Based on Dual-Channel Feature Selection for NOx Concentration Prediction Using Informer

Aiming at the problem that the large lag in the measurement of inlet NOx concentration of Selective Catalytic Reduction (SCR) in coal-fired power plants causes the delayed response of the measurement system, and the data are nonlinear and time-varying, we propose a NOx concentration prediction method based on the two-channel feature screening Informer. The model first uses the maximum information coefficient to calculate the correlation coefficient between the feature variables and the inlet NOx concentration, and screens the feature variables with a high degree of correlation, and then adopts the variational modal decomposition algorithm to decompose the inlet NOx concentration, retaining most of the local features and detailed information to reduce the influence of noise and interference on the model prediction, and the data set of the two-channel screening is reorganized to get the reconstructed data set, and finally the reconstructed data set is input into the Informer model to predict the NOx concentration. Finally, the reconstructed data set is input to the Informer model to predict the inlet NOx concentration. The DCF - Informer prediction model is validated with the historical operation data of a 600MW power boiler in a power plant. The experimental results show that the evaluation index R2 is improved by at least 6% compared with other prediction models, which verifies that the model has high prediction accuracy and better generalization effect. It provides strong support for the SCR denitrification system to realize accurate ammonia inj ection.