Prediction of syngas properties of biomass steam gasification in fluidized bed based on machine learning method

Abstract

Steam gasification of biomass is a promising technology to produce hydrogen-rich syngas. While the complex correlation between gasification process and syngas properties has not been effectively explained. In this study, 222 relevant experimental data from peer-reviewed publications was collected to predict the syngas properties in fluidized bed reactors by machine learning (ML) method. Five algorithms were adopted and the Extreme Gradient Boosting regression (XGBoost) showed the most consistent predictive performance with the testing R2 of 0.89–0.92. The equivalent ratio (ER) and steam to biomass ratio (S/B) were the most significant determinants of H2 concentration. Optimal H2 production (around 45%) was achieved by favorably choosing operating conditions characterized by the range of ER < 0.08 and 1 < S/B < 2.5. The valuable insights provided by ML models may contribute to the better understanding of biomass steam gasification process for the H2-rich syngas production.