Machine learning prediction of biochar-specific surface area based on plant characterization information

Biochar has a wide range of environmental applications. The use of machine learning technology to predict the characteristics of biochar adsorption is a hot topic in current research. This paper proposes the concept of plant organs (PO) information for the first time. Based on a deep neural network (DNN) and light gradient boosting machine (LightGBM), a machine learning prediction model was created for the specific surface area of biochar, with 70 % of the 187 collected data points used as the training set and 30 % as the test set. The findings demonstrate that adding PO characteristics enhances the model's performance and accuracy, with the DNN model outperforming the LightGBM model. With the DNN model, the RMSE is lowered by 7.58, the MAE is reduced by 4.52, and the R2 increases from 0.897 to 0.917. Combined with Pearson correlation coefficient (PCC) analysis and Shapley additive explanations (SHAP) analysis, pyrolysis temperature (PT) is a key feature affecting the specific surface area of biochar, and PO significantly impacts the model's performance. By identifying and selecting specific plant organs for biochar production, researchers can more effectively optimize the biochar production process.