SO2 Prediction for Wet Flue Gas Desulfurization Based on Improved Long and Short-Term Memory

The general thermal power plant wet flue gas desulfurization (WFGD) process suffers from slurry resource wastage and unstable SO2 content in the outlet flue gas. Considering the complicated mechanistic modeling of conventional WFGD systems and the time-series characteristics of data., this paper investigates the construction of a long and short-term memory (LSTM) neural network., including two long and shortterm memory layers., two rectified linear unit (ReLU) function layers., a fully connected layer., and input and output layers., for the prediction of the main indicators of WFGD systems. Among them., data processing techniques are used to determine each input variable of the model and the output variable with SO2 exported; a certain percentage of data is used to verify the reliability of the model. 20.,000 sets of data are used for training., 1.,000 sets of data are used for testing., and 1.,000 sets of data are used to verify the accuracy of the model. The results show that the established improved LSTM model has higher prediction accuracy., which is 97.7%., compared with back propagation (BP) neural network., recurrent neural network (RNN)., and the basic LSTM model., which can achieve more accurate control over the use of relevant resources and reduce waste., and can be used as one of the scientific methods for system optimization.