Machine Learning Model Insights into Base-Catalyzed Hydrothermal Lignin Depolymerization.

Area of Science:

• Biomass Conversion
• Chemical Engineering
• Machine Learning Applications

Background:

• Lignin, a plant biomass component, offers potential for renewable chemicals and biofuels but is underutilized.
• Homogeneously catalyzed lignin depolymerization in water is promising but lacks standardized yield reporting.
• Inconsistent reporting of bio-oil and solid residue yields hinders process optimization and comparability.

Purpose of the Study:

• Develop predictive machine learning models for bio-oil and solid residue yields in lignin depolymerization.
• Validate model predictions through experimental work.
• Identify key reaction variables influencing depolymerization outcomes.

Main Methods:

• Utilized machine learning algorithms to build predictive models for bio-oil and solid residue yields.
• Employed reaction variables as input features for the models.
• Validated model performance using experimental data and calculated coefficients of determination (R²).
• Assessed variable importance using two distinct methodologies, including Shapley additive explanation (SHAP).

Main Results:

• Achieved R² scores of 0.83 for bio-oil yield and 0.76 for solid residue yield.
• Identified temperature and reaction time as the most significant predictors of experimental outcomes.
• Variable importance analysis aligned with existing literature, providing mechanistic insights.

Conclusions:

• Machine learning models offer reliable predictions for lignin depolymerization yields.
• Established clear guidelines for reporting are crucial for advancing lignin depolymerization research.
• Temperature and reaction time are critical parameters requiring careful control and reporting.