Abstract: In the limestone-gypsum wet desulfurization systems of thermal power plants， the pH value of circulating slurry in absorption towers is an important parameter that affects the performance of desulfurization systems. Therefore， establishing an effective pH value prediction model is fundamental for improving the desulfurization efficiency. The absorption tower system has the characteristics of massive data volume and variable parameters which have strong correlations. Thus，feature extraction and Pearson's r correlation analysis were performed on the data from a plant-level SIS （Supervisory Information System） database.Then， the extracted features were taken as the input of the Long Short-Term Memory （LSTM） neural network to obtain a prediction model for the pH value of desulfurization slurry. The model was applied to a supercritical 330 MW unit coal-fired unit desulfurization system in predicting the pH value of the absorption tower. The root mean square error （RMSE） of the pH values predicted by LSTM neural network model is 0.004， and the mean absolute error （MAE） is 0.003； The test shows that the data tracking results made by LSTM neural network model is of little fluctuations， small errors and high stability.

Key words: neural networks, prediction model, wet desulfurization, pH value of the absorption tower, artificial intelligence, LSTM, SIS, big data