Prediction of activation energy of lignocellulosic biomass pyrolysis through thermogravimetry-assisted machine learning

Abstract

Activation energy is a crucial indicator to explain the pyrolysis process of lignocellulosic biomass that can be converted into clean fuels and high-value-added chemicals. The present work employs machine learning algorithms to explore the hidden correlations between feedstock characteristics, thermogravimetry performance, reaction kinetic equation, and activation energy, which allows these kinetic parameters to be obtained without the Arrhenius equation. Bayesian optimization improves the Gradient boosting decision tree model performance: MSE of 198.67, R² of 0.991. Feature importance analysis and partial dependence analysis show that feedstocks with high lignin content and low nitrogen and oxygen content usually have higher activation energy. Meanwhile, the temperature when the mass is reduced to 80% of the initial mass during the thermogravimetry process plays a very important role in activation energy. The results show that the machine learning model can accurately learn the relationship between thermogravimetry curve characteristics and activation energy, so it can accurately predict the activation energy of lignocellulosic biomass pyrolysis. This approach can potentially save time and effort that would otherwise be spent on tedious calculations required by kinetic reaction equations. Additionally, it can provide useful guidance for experimental studies.