Understanding and optimizing the gasification of biomass waste with machine learning

Abstract

Gasification is a sustainable approach for biomass waste treatment with simultaneous combustible H2-syngas production. However, this thermochemical process was quite complicated with multi-phase products generated. The product distribution and composition also highly depend on the feedstock information and gasification condition. At present, it is still challenging to fully understand and optimize this process. In this context, four data-driven machine learning (ML) methods were applied to model the biomass waste gasification process for product prediction and process interpretation and optimization. The results indicated that the Gradient Boosting Regression (GBR) model showed good performance for predicting three-phase products and syngas compositions with test R2 of 0.82–0.96. The GBR model-based interpretation suggested that both feed and gasification condition (including the contents of feedstock ash, carbon, nitrogen, oxygen, and gasification temperature) were important factors influencing the distribution of char, tar, and syngas. Furthermore, it was found that a feedstock with higher carbon (> 48%), lower nitrogen (< 0.5%), and ash (1%–5%) contents under a temperature over 800 °C could achieve a higher yield of H2-rich syngas. It was shown that the optimal conditions suggested by the model could achieve an output containing 60%–62% syngas and achieve an H2 yield of 44.34 mol/kg. These valuable insights provided from the model-based interpretation could aid the understanding and optimization of biomass gasification to guide the production of H2-rich syngas.