Prediction of preparation conditions for low PAHs corn straw biochar guided by CatBoost model optimized via genetic algorithm and molecular dynamics simulation.

Biochar shows great potential in cultivated soil improvement, but its polycyclic aromatic hydrocarbons (PAHs) are toxic and may threaten ecological security and human health via the soil-plant system. To optimize low-PAHs biochar production, this study used corn straw as raw material, exploring the correlation between preparation conditions and PAHs formation through hydrothermal carbonization experiments, machine learning, and molecular dynamics simulation. A database with 10 input variables (hydrothermal temperature, time, etc.) and PAHs content as output was built. Among four machine learning models, CatBoost performed best. After genetic algorithm (GA) optimization, its test set R2 reached 0.9914, enhancing analysis of multi-factor nonlinear relationships. Feature importance showed hydrothermal time (30.35%), nitrogen content (18.38%), and hydrogen content (13.37%) were key factors. Partial correlation analysis indicated that beyond 4 h, prolonged high temperature reduced PAHs via dealkylation, while high nitrogen and hydrogen promoted accumulation. Molecular dynamics confirmed prolonged time accelerated carbon chain fragmentation, with energy changes reflecting reaction shifts. Optimal conditions (240-300 °C, 6 h) controlled PAHs below 347.04 μg/kg. Validation showed < 2.5% error, confirming GA-CatBoost reliability. This study provides a data-driven strategy for biochar optimization, aiding its safe agricultural application.