Machine learning application to predict yields of solid products from biomass torrefaction. (April 2021)

• Record Type: Journal Article
• Authors: Onsree, Thossaporn; Tippayawong, Nakorn
• Is Part Of: Renewable energy. Volume 167(2021)
• Page Start: 425
• Page End: 432
• Publication Date: 2021-04
• Subjects: Biomass -- Gradient tree boosting -- Machine learning -- Solid fuels -- Torrefaction; Renewable energy sources -- Periodicals; Power resources -- Periodicals; Énergies renouvelables -- Périodiques; Ressources énergétiques -- Périodiques; 333.794
• DOI: 10.1016/j.renene.2020.11.099
• Languages: English
• ISSNs: 0960-1481

Abstract

Abstract: Machine learning was used to develop a model that had the capability to predict yields of solid products from biomass torrefaction using input features of biomass properties and torrefaction conditions. With ten-fold cross-validation, several machine learning algorithms were evaluated, and their hyper-parameters were optimized by a full-factor grid search. Gradient tree boosting algorithm was found to have the highest prediction accuracy with R 2 of about 0.90 and an average error of 0.07 w/w. Six highly important features on making predictions of the model were torrefaction temperature, residence time, and O2 concentration in the reacting gas for torrefaction conditions, as well as volatile matter, carbon content, and oxygen content for biomass properties. Unlike the carbon content, the other features were found to have a negative effect on the yields of torrefied biomass. The biomass property features contributed to the solid yields for about 30%, with approximately one-third accounted by the volatile matter. Graphical abstract: Image 1 Highlights: Machine learning used to predict yields of torrefied biomass by 14 features input. Partial dependence plot showed inside into impacts of each variable on torrefaction. Gradient tree boosting offered the highest prediction accuracy at about 0.90 R 2 . Biomass properties contributed to the yields for 30% with 1/3 from volatile matter.