Machine-learning-aided prediction and engineering of nitrogen-containing functional groups of biochar derived from biomass pyrolysis

Biochar, derived through biomass pyrolysis, is an emerging carbonaceous material with significant potential in various applications. The performance of biochar in applications depends heavily on its surface characteristics, including its content of N-containing functional groups. In this study, machine learning (ML) was used for the prediction and engineering of biochar N-containing functional groups, specifically amine-N (N-A), pyrrolic-N (N-5), and pyridinic-N (N-6). The single-target random forest model accurately predicted N recovery in biochar (char-N yield) and N-A, N-5, and N-6 contents, achieving test R2 of 0.91–0.97. Model interpretations revealed that pyrolysis temperature was the most influential factor in predicting char-N yield, N-A, N-5, and N-6. The multi-target random forest model achieved an average test R2 of 0.93, and it was used to optimize pyrolysis parameters for designing biochar N-functional groups. This process was followed by experimental verification. Relative errors for the three N-containing functional groups were mostly within 15% (18.4% on average). The favorable results obtained from experimental verification demonstrate the significant potential of ML in biochar engineering.