Abstract:

Accurately predicting the NOx emission concentration at the outlet of the selective catalytic reduction (SCR) denitrification system in the waste incineration process is of great significance for enhancing data quality and optimizing ammonia injection. However, the waste incineration process exhibits significant nonlinearity, multivariate coupling, and time-series characteristics. These factors pose substantial challenges to achieving accurate predicting of NOx emissions. This paper presents a prediction model for NOx emission concentration at the outlet of the SCR denitrification system by integrating maximum information coefficient (MIC), principal component analysis (PCA), and long short-term memory (LSTM) neural networks. First, MIC is employed to assess the nonlinear correlations among variables, selecting input variables that exhibit the strongest correlation with NOx emission concentration while eliminating redundant variables based on the principle of maximum redundancy. Then, PCA is utilized to address the coupling characteristics among variables and to reduce information redundancy. Finally, a predicting model of NOx emission at the outlet of the SCR denitrification system is developed based on the LSTM model. The results indicate that the proposed model exhibits high accuracy and generalization ability, achieving an average absolute percentage error of 6.33%, a root mean square error of 4.71mg/m3, and 0.90 of R2. It outperforms both the back propagation neural network (BPNN) model and the support vector machine (SVM) model, thereby laying a foundation for achieving intelligent control of the SCR denitrification system in the waste incineration process.

Key words: waste incineration, selective catalytic reduction, maximum information coefficient, principal component analysis, long short-term memory neural network