Construction of chemical looping gasification model based on machine learning: focusing on performance prediction, feature importance and parameter optimization.

Chemical looping gasification (CLG) is a promising biomass conversion technology, but its complex multi-factor coupling mechanisms hinder optimal condition screening. This study developed an integrated framework combining machine learning, Shapley Additive exPlanations (SHAP) interpretability analysis, and particle swarm optimization algorithm to address this challenge. Based on 134 valid CLG datasets, three machine learning models, namely Random Forest (RF), Support Vector Machine (SVM), and Artificial Neural Network (ANN), were constructed to predict key outputs including syngas components, gasification efficiency, and carbon conversion efficiency. The RF model outperformed others with an average R2 of 0.945. SHAP analysis revealed that the mass ratio of steam to biomass is the core factor regulating H2 and CO generation, while substrate carbon content provides the material basis for carbon conversion efficiency and, mechanistically, for CH4 production, although CH4 yield is influenced by multiple weaker factors. Ash content inhibits gasification efficiency, but its statistical contribution is relatively minor compared to fixed carbon and temperature. The particle swarm optimization algorithm further determined optimal parameter combinations for targeted outputs, such as mass ratio of oxygen carrier to biomass = 1, mass ratio of steam to biomass = 3.6, reaction temperature = 850.6 ℃ for H2 enrichment. This framework overcomes traditional experimental limitations, providing theoretical and technical support for CLG industrial application.