Fuente: PubMed "swarm" Waste Manag. 2026 Jun 16;222:115670. doi: 10.1016/j.wasman.2026.115670. Online ahead of print.ABSTRACTAccurate prediction of fly ash generation from municipal solid waste incineration is critical for source reduction and cost-effective disposal. However, conventional models fail to capture the complex nonlinear relationships governing fly ash formation. In this study, a machine-learning-based prediction framework was developed using six algorithms (Lasso, KNN, DT, SVM, RF, and XGBoost), with XGBoost identified as the optimal model, achieving an R2 of 0.896, RMSE of 178.49, and MAE of 101.44 on the test set. Model interpretability analysis using SHapley Additive exPlanations and Partial Dependence-Individual Conditional Expectation analysis methods revealed that incinerator type, designed disposal capacity, operating load, and lime injection rate were the dominant factors influencing fly ash yield. Based on these insights, optimized operational ranges were proposed, including an operating load of 95 ∼ 105 %, deacidification efficiency of 85 ∼ 95 %, denitrification efficiency of 60 ∼ 70 %, high-temperature flue gas residence time of 2.7 ∼ 3.0 s, average furnace temperature of 860 ∼ 870 °C, and lime injection rate of 50 ∼ 100 kg/h. By applying Particle Swarm Optimization to optimize incineration parameters in a specific case study, the fly ash yield was reduced by 22.40 %, corresponding to a maximum annual reduction of 1.229 thousand tons. This optimization could lower the enterprise's annual fly ash disposal costs by approximately 2.089 ∼ 2.458 million CNY, while achieving a minimum reduction of 682 tons of CO2 emissions per year.PMID:42308755 | DOI:10.1016/j.wasman.2026.115670