Prediction of NOx emissions for coal-fired power plants with stacked-generalization ensemble method

By Yuan Zhaowei, Meng Lei, Gu Xiaobing, Bai Yuyong, Huanmin Cui, Cheng Jiang. 2020

76 citations recorded by the source corpus.

Abstract Measuring the nitrogen oxides (NOx) concentration accurately at the inlet of the denitration reactor plays an important role in controlling the NOx emissions for coal-fired power plants. Therefore, a NOx emissions prediction method based on stacked-generalization ensemble method (SGEM) was explored. Firstly, principal component analysis (PCA) method was used to eliminate the correlations between the original auxiliary variables. Subsequently, the mutual information (MI) method was used to select the auxiliary variables that have a greater impact on the inlet NOx emissions. Finally, The NOx emissions model was established based on SGEM with the selected auxiliary variables. Among the SGEM method, the back-propagation neural network (BPNN), support vector regression (SVR) and decision tree (DT) were used as the base models and the linear regression (LR) as the meta-model. The optimal hyper-parameters of each base model were determined by the grid search and 10-folds cross-validation methods. Based on the historical data with variable conditions from distributed control system (DCS), the comparison results showed that the predicted and measured results of SGEM were concentrated in the direction of the 45-degree, and the relative error (RE) was mainly distributed between −5 mg/Nm3 and 5 mg/Nm3. Simultaneously, the SGEM achieved the highest square of correlation coefficient (R2) and the smallest root mean square error (RMSE), mean absolute error (MAE) and mean absolute percentage error (MAPE) than any individual method. The 10-folds cross-validation results also showed that the SGEM method had strong robustness and generalization ability.

Subfields: Biomedical Engineering, Environmental Engineering, Industrial and Manufacturing Engineering

View Paper: https://www.semanticscholar.org/paper/3c3515ac621323492588a6f4d73e9222ea96b2ed