A comparative study of machine learning methods for bio-oil yield prediction - A genetic algorithm-based features...

Area of Science:

• Biomass conversion and pyrolysis
• Machine learning applications in chemical engineering
• Renewable energy research

Background:

• Bio-oil yield prediction is crucial for optimizing biomass pyrolysis.
• Accurate prediction can reduce time-consuming and expensive experimental efforts.
• Understanding complex correlations between biomass properties and bio-oil yield is challenging.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting bio-oil yield.
• To identify the most effective machine learning method for this prediction task.
• To provide a practical tool for estimating bio-oil yield and insights into the pyrolysis process.

Main Methods:

• A novel genetic algorithm was used for feature selection.
• Four distinct machine learning methods were investigated: Random Forest (RF) and Multi-Linear Regression were highlighted.
• Partial dependence analysis was employed to understand variable influence on bio-oil yield.

Main Results:

• Machine learning models demonstrated reliable prediction of bio-oil yield.
• The Random Forest model achieved high prediction accuracy (R2 ≈ 0.98).
• Multi-Linear Regression showed lower generalization performance (R2 ≈ 0.75).

Conclusions:

• Machine learning, particularly Random Forest, is effective for predicting bio-oil yield.
• The developed RF-based software package offers a practical tool for researchers.
• This approach can significantly reduce the need for extensive experimental work in biomass pyrolysis research.