Unveiling the relationships between process parameters and nitrogen-containing chemicals via machine learning: A...

Machine learning (ML)-driven analysis of biomass fast pyrolysis provides a promising strategy to uncover the complex relationships between process parameters and nitrogen-containing chemicals (NCCs), thus facilitating experimental optimization and enhancing the valorization of biomass resources. In this work, the optimized random forest model achieved satisfactory predictive performance (Training R² = 0.91, Test R² = 0.80), which was further verified experimentally with an accuracy of 78.09%. Visual interpretation of the ML results indicated that a high nitrogen content in the feedstock favored NCCs formation, while high temperatures imposed an inhibitory effect. To improve the production of NCCs, both intrinsic nitrogen (derived from macroalgae) and extrinsic nitrogen sources (urea, melamine, and ammonium bicarbonate) were introduced to elevate the nitrogen content of feedstock, and the product distribution of NCCs was explored at relatively low pyrolysis temperatures. Results demonstrated that urea addition at 550 °C exhibited high selectivity for pyridines (> 70% of N-heterocycles). Melamine introduction at 650 °C was more suitable for amides production, while ammonium bicarbonate incorporation at 350---450 °C facilitated azoles preparation (pyrazoles, imidazoles). Finally, probable reaction pathways for extrinsic-nitrogen-driven NCCs synthesis were proposed, providing mechanistic insights for selective NCCs production from macroalgae N- enriched pyrolysis.