Prediction of product yields and heating value of bio-oil from biomass fast pyrolysis: Explainable predictive modeling and evaluation

In this study, optimized ensemble learning algorithms were employed to predict and analyze the product distribution and higher heating value (HHV) of bio-oil from biomass fast pyrolysis, based on feedstock characteristics, operating conditions, and reactor parameters. The results reveal that pyrolysis temperature, biomass carbon and hydrogen content, and feedstock volatile matter are the most influential factors for achieving high bio-oil yield, while deoxygenation pretreatment and moderate pyrolysis temperatures (approximately 500 degrees C) are critical for enhancing HHV. SHapley Additive exPlanations (SHAP) and Partial Dependence Plot (PDP) analyses further elucidated the complex interactions among these parameters, providing actionable insights for optimizing pyrolysis processes. Additionally, the developed ML models demonstrated robust predictive accuracy, with R2 values exceeding 0.93 for bio-oil yield prediction, and a user-friendly graphical user interface (GUI) was developed to facilitate practical applications. Finally, when evaluated on the external dataset, the optimized LightGBM model demonstrates a moderate linear relationship between predicted and true values, achieving an accuracy of approximately 80 %, with a peak of 84 %. The residual distribution reflects strong generalization capabilities, validating the effectiveness of the optimization strategy. This work provides a comprehensive understanding of biomass pyrolysis behavior and valuable guidance for industrial process optimization.