Machine learning-driven optimization of iron-based oxygen carriers for enhanced hydrogen production from biomass chemical looping gasification

Fe-based oxygen carriers (OCs) exhibit significant potential for biomass chemical looping gasification (BCLG) to produce hydrogen. However, variations in OC composition and operating conditions strongly affect BCLG performance. In this study, Fe-based OCs were optimized by integrating experimental results with machine learning (ML) techniques, considering both material composition and operational parameters. Experimental evaluation identified Fe8Al2 as the most effective OC, achieving a hydrogen yield of 22.83 mmol/g biomass. These experimental data were combined with literature datasets to train an XGBoost model, yielding a robust predictive performance (R2 > 0.824). Interpretable ML analyses using Shapley Additive Explanations (SHAP) and partial dependence plots (PDP) revealed that the steam-to-biomass ratio and Fe content were the most influential factors for hydrogen production. This integrated approach demonstrates a viable pathway for OC optimization by supplementing limited datasets with targeted experimental data, thereby advancing hydrogen production from BCLG.