High Temperature Superheater Wall Temperature Prediction Method Based on ARLD-LSTM Recurrent Neural Network

To effectively predict the temperature trend of superheater walls in advance for preventing forced shutdown of thermal power units caused by superheater overheating, a data-driven superheater wall temperature prediction model using historical data is proposed in this paper. Using real data obtained from a thermal power plant in Shanghai, an autoregressive distributed lag (ARDL) time series prediction model of superheater wall temperature is constructed. Grey correlation was employed to obtain the correlation coefficients between each influencing factor and the superheater wall temperature to find the best number of time-lag steps. Then, the lagged time window predicted by the ARDL model was embedded into the input layer of the long-and-short term memory (LSTM) recurrent neural network (RNN) to predict the superheater wall temperature. The whole data-driven approach is abbrivated as ARDL-LSTM RNN model. Compared with a conventional LSTM model, the proposed ARDL-LSTM RNN prediction model shows superior performance with less volatility and lag. Hence it is promising to be used for superheater wall over-heating prediction and in-advance warning.