Prediction and optimization of flue gas desulfurization in waste incineration using catboost and permutation-based feature importance

With the widespread adoption of waste incineration technology in urban environments, the accurate prediction and control of flue gas pollutants such as sulfur dioxide (SO₂) and hydrogen sulfide (H₂S) have become essential for sustainable development within smart city ecosystems. Traditional desulfurization system modeling methods, which rely on physical mechanisms or linear assumptions, are inadequate for capturing the nonlinear, multivariable dynamics in real-world operation. To address these challenges, this study proposes a data-driven, intelligent regulation framework that aligns with smart city principles. The framework integrates CatBoost (Categorical Boosting)-based nonlinear regression for high-precision pollutant prediction, permutation importance for interpretable feature analysis, and genetic algorithm-based optimization for process regulation. Experimental results show that this integrated approach significantly enhances prediction accuracy and reduces emissions, achieving an average reduction of over 12%. By embedding machine learning into urban environmental infrastructure, the proposed model supports intelligent, adaptive emission control and provides a scalable solution for smart ecological management in future urban systems.