LSTM-Based Prediction Method for NOx Emissions of Coal-Fired Boilers
作者
Dongqing Ren,Fang Wu
出处
期刊: [Institute of Electrical and Electronics Engineers] 日期:2026-01-16卷期号:: 369-373
标识
DOI:10.1109/iceaai68945.2026.11442500
摘要
Accurate prediction of NOx emissions is the prerequisite for achieving closed-loop control and meeting ultralow emission standards in coal-fired power plants. Due to the nonlinearity, strong coupling, and large time-delay characteristics of the boiler combustion process, traditional mechanisms and static models struggle to ensure prediction accuracy. This paper proposes a soft sensing model based on a Dual-Layer Long ShortTerm Memory (LSTM) neural network for a 660 MW supercritical unit. By combining mechanism analysis with Mutual Information (MI), 12 key input variables were selected. A duallayer LSTM topology was constructed to capture the long-term temporal dependencies of the combustion system. Experimental results show that compared with BP neural networks and standard RNN models, the proposed dual-layer LSTM model significantly reduces the Root Mean Square Error (RMSE) to 6.67 and improves the coefficient of determination ($\boldsymbol{R}^{\mathbf{2}}$) to 0.72. The model effectively overcomes the “gradient vanishing” problem and demonstrates superior generalization ability under variable load conditions.