Machine learning-based prediction of methane production from lignocellulosic wastes

Area of Science:

• Biotechnology
• Renewable Energy
• Environmental Science

Background:

• The biochemical methane potential (BMP) test is standard for assessing lignocellulosic waste biodegradability in anaerobic digestion (AD).
• BMP tests are time-consuming and costly, necessitating alternative prediction methods.
• Optimizing AD processes requires accurate prediction of methane yield.

Purpose of the Study:

• To develop a machine learning (ML) model for predicting cumulative methane yield (CMY) from lignocellulosic wastes (LWs).
• To identify key feedstock characteristics and digestion parameters influencing CMY.
• To provide guidance for feedstock selection and AD plant operation.

Main Methods:

• Utilized a dataset of 157 LWs, including physicochemical properties and digestion conditions.
• Developed and validated an ML model to predict CMY.
• Performed model interpretability analyses to identify significant predictors.

Main Results:

• Achieved a coefficient of determination (R²) of 0.869 for CMY prediction.
• Identified lignin content, organic loading, and nitrogen content as critical factors for CMY.
• Found that high cellulose content (>50%) initially lowers CMY but prolonged digestion increases it.
• Observed that lignin content above 15% significantly inhibits methane production.

Conclusions:

• The developed ML model offers a reliable and efficient alternative to traditional BMP tests.
• Feedstock composition, particularly cellulose and lignin content, critically impacts methane yield.
• Optimized feedstock selection and digestion parameters can enhance biogas production in AD plants.