Process modelling integrated with interpretable machine learning for predicting hydrogen and char yield during chemical looping gasification

Abstract

Chemical looping gasification (CLG) is a promising thermochemical process for the production of H2. CLG process is mainly based on oxygen transfer from an air reactor to a gasification reactor using solid metal oxides (also called oxygen carriers, (OC)) as oxidants. The unique oxygen separation system of CLG makes it an advanced process with a smaller carbon footprint compared to the conventional gasification process. The other advantages of CLG includes increased efficiency, reduced greenhouse gas emissions, and improved process stability compared to conventional biomass gasification. Although CLG is a promising technology, it still faces several challenges such as high capital cost, OC durability, complex reaction mechanism and scalability issues. Some of these challenges can be addressed by understanding the impact of various process conditions on H2 yield and char formation during CLG. The present study proposes a novel integrated process simulation and experimental studies to generate large dataset used for interpretable machine learning (ML) analysis. Three different ML models including support vector machine (SVM), random forest (RF), and gradient boost regression (GBR) were used to develop models for predicting the H2 and char yield during CLG. The GBR outperformed other models for the prediction of H2 and char yield during CLG with R2 value > 0.9. Among the experimental conditions, the temperature (T) and steam to biomass ratio (SBR) were the most relevant parameters affecting H2 and char production. Biomass ash, C, volatile matter (VM) and H content also influenced H2 and char formation. Overall, a combination of SHAP and partial dependence plot helped address the black box challenges of ML models.