It is of great environmental significance to promote the deep treatment of thermal power flue gas. Therefore, this paper proposes a prediction model based on CEEMDAN-SE-RF-GWO-LSTM for the difficulty of accurately detecting the SO2 concentration at the inlet of the desulfurization system of thermal power plants. First, the process parameters related to the inlet SO2 concentration are determined as the original input variables through mechanism analysis, and a random forest algorithm (RF) is used to evaluate the importance of the variables for feature selection. Secondly, the SO2 concentration data is decomposed by complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN), and the sample entropy (SE) is used to merge and reconstruct the modal components as the final predictor variable. After determining the input variables and predictors, the long-term short-term memory network (LSTM) optimized by the gray wolf algorithm is used to establish a predictive model of the inlet SO2 concentration. The experimental results show that the grey wolf optimizer (GWO) significantly affects optimizing hyperparameters. Decomposed variables can be separately predicted and then combined to improve the prediction accuracy, while reconstruction after decomposition can improve the model training efficiency while improving the prediction accuracy.

Date Published: 6 May 2022
PDF: 11 pages
Proc. SPIE 12256, International Conference on Electronic Information Engineering, Big Data, and Computer Technology (EIBDCT 2022), 122560J (6 May 2022); doi: 10.1117/12.2635804

Published in SPIE Proceedings Vol. 12256:
International Conference on Electronic Information Engineering, Big Data, and Computer Technology (EIBDCT 2022)
Guoqiang Zhong, Editor(s)