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

Area of Science:

• Biomass energy conversion
• Thermochemical processing
• Sustainable fuels

Background:

• Gasification of lignocellulosic residues offers a route to valuable energy and fuels.
• Challenges in biomass thermoconversion include feedstock variability and tar formation, hindering widespread adoption.
• Modeling is crucial for optimizing gasification processes and minimizing experimental costs.

Purpose of the Study:

• To develop a hybrid machine learning (ML) model for predicting gasification performance.
• To predict gas yield, syngas composition, and tar concentration for various lignocellulosic feedstocks.
• To reduce experimental efforts in optimizing biomass gasification.

Main Methods:

• A hybrid ML model combining Gradient Boosting Machines (GBM), XGBoost, CatBoost, and NGBoost was developed.
• The model was selected from ten ML methods using k-fold cross-validation and performance evaluation.
• Training data comprised 270 experimental points from 31 studies with diverse lignocellulosic compositions and processing conditions.

Main Results:

• Initial R2 values for individual algorithms ranged from 0.59 to 0.93.
• Hyperparameter optimization improved predictive accuracy, achieving R2 values between 0.77 and 0.94.
• Model interpretability tools quantified feature influence, offering insights into the prediction mechanism.

Conclusions:

• The hybrid ML model demonstrates high predictive accuracy for lignocellulosic biomass gasification.
• This approach can significantly reduce the experimental workload for process optimization.
• The model facilitates the efficient conversion of diverse lignocellulosic residues into energy and fuels.