Abstract
N-doped biochar has garnered significant attention for its potential applications in environmental remediation and energy storage. However, their performance is affected by the complex interaction of multiple features, and conventional preparation methods (PM) face challenges such as long experimental cycles and complex parameter optimisation. This study employs machine learning (ML) to systematically predict the specific surface area (SSA), total pore volume (Vt), and nitrogen content of N-doped biochar (NC), aiming to establish quantitative relationships between preparation parameters and performance metrics. Using input features, including biomass elemental composition, carbonization conditions, activation conditions, nitrogen doping conditions, and PM, single-target and multi-target random forest-extreme gradient boosting (RF-XGB) models were developed. Results demonstrated that the single-target models achieved test regression coefficient (R2) values of 0.86, 0.84, and 0.85 for SSA, Vt, and NC, respectively, validating the model's accuracy. SHAP value analysis revealed that activation temperature (AT) predominantly influenced SSA, activating agent ratio (AR) governed Vt, and nitrogen doping ratio (DR) critically determined NC. Among the PM, the one-step method significantly enhanced nitrogen content, while the multi-step method favored pore structure development. Partial dependence plots further quantified critical parameter thresholds: SSA and Vt peaked at AT = 800 °C and AR = 0.2, whereas NC reached its maximum at DR = 0.6. This study provides a data-driven strategy for the targeted design and process optimization of N-doped biochar. © 2025 Elsevier Ltd