Comparative study of machine learning algorithms for predicting syngas yield in chemical looping gasification systems

Abstract

An effective method for creating pure syngas with built-in CO₂ collection is chemical looping gasification (CLG). In order to forecast the composition of syngas in CLG systems with metal oxide carriers, this study assesses seven machine learning algorithms. Models such as Support Vector Machines, Ensemble Methods, and Neural Networks were evaluated using R2 (0.81 – 0.96), RMSE, and cross-validation using an extensive dataset from various reactor settings. The results demonstrate the performance of target-specific algorithms: tree-based ensembles perform best for methane, whereas kernel-based approaches perform best for hydrogen. The steam-to-biomass ratio, temperature, and carrier particle properties are identified as the main determinants via feature importance analysis. Multi-objective optimization identifies balanced operating conditions achieving 29.0% hydrogen yield and 16.9% carbon monoxide yield. The framework provides useful advice for controlling and optimizing CLG processes.