Machine learning-driven predictive modeling and process optimization of one-pot biomass conversion to FDCA via...

Area of Science:

• Chemical Engineering
• Biomass Conversion
• Machine Learning Applications

Background:

• Cascade reaction modeling is crucial for industrial scale-up, especially in biomass valorization.
• 2,5-furandicarboxylic acid (FDCA) is a key biobased platform chemical derived from biomass.
• Optimizing one-pot synthesis of FDCA requires efficient reaction condition prediction.

Purpose of the Study:

• To evaluate machine learning (ML) regressor models for optimizing FDCA synthesis.
• To predict FDCA yield and selectivity using limited experimental data.
• To identify the most suitable ML model for biomass conversion process optimization.

Main Methods:

• Investigated three ML models: ridge regression, support vector regressor (SVR), and gradient boosting regression (GBR).
• Utilized data from box-Behnken design of experiments for model training.
• Assessed model performance using mean absolute error (MAE) and coefficient of determination (R²).

Main Results:

• Ridge regression achieved the lowest MAE (0.595) and highest R² (95.6%) for FDCA yield prediction.
• SVR optimized reaction conditions (165.65 °C, 5.41 h, 0.80 g catalyst) yielding 66.65% FDCA.
• ML models demonstrated effectiveness in handling small datasets for process optimization.

Conclusions:

• ML techniques, particularly ridge regression, offer reliable prediction for FDCA synthesis.
• The study highlights the potential of ML in optimizing biomass conversion processes.
• The findings provide insights for efficient scale-up of FDCA production from sugarcane bagasse.