A Data-Driven Modeling Framework Integrated with Thermodynamic Validation for Many-Objective Optimization of Syngas Production in Biomass Chemical Looping Gasification

Accurate prediction and systematic optimization of syngas production in biomass chemical looping gasification (BCLG) are challenged by feedstock heterogeneity and operational variability and the complex coupling of multiple reactions. Herein, a data-driven many-objective optimization framework integrated with mechanistic modeling is proposed. Initially, a thermodynamic equilibrium model was constructed to simulate the BCLG process. Subsequently, a comprehensive data set compiled from publications was utilized to both validate the thermodynamic model and train the machine learning (ML) model. Among the algorithms evaluated, gradient boosting regression (GBR) achieved the highest predictive accuracy ( R 2 = 0.903) for gas composition and syngas yield. SHAP analysis based on the GBR model revealed temperature and steam-to-biomass (S/B) ratio as the dominant factors influencing syngas production. Moreover, GBR-based optimization identified the compromise optimal conditions for high-quality syngas production, reintroducing the optimal operating conditions into the thermodynamic model verified the primary gas species attainment with relative errors below 8%, and noticeable discrepancies in CH 4 concentration and syngas yield indicated the importance of nonequilibrium effects, further demonstrating the reliability of the proposed framework. This study presents a synergistic optimization framework that accelerates the experimental design of the BCLG system and establishes a transferable optimization paradigm applicable to other thermochemical conversion processes.