Multi-objective optimization framework for nitrogen-containing compounds generation in nitrogen-enriched pyrolysis: Integrating transfer learning and experimental validation

Abstract

A multi-objective optimization approach, integrating machine learning and transfer learning, was proposed to optimize the generation of nitrogen-containing compounds in nitrogen-enriched pyrolysis of biomass. A high-accuracy Gradient Boosting Regression Tree (GBRT) model was developed using 827 experimental data sets, with transfer learning employed to accelerate training on specific target variables. This approach significantly enhanced both learning efficiency and predictive performance. The model achieved a Coefficient of Determination (R²) of 0.968 and a Mean Absolute Error (MAE) of 1.047 on the test set, demonstrating exceptional predictive capability. Through Principal Component Analysis (PCA) and model interpretability methods such as SHapley Additive exPlanations (SHAP) and Local Interpretable Model-agnostic Explanations (LIME), key influencing factors were identified. The critical factors include nitrogen source ratio, pyrolysis temperature, and protective gas. The study identified a synergistic effect when the nitrogen source ratio was 50.00 % and the pyrolysis temperature was 550°C. This condition led to the maximum generation of nitrogen-containing compounds. Additionally, increasing the nitrogen source ratio reduced the formation of volatile compounds, while higher lignin content promoted the formation of aldehydes and ketones. Experimental validation via nitrogen-enriched pyrolysis of corn stover confirmed the practical applicability of the model. The model accurately predicted nitrogen-containing compounds generation, with the maximum prediction error constrained to within 6.20 %. This study combines data-driven methods with experimental validation. The approach provides a novel technological framework for optimizing complex chemical reactions and supporting the sustainable production of high-value nitrogen-based chemicals.