Abstract: The selective non-catalytic reduction (SNCR) denitrification system in waste incinerators involves multiple influencing factors, strong nonlinearity, and response delays, posing challenges to the accurate measurement of NO_<i>x emissions. A dynamic modeling method integrating Transformer with bidirectional gated recurrent unit (BiGRU) was proposed, with model parameters optimized using the black-winged kite algorithm (BKA). First, mechanism analysis was conducted to identify primary denitrification factors for preliminary feature selection. Then, the combined feature extraction algorithm—MRX-SynFilter algorithm was employed to select highly correlated variables and remove redundant ones. Subsequently, variable delay times were estimated by data trend analysis combined with the maximal information coefficient (MIC). Next, the BKA was used to search for optimal variable order combinations to complete dynamic data reconstruction, enhancing the model's ability to capture temporal characteristics. Finally, the dynamic model was built on the Transformer-BiGRU network, with network hyperparameters simultaneously optimized by BKA. Experimental results demonstrate that variable selection, delay estimation, and order selection all improve modeling accuracy. The BKA-Transformer-BiGRU model outperforms other comparative models in fitting accuracy and error control, effectively capturing the complex nonlinear dynamic characteristics of the SNCR denitrification system.