Novel method for predicting concentrations of incineration flue gas based on waste composition and machine learning.

The complex composition of solid waste leads to the variability of flue gas emissions during its incineration, which poses a challenge to the stable operation of incineration and pollution control systems. To address this problem, the study explored a new method to predict the concentrations of flue gas pollutants during incineration based on the composition of mixed solid waste using machine learning. Through comprehensive model interpretation and analysis, the important influence of waste composition characteristics on the generation of flue gas pollutants during incineration was deeply explored. The study found that rubber and plastic components significantly promoted the conversion of C to CO during waste incineration; N content and C/N ratio had a significant effect on the generation of NO X ; S content and C/S ratio affected the generation of SO 2 ; Cl content and C/Cl ratio had a significant effect on the generation of HCl, especially with PVC components. The extreme gradient boosting tree (XGBOOST) model optimized by feature engineering showed more excellent R2-validation (0.98, 0.94, 1.00, 0.98, 1.00, 0.98) for predicting CO 2 , CO, N 2 O, NO, SO 2 , and HCl concentration, than K-nearest neighbor (KNN), random forest (RF), and light gradient boosting machine (LGBM). This study provides a new prediction and optimization method for waste incineration plants, which can guide the regulation of feedstock and incineration parameters, thus improving operating efficiency and pollution control. It is of great significance to promote sustainable waste management and environmental protection. • A novel ML-based approach predicts flue gas concentrations from waste composition. • Feature engineering identifies key predictors and uncovers nonlinear relationships. • XGBOOST outperforms other models in predicting flue gas pollutant concentrations. • The approach optimizes waste composition and enhances pollution control efficiency. [ABSTRACT FROM AUTHOR]