Hybrid data-driven machine learning model for predicting performance of lignocellulosic biomass gasification

Gasification is a promising technology for converting lignocellulosic residues into valuable energy and fuels. However, widespread adoption is challenging due to technical barriers in biomass thermoconversion, notably feedstock variability and tar formation. To advance in this field, modeling strategies have been employed to optimize the conversion process and reduce the need for costly and time-consuming experimental tests. This study introduces a hybrid machine learning (ML) model designed to predict gas, yield, syngas composition, and tar concentration from the gasification of different lignocellulosic feedstocks. The model combines advanced ensemble ML methods: gradient boosting machines (GBM), eXtreme Gradient Boosting (XGBoost), Categorical Boosting (CatBoost), and Natural Gradient Boosting (NGBoost), which were rigorously selected from a pool of ten ML methods using k-fold cross-validation and extensive performance evaluation. The model was trained on a comprehensive dataset of 270 experimental data points collected from 31 peer-reviewed studies, featuring varied levels of cellulose (0-69.85 %), hemicellulose (5.15-87 %), and lignin (8.55-62.86 %) under different processing conditions. Initial performance metrics (R2) of individual algorithms ranged from 0.59 to 0.93, and after hyperparameter optimization, the model's predictive accuracies significantly improved, with R² values ranging from 0.77 to 0.94. Model interpretability tools were then employed to quantify the influence of each input feature, providing insight into the prediction mechanism. Thus, this hybrid ML model holds the potential to greatly reduce the experimental efforts required to optimize the gasification process for diverse lignocellulosic residues.