Machine Learning Approach for Predicting Hydrothermal Liquefaction of Lignocellulosic Biomass

Hydrothermal liquefaction (HTL) of lignocellulosic biomass has gained attention as a promising technology for the production of biofuels and other value-added products. HTL process optimization is complex and involves various parameters such as reaction time, temperature, and pressure. In recent years, machine learning (ML) approaches have been adopted as a tool to optimize and predict the HTL process performance. The purposes of this study were to investigate the ML-based prediction of bio-crude yield (BCY) and their higher heating values (HHVs) from HTL of lignocellulosic biomass and to elucidate the relationship of features affecting the output of interest. Pre-processing and normalization were applied to a dataset of 215 data points with 17 input features. Feature selection using the Shapley value method identified key predictors. ML models including multilayer perceptron, kernel ridge regression, random forest, and extreme gradient boosting (XGB) were trained and evaluated. XGB algorithm shows superior performance in predicting the yields and their calorific values to within 5–8% of experimental values. Temperature was the most influential feature for both BCY and HHV prediction accounting for about 30%, followed by other feedstock and operational characteristics. In addition, a user interface was presented for ease of use in the ML modeling.