Abstract: Due to the complexity and variability of control parameters in the incineration process, it is difficult to quickly identify the most critical control parameters during manual operation. XGBoost-SHAP is often used for interpretability analysis of machine learning models and can analyze the importance ranking of specific features relative to predicted values. In this study, operational data of the co-incineration of biogas residue, leachate, and sludge were collected from a grate-type incinerator at an actual municipal solid waste incineration power plant. By actively adjusting parameters commonly regulated in industrial practice (such as air flow rate and grate residence time), the XGBoost-SHAP importance interpretation method was employed to analyze the effects of parameter adjustments on the main steam flow rate and gaseous pollutant emissions. The results indicated that the furnace temperature and total air flow rate had the most significant impact on the main steam flow rate. The secondary air flow rate exerted the greatest influence on NOx concentration, and the co-incineration of leachate had the most pronounced effect on parameter variations, followed by sludge and biogas residue. This study clarified the priority order of potential relationships among regulatory parameters during the combustion process. Furthermore, by integrating real-time operational data with principal component analysis, it identified positive and negative correlations among various parameters, offering valuable references for practical engineering operations.

Key words: XGBoost-SHAP, importance order, co-incineration, parameter optimization