Dynamic NO _x Prediction Model for SCR Denitrification Outlet of Coal-Fired Power Plants Based on Hybrid Data-Driven and Model Ensemble

Aiming at the difficulty of accurate modeling of selective catalytic reduction (SCR) systems in coal-fired power plants, this paper proposes a modeling scheme based on hybrid data-driven and model ensemble to achieve the dynamic estimation of NO (x) concentration at the outlet of SCR systems. First, the LOF algorithm is adopted to detect outliers in the raw production data. The maximum information coefficient (MIC) algorithm is employed to calculate the time delay of each input variable, and phase space reconstruction is performed to enhance the nonlinear correlation between inputs and outputs. Second, the elastic net algorithm is employed to select potential input variables to determine the optimal input variables. Third, to further improve the modeling accuracy, the kernel fuzzy C-means (KFCM) clustering algorithm is utilized to cluster the selected input variables, and the Xie-Beni index is introduced to determine the optimal number of clusters. Finally, a model for predicting the NO (x) concentration at the outlet of the SCR system is established by combining the model ensemble and error correction strategies, and the sublearners of the ensemble model are the genetic algorithm optimized back-propagation neural network (GA-BP) model, least-squares support vector machine (LSSVM) model, and extreme gradient boosting (XGBoost) model. The validation results based on actual field operation data show that the established prediction model can accurately predict the outlet NO (x) emission concentration of the SCR system with excellent dynamic tracking performance and prediction consistency, which meets the operational requirements of industrial applications and can provide excellent guidance for the production field.