Machine learning-based prediction of heating values in municipal solid waste.

Area of Science:

• Environmental Science
• Chemical Engineering
• Data Science

Background:

• Accurate prediction of municipal solid waste (MSW) heating values is crucial for waste-to-energy processes.
• Traditional methods often lack the precision required for optimizing energy recovery.

Purpose of the Study:

• To employ and compare machine learning models for predicting MSW heating values.
• To identify the most effective machine learning technique and key input parameters.

Main Methods:

• XGBoost, Extra Trees, CatBoost, and Multiple Linear Regression (MLR) were utilized.
• Models were trained using dry sample weight and elemental composition (C, H, O, N, S, ash).
• Hyperparameter tuning was performed for the Extra Trees model.

Main Results:

• The tuned Extra Trees model achieved R² values of 0.999 (training) and 0.979 (testing) with low MSE, MAE, and MAPE.
• Extra Trees significantly outperformed XGBoost, CatBoost, and MLR in predictive accuracy.
• Nitrogen content, sulfur content, ash content, and dry sample weight were identified as key predictors.

Conclusions:

• Machine learning, particularly the Extra Trees algorithm, offers a highly accurate and reliable method for predicting MSW heating values.
• Elemental composition, especially nitrogen, is a critical factor in determining heating values.
• This approach can enhance the efficiency of waste-to-energy technologies.