A Soft-Sensor Model for NOx Concentration Based on Enet-GPR

Abstract

The control and optimization of Selective Catalytic Reduction (SCR) system has been one of the research hotspots of thermal power units. Accurate measurement of the Nitrogen Oxide (NOx) concentration at the entrance of SCR is of great significance for SCR control and optimization. Firstly, Elastic Net (Enet) method is used to variable selection. This method improves the penalty coefficient by convex combination of $L_{1}$ and $L_{2}$ norm, which has the advantages of ridge regression (RR) and Least Absolute Shrinkage and Selection Operator (LASSO), and overcome the problem of collinearity and group effects in the data when using the LASSO Method. Then, focusing on the advantages of the Gauss process regression (GPR) model, such as the easy acquisition of the super parameters, the flexibility of non parametric inference and the probability significance of output, the Enet-GPR soft-sensor model is established. Field data simulation results show that the proposed method has excellent prediction accuracy and generalization performance.