On the construction of hybrid algorithms to predict biochar yield as a function of pyrolysis parameters

Biochar yield prediction is essential for adjusting pyrolysis progressions and proceeding justifiable biomass utilization. This research develops a unified, interpretable framework that join in Gradient Boosting Decision Trees (GBDT) with metaheuristic optimizing algorithms, Ant Colony Optimization (ACO), Whale Optimization Algorithm (WOA), Coupled Simulated Annealing (CSA), and Particle Swarm Optimization (PSO), to enhance predictive accuracy. A comprehensive dataset of 211 experimentally validated pyrolysis observations, comprising 14 physicochemical and operational parameters, was curated from peer–reviewed literature. After preprocessing through outlier detection and normalization, all models were trained using 5–fold cross–validation and evaluated via R 2 , MSE, and AARE%. SHAP analysis was employed to quantify feature contributions and elucidate mechanistic relationships between pyrolysis conditions and biochar yield. Among the optimized models, the GBDT-ACO model achieved the highest predictive accuracy, yielding test R 2 values of 0.709 and test MSE of 16.284. The results highlight the critical influence of ash content, residence time, and peak temperature on yield formation, while also revealing the trade–off between computational efficiency and predictive robustness across optimization strategies.