(370m) Machine Learning for Biorefining: Towards a Universal Kinetic Model of Wood Deconstruction

Hemicelluloses are amorphous polymers of various sugar molecules and Hemicellulose content ranges from 25 to 35 wt% in wood [1]. Hemicellulose has applications in the bioenergy, textile, mining, cosmetic and pharmaceutical industries [2]. Industrial use of hemicellulose often requires that the polymer be hydrolyzed into constituent oligomers and monomers. Hemicellulose hydrolysis has been studied extensively, yet there lacks a universal model of this process that can be applied to multiple species and reactor conditions. Most current models are based on Arrhenius type kinetics [3] and are only effective when applied to the conditions under which the empirical parameters were fitted. An alternative to first principles kinetic models is to use a holistic data driven approach, specifically machine learning. Due to the attention this field has received, there is substantial data in the literature pertaining to hemicellulose hydrolysis. In this project, the literature on hardwood hemicellulose hydrolysis is mined and used to build machine learning models. The kinetic and machine learning models (e.g. linear regression, support vector regression, and deep neural networks) are assessed on their ability to predict xylose yield. Predictions of xylose yield within 15% were obtained.

[1] Isikgor, F. H., & Becer, C. R. (2015). Lignocellulosic biomass: A sustainable platform for the production of bio-based chemicals and polymers.6(25), 4497-4559.

[2]Spiridon, I., & Popa, V. I. (2008). Hemicelluloses: Major Sources, Properties and Applications. Monomers, Polymers and Composites from Renewable Resources.

[3] Kapu, S. N., & Trajano, H. L. (2014). Review of hemicellulose hydrolysis in softwoods and bamboo. Biofuels, Bioproducts and Biorefining, 8(6), 857-870.