Experimental simulation and analysis of Acacia Nilotica biomass gasification with XGBoost and SHapley Additive Explanations to determine the importance of key features

Abstract

Biomass gasification is a versatile and environmentally friendly process that turns biomass feedstocks such as agricultural waste, wood, or organic waste into a combustible gas known as producer gas. This technique has several significant advantages, including renewable energy sources, waste utilization, and reduction in greenhouse gases. The biomass gasification process in a special-purpose reactor known as a gasifier is complex and highly nonlinear. The process modeling in such cases becomes complex and difficult. Stakeholders find black-box models produced by traditional machine-learning approaches hard to understand. XGBoost and SHapley Additive Explanations (SHAP) approaches are combined in this research work to improve the prediction accuracy and interpretability of the biomass gasification process. The prediction models for the main constituents of producer gas (hydrogen and carbon monoxide), lower heating value, and cold gas efficiency were developed. The robust prediction ability of XGBoost ML was demonstrated with a higher coefficient of determinant values in the range of 0.9558-0.9968 with a low mean squared error (0.0029-1.3928) during model testing. The combined use of XGBoost and SHAP values helped to enhance the comprehensible understanding of the influence of each attribute.