Predicting municipal solid waste gasification using machine learning: A step toward sustainable regional planning

Abstract

The gasification process can treat and valorize municipal solid waste (MSW) in an environmentally and economically friendly way. Using this process, MSW can be safely disposed of and sustainably converted into bioenergy as part of regional planning. Experimental laboratory data is a key component in designing, optimizing, controlling, and scaling up MSW gasifiers. However, most researchers lack the resources and time to conduct experiments. Machine learning (ML) technology can resolve this issue by detecting patterns and hidden information in published data. Hence, the present study aims to construct an inclusive ML model to predict and understand the MSW gasification process. The objective is to establish a consistent and homogeneous database containing MSW sources under different gasification conditions, followed by an analysis of the database using statistical methods. Three ML models are used to predict the distribution of syngas, char, and tar and the quality of syngas in MSW gasification using feedstock characteristics and gasification parameters. When a gradient boost regressor is used to model the process, the prediction accuracy is highest (R2 > 0.926, RMSE <6.318, and RRMSE <0.304). SHAP analysis is successfully used to understand the significance and contribution of descriptors on targets in the modeling process.