Dioxin emission prediction based on improved deep forest regression for municipal solid waste incineration process

Area of Science:

• Environmental Science and Engineering
• Chemical Engineering
• Data Science and Machine Learning

Background:

• Dioxin (DXN) emission concentration is a critical environmental indicator in municipal solid waste incineration (MSWI).
• Accurate DXN emission prediction is essential for effective pollution control and optimizing MSWI operations.

Purpose of the Study:

• To propose an improved deep forest regression (ImDFR) model for predicting DXN emission concentration in MSWI.
• To enhance prediction accuracy and generalization ability compared to existing models.

Main Methods:

• Introduced a feature reduction layer using out-of-bagging error to eliminate redundant variables.
• Developed a deep ensemble stacking model integrating random forests, completely random forests, GBDT, and XGBoost.
• Validated the model using historical data from two 800-ton daily capacity incinerators.

Main Results:

• The proposed ImDFR model demonstrated higher accuracy in DXN emission prediction.
• The model exhibited superior generalization ability compared to state-of-the-art prediction models.
• Feature reduction and ensemble stacking effectively improved the prediction performance.

Conclusions:

• The developed ImDFR model provides a reliable tool for predicting DXN emissions in MSWI.
• The model supports enhanced pollution control strategies and operational optimization in waste incineration.
• The approach offers a significant advancement in environmental monitoring for MSWI processes.