Machine learning prediction of biocrude yields and higher heating values from hydrothermal liquefaction of wet biomass and wastes

Abstract

Machine learning (ML) approach was applied for the prediction of biocrude yields (BY) and higher heating values (HHV) from hydrothermal liquefaction (HTL) of wet biomass and wastes using 17 input features from feedstock characteristics (biological and elemental properties) and operating conditions. Several novel ML algorithms were evaluated, based on 10-fold cross-validation, with 3 different sets of input features. An extreme gradient boosting (XGB) model proved to give the best prediction accuracy at nearly 0.9 R2 with normal root mean square error (NRMSE) of 0.16 for BY and about 0.87 R2 with NRMSE of about 0.04 for HHV. Temperature was found to be the most influential feature on the predictions for both BY and HHV. Meanwhile, feedstock characteristics contributed to the XGB model for more than 55%. Individual effects and interactions of most important features on the predictions were also exposed, leading to better understanding of the HTL system.