Machine-learning-aided predicting and engineering of hydrochar nitrogen-containing functional groups.

Hydrochar produced via hydrothermal treatment (HTT) is an emerging carbonaceous material whose properties are strongly governed by N-containing functional groups (NCFGs). In this study, machine learning (ML) was applied to predict and regulate four key NCFGs: amine-N (N-A), pyrrolic-N (N-5), pyridinic-N (N-6) and quaternary-N (N-Q). Four single-target random forest (RF) models achieved test coefficient of determination (R2) values of 0.904-0.943 with average root mean square error (RMSE) of 2.916-7.204. A multi-target RF model was subsequently constructed with an average test R2 of 0.882 and RMSE of 4.956. External validation using 33 additional hydrochars resulted in an R2 of 0.877 and RMSE of 7.494, confirming the model's robustness. Incorporation of validation data further improved the final multi-target model (R2 = 0.894, RMSE = 4.747). Subsequently, an online system was developed to enable prediction and process optimization. These findings demonstrated the potential of ML in guiding NCFGs-targeted hydrochar production.