Applying machine learning to predict torrefaction and pyrolysis activation energy based on biomass characteristics and heating conditions

Pyrolysis of biomass is a complex process involving many reactions and components. Many experiments have been carried out to determine the kinetics of pyrolysis and low-temperature torrefaction. In this paper, the multiple linear regression and random forest regression methods were used to predict the activation energy of biomass pyrolysis and torrefaction as determined by Kissinger-Akahira-Sunose (KAS) and Flynn-Wall-Ozawa (FWO) methods. The input variables were biomass characteristics, heating conditions and conversion. The hyper-parameters of the model were optimized by grid search and 5-fold cross validation. In all cases performed in this study, the prediction accuracy of the random forest models was found to be acceptable with R 2 > 0.83 and RMSE < 0.04. The contribution of each variable to the activation energy was analyzed by Pearson analysis, feature importance analysis and partial dependence analysis. This study recommended that machine learning can be applied for predicting kinetic parameters of the thermochemical process, providing a new tool for understanding, simulating and evaluating biomass pyrolysis and torrefaction processes. • Activation energy of biomass pyrolysis were predicted by machine learning. • Random forest method gave a better performance than linear regression method. • Effects of input variables on activation energy were analyzed. • Pyrolysis activation energy was mostly affected by conversion and fixed carbon. • Torrefaction activation energy was mostly affected by carbon and moisture contents.