditions of biochar using deep neural network (DNN), random forest (RF), gradient boosted decision tree (GBDT), extreme gradient boosting (XGB), and light gradient boosting machine (LGBM) algorithms. A comparison of seven different combinations of input features revealed that the best prediction accuracy was found for the combination that retained all the activation condition features and that the introduction of the activation condition had an effect on the other features as well. The LGBM model had a higher prediction accuracy and prediction performance for the dataset (R² of 0.956, MAE of 0.245, and RMSE of 0.350). The importance of the features under this model was also analyzed, and the results showed that the activation temperature and the flow rate of the protective gas were more important than the type of activator and the activation ratio for CO₂ adsorption, and activation time had the least effect. This study provides new perspectives and methods for the prediction study of CO₂ adsorption by biochar and also provides a reference for a deeper understanding of the related mechanism of action.