Prediction Model of NO_x Emission from Thermal Power Plants based on Stacking Ensemble Learning

Because of the delay in the denitrification reaction and the presence of a long gas pipeline in the selective catalytic reduction (SCR) system, there is a significant delay in the measurement of NO<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">x</inf> content in the discharged flue gas in thermal power plants. A prediction modeling scheme based on the Stacking ensemble learning was proposed to improve accuracy of SCR model. Firstly, data preprocessing of the historical data of the field station was obtained to select the parameters that would affect the NO<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">x</inf> content at the outlet. Then the delay time of each variable was estimated, the historical data were reconstructed, and the input variables were screened for features. Finally, a Stacking ensemble model was constructed using light gradient boosting machine (LightGBM), support vector machine (SVM), gated recurrent unit (GRU), and random forest as base learners and logistic regression as meta-learner. Experimental results based on actual production data show that this model's mean absolute percentage error and root mean squared error are 2.61% and 0.784, respectively, and the improved techniques also show performance advantages in the comparison experiments, which proves that the model proposed in this paper has excellent prediction ability for NO<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">x</inf> emission.