Estimation of pyrolysis liquid product yield and its hydrogen content for biomass resources by combined evaluation of pyrolysis conditions with proximate-ultimate analysis data: A machine learning application

Abstract

This study, it was aimed to predict the pyrolysis liquid product yield (Liquid P. Yield (%wt.)) and its hydrogen content (H-oil (%wt.)) based on the biomass composition and pyrolysis conditions. For this purpose, firstly, serial previous studies on biomass pyrolysis in literature were collected in a data set, and the data set used in the study was presented in the Supplementary information document. Then machine learning application was made. Within the scope of the study, Multiple Linear Regression Theory (MLR), Decision Three Theory (DT), Gaussian Model Theory (GM), and Random Forest Theory (RF) were applied to the data set obtained from the literature, and the success rates were compared. As a result, it has been determined that the RF model provides more successful results than other models in predicting the pyrolysis liquid product yield and hydrogen content. In addition, the importance feature information was obtained by using the average pollution reduction method in the whole models of RF. A detailed partial dependency analysis was performed to investigate in-depth the relationship between biomass composition and pyrolysis conditions and target variables. Thus, the amount of liquid product and the hydrogen content (quality) of the liquid product was optimized depending on the biomass composition and pyrolysis conditions. In addition, it was determined that the biomass composition was more effective on the liquid product yield and H-content than the pyrolysis conditions. It is thought that the results and comments of this study will guide the researchers in the process of pyrolysis of different biomass wastes that emerge seasonally and contribute to the more effective evaluation of biomass resources. In addition, The RF model can be used as a good reference for the modeling studies on biomass pyrolysis liquid product distribution and its quality.