Nonlinear Dynamic Soft-Sensing Modeling of NOx Emission of a Selective Catalytic Reduction Denitration System

Abstract

The development of accurate data-driven models is of great significance for online monitoring and control of industrial processes. In this study, a nonlinear dynamic soft-sensing algorithm for measuring and predicting the nitrogen oxides (NOx) emissions of the denitration system of a power plant was developed. First, the processes and principles of the denitration system were studied. Second, a soft-sensing algorithm based on long short-term memory (LSTM) and least absolute shrinkage and selection operator (LASSO) was proposed. The dynamics of the process were captured by LSTM, and the selection of influential input variables was achieved by LASSO. A moving window cross validation (MWCV) strategy was applied to determine the hyperparameters of the proposed algorithm. Third, the proposed algorithm was implemented to predict NOx emissions from a denitration system. The experimental results demonstrated that the proposed approach had superior accuracy and higher precision of variable selection compared with other state-of-the-art soft-sensing algorithms. In addition, the variable importance analysis provided by the proposed algorithm is consistent with the principles of chemical reactions and field experiences. The proposed algorithm could provide theoretical and technical support for the optimization of the denitration system and predictive control design of the denitration process.