Dynamic modeling for NOx emission sequence prediction of SCR system outlet based on sequence to sequence long short-term memory network. (1st January 2020)

• Record Type: Journal Article
• Authors: Xie, Peiran; Gao, Mingming; Zhang, Hongfu; Niu, Yuguang; Wang, Xiaowen
• Is Part Of: Energy. Volume 190(2020)
• Publication Date: 2020-01-01
• Subjects: Selective catalytic reduction -- NOx sequence prediction -- Sequence to sequence model -- Long short-term memory; Power resources -- Periodicals; Power (Mechanics) -- Periodicals; Energy consumption -- Periodicals; 333.7905
• DOI: 10.1016/j.energy.2019.116482
• Languages: English
• ISSNs: 0360-5442

Abstract

Abstract: As environmental protection policies become more stringent, lower and lower NOx emission targets are required. Accurate NOx concentration prediction model plays an important role in low NOx emission control in power stations. This study aims to accurately predict the future sequence of NOx emission in the next horizon. Through the analysis on formation mechanism of NOx and the reaction mechanism of SCR reactor, a sequence to sequence dynamic prediction model is proposed, which can fit multivariable coupling, nonlinear and large delay systems. In particular, considering the different effects of multivariate on NOx, a new attention mechanism is necessary to be put forward. A large amount of historical data is used to fully train this dynamic prediction model. The results show that, the prediction accuracy of the NOx concentration and fluctuation trend based on this model is superior to comparison algorithms. Furthermore, some interesting features of this prediction model, such as error accumulation and bidirectional encoder, are also discussed in depth. Highlights: The prediction of NOx is a sequence in the next horizon, that can reflect the trend of fluctuation. Variable selection attention mechanism is designed, considering the different effects of multivariate on NOx emission. Maximal Information Coefficient-based feature selection is used for selecting optimal inputs. A modeling method for multivariable nonlinear large delay systems with attention mechanism is proposed.