Machine learning-based stacked ensemble model for predicting and regulating oxygen-containing compounds in nitrogen-rich pyrolysis bio-oil

The production of low-carbon liquid biofuels and chemicals is a method for increasing the added value of biomass through pyrolysis technology. However, compared to fossil fuels, the high oxygen content in biomass results in a higher proportion of oxygen-containing compounds in bio-oil, which adversely affects its storage stability and combustion performance. This study tackles the challenge of regulation and prediction of various oxygen-containing compounds in nitrogen-rich pyrolysis bio-oil by developing a machine learning-based, multi-model stacked ensemble learning framework. The ensemble model incorporates 853 data points, using Decision Trees, Random Forests, and Gradient Boosting Decision Trees as base learners, combined with Linear Regression as the meta-learner. The model demonstrated exceptional performance in both cross-validation and independent dataset tests, achieving an R2 value of 0.97 and an RMSE of 0.49. Through SHAP value analysis and Monte Carlo simulations, the study identifies critical factors influencing the generation of oxygen-containing compounds, such as cellulose content, fixed carbon content, and pyrolysis temperature. This work enhances the predictive accuracy of oxygen-containing compounds in nitrogen-rich pyrolysis bio-oil, as well as the generalization and interpretability of the model. It offers scientific evidence and strategies for optimizing the pyrolysis process and improving bio-oil quality.