Prediction of nitrogen product distribution in algal pyrolysis bio-oil driven by machine learning

Algal biomass pyrolysis shows promise for bio-oil production, but the presence of nitrogen-containing compounds in bio-oil hinders its application. Thus, regulating these nitrogenous chemicals is crucial for environmental protection and efficient utilization of algal resources. This study aimed to address the challenge of predicting nitrogenous product distribution in algal pyrolysis. Machine learning methods like Random Forest (RF) and Extreme Gradient Boosting (XGB) were employed to build a high-performance prediction model for nitrogenous products (e.g., pyridines, pyrroles, nitriles, amines) from algal pyrolysis. The model was trained with 210 sets of pyrolysis data. Using box plot outlier detection, SHAP analysis, and PDP bivariate interaction analysis, it revealed the nonlinear regulatory mechanisms of nitrogen content, algal feedstock properties (C, H, O, N, S, ash), and pyrolysis conditions (temperature, etc.) on nitrogenous product distribution. Results showed bio-oil yield reached its peak predicted by the model at 500-550 °C with reaction time over 15 min. At temperatures below 500 °C, amines and amides are the main nitrogen-containing products. As temperature rises, pyrroles and nitriles become dominant, with nitriles production showing an oscillating trend with temperature. Pyrroles and pyridines content increase at higher temperatures. The N/C ratio affects nitrogen product formation through feedstock nitrogen content: higher N/C promotes nitrogen product generation, while higher O/C inhibits amines and amides and higher H/C benefits amines/nitriles production. This study provides data to optimize algal pyrolysis parameters.