Prediction of a fly ash-based proxy for PCDD/F emissions from municipal solid waste incineration plants using a deep...

To address the high cost, long detection cycles, and limited predictive accuracy associated with conventional monitoring of polychlorinated dibenzo-p-dioxins and dibenzofurans (PCDD/Fs) in municipal solid waste incineration (MSWI) facilities, this study develops a multi-parameter deep forest (DF) regression model to predict a fly ash-based proxy indicator of PCDD/F emissions. The dataset covers 100 MSWI facilities across China. The studied units are grate-furnace plants equipped with broadly similar air-pollution-control devices, and the model is therefore intended to generalize primarily within this technology envelope. Among the tested models, the DF framework using XGBoost as the base predictor yielded the best within-dataset performance, achieving a test-set coefficient of determination of 0.982 and a mean absolute percentage error of 8.95% when 60% of the data were used for training. HCl emerged as the dominant predictive feature (>80% model weight), whereas SO₂ showed a strong negative association with the target proxy; furnace temperature and incineration capacity each contributed less than 5%. Because the target series was derived from quarterly fly ash disclosures and temporally aggregated, the reported metrics should be interpreted as performance for a smoothed proxy indicator rather than direct proof of stack-emission compliance or broad transferability to unseen plants. The framework can therefore serve as a fleet-level screening and hypothesis-generating tool for identifying candidate control priorities in MSWI operations.