Simulation and optimization of co-pyrolysis biochar using data enhanced interpretable machine learning and particle swarm algorithm

Co-pyrolysis is an advanced technique used to optimize biochar by combining different biomass feedstock and adjusting process conditions. The H/N ratio and yield of biochar, along with factors such as nutrient content and specific surface area, affect the adsorption and pore structure of soil amendments. This study proposed a modeling framework to predict these indicators by the feedstock properties and process conditions. Data enhancement method was used to solve the limited data size problem, which improved the average accuracy from 7.3% to 13.3%. Grid search method was utilized to determine the optimal parameters in three typical machine learning models, namely artificial neural networks, support vector machine, and random forest. The results revealed that the random forest model achieved the highest performance, with an accuracy rate of 92.8% and an average R2 of 0.95. Sensitivity analysis indicated that the reaction temperature and specific elements of the biomass feedstock were the primary factors influencing the biochar prediction model. Consequently, the random forest model was integrated into a heuristic algorithm to find the optimal H/N ratio and yield conditions. The optimal values of H/N ratio and yield obtained were 7.14 and 29.7%. The prediction and optimization of yield and H/N ratio will provide new visual insights into the use of biochar in soil improvement. [Display omitted] •Data enhancement improved model accuracies from an average of 7.3%–13.3%.•Interpretable machine learning was used to predict co-pyrolysis biochar properties.•The optimal RF model achieved the highest accuracy of 92.8%, outgoing SVM and ANN.•The RF model integrates into PSO to find optimal production conditions for biochar.