Development of robust machine learning models to estimate hydrochar higher heating value and yield based upon biomass...

Area of Science:

• Biomass valorization and conversion technologies.
• Machine learning applications in chemical engineering.
• Sustainable energy and materials science.

Background:

• Hydrothermal carbonization (HTC) converts biomass into hydrochar, a valuable material.
• Predicting hydrochar yield and higher heating value (HHV) is crucial for optimizing HTC processes.
• Accurate prediction models require comprehensive datasets and robust analytical methods.

Purpose of the Study:

• To develop and validate a machine learning framework for predicting hydrochar yield and HHV.
• To compare the performance of various machine learning algorithms for HTC outcome prediction.
• To identify key biomass proximate analysis parameters influencing hydrochar properties.

Main Methods:

• A dataset of 481 biomass samples was curated, including proximate analysis and HTC parameters.
• Monte Carlo Outlier Detection (MCOD) was used for data cleaning.
• Thirteen machine learning algorithms were evaluated, including CNN, ensemble methods (CatBoost, LightGBM, XGBoost), and others.
• SHAP (SHapley Additive exPlanations) analysis was employed to interpret model predictions.

Main Results:

• CatBoost demonstrated superior performance, achieving R² of 0.98 for HHV prediction and R² of 0.94 for yield prediction.
• Key predictors for HHV included ash content, while temperature, water content, and fixed carbon were significant for yield.
• The framework effectively models HTC outcomes, supporting data-driven biomass valorization.

Conclusions:

• Gradient boosting models, specifically CatBoost, are highly effective for predicting hydrochar yield and HHV.
• Biomass proximate analysis data, combined with machine learning, provides a powerful tool for optimizing HTC processes.
• This approach facilitates efficient and sustainable biomass valorization strategies.