Exploring the Critical Factors of Biomass Pyrolysis for Sustainable Fuel Production by Machine Learning

Abstract

The goal of this study is to use machine learning methodologies to identify the most influential variables and optimum conditions that maximize biochar, bio-oil, and biogas yields for slow pyrolysis. First, experimental results reported in 37 articles were compiled into a database. Then, an explainable machine learning approach, Shapley Additive exPlanations (SHAP), was employed to find the effects of descriptors on the targets, and it was found that higher biochar yields can be obtained at lower temperatures using biomass with low volatile matter and high ash content. Following that, decision tree classification was used to discover the variables leading to high levels of the targets, and the most generalizable path for high biogas yield was found to be where the maximum particle diameter was less than or equal to 6.5 mm and the temperature was greater than 912 K. Finally, association rule mining models were created to find associations of descriptors with very high levels of yields, and among many findings, it was discovered that biomass with larger particles cannot be converted into bio-oil efficiently. It was then concluded that machine learning methods can help to determine the best slow pyrolysis conditions for the production of renewable and sustainable biofuels.