Integrated learning framework for enhanced specific surface area, pore size, and pore volume prediction of biochar

Specific surface area, pore size, and pore volume are essential biochar properties. Optimization typically reduces yield by focusing on per gram of biochar. This work introduces new indicators and an integrated model to balance quality and quantity, emphasizing overall adsorption potential per gram of raw biomass. The integrated model outperformed nine machine learning models with 91.93% accuracy, RMSE of 0.73, and R² of 0.965. SHAP analysis identified temperature, volatile matter and ash content as the most influential factors. PDP analysis provided insights into their interactions, while PSO determined the optimal conditions for maximizing adsorption efficiency. Among three indicators, temperature emerged as the common key parameter, with optimal averages identified at 720℃. Furthermore, A user-friendly interface was developed for visualizing training and prediction, enhancing model applicability. This work achieves a quality-quantity balanced biochar design with interpretable mechanisms, advancing adsorption optimization and practical implementation.