A hybrid VMD-TCN-LSTM model for NOX prediction at the SCR inlet of a municipal solid waste incineration plant

Municipal solid waste (MSW) incineration plants commonly employ a two-stage denitrification route consisting of selective non-catalytic reduction (SNCR) followed by selective catalytic reduction (SCR). After partial NOx removal by SNCR, the flue gas entering the SCR unit may still contain highly variable residual NO, which directly determines the ammonia demand of the downstream catalyst reactor. Accurate prediction of residual NO at the SCR inlet is therefore important for feedforward ammonia-injection control. To address this issue, this study proposes a hybrid VMD-TCN-LSTM model for short-term prediction of NO concentration upstream of the SCR reactor. VMD is first used to separate the non-stationary signal into multi-scale components, after which a temporal convolutional network (TCN) and a long short-term memory (LSTM) network are combined to extract local temporal patterns and sequential dependencies. The proposed model achieved an R2 of 0.9891, an RMSE of 0.6660 mg/m3, an MAE of 0.5218 mg/m3, and a MAPE of 8.04% on the test set, outperforming the benchmark models. The results show that the proposed framework effectively captures the non-stationary and multi-scale characteristics of the SCR-inlet residual NO signal and provides practical support for precise SCR ammonia dosing.