A practical ML framework for biomass torrefaction analysis and simulator deployment.

Area of Science:

• Biomass energy conversion
• Thermochemical processing
• Sustainable biofuels

Background:

• Torrefaction is a key pretreatment for biomass to solid biofuels.
• Selecting optimal torrefaction conditions is complex due to yield-energy trade-offs.

Purpose of the Study:

• Develop a machine learning (ML) framework to predict biomass mass yield (MY) and higher heating value (HHV).
• Create a user-friendly simulator for optimizing torrefaction operating conditions.

Main Methods:

• Collected and preprocessed experimental data from diverse torrefaction studies.
• Compared various regression algorithms (linear, ensemble, boosting, kernel-based) with hyperparameter tuning.
• Applied domain-informed feature engineering for enhanced model reliability.
• Integrated best-performing ML models into a GUI-based simulator.

Main Results:

• Tree-based ensemble and boosting algorithms, especially CatBoost, demonstrated robust performance in predicting MY and HHV.
• The developed simulator effectively visualizes trade-offs and identifies optimal operating windows.
• ML models provided stable feature attribution for key torrefaction parameters.

Conclusions:

• Machine learning offers a practical engineering tool for torrefaction process screening and condition selection.
• The ML-driven simulator complements traditional empirical methods, reducing trial-and-error.
• This approach facilitates efficient development of high-quality solid biofuels from biomass.