Application of LSTM in Metal Temperature Forecasting for a 1000 MW Double-Reheat Coal-Fired Boiler

Since more renewable energy is being incorporated into the electrical grid, coal-fired power units must operate flexibly rather than as base-load. Sharp changes in the reheat steam temperature during flexible operation of the secondary reheat units can cause thermal stress on the reheater, resulting in tube breakage or unplanned shutdown. To avoid the risk of this issue occurring, the operating temperature of the reheater should be precisely monitored in real time. A novel long short-term memory (LSTM) for predicting metal temperatures is developed using real operating data obtained from a 1000 MW doublereheat power plant boiler. The soot blower action signals are added for more accurate prediction. The results indicate that the LSTM has adequate accuracy. The root mean square error for the test dataset is 0.861 ℃, the mean absolute percentage error is 0.105% and the Pearson correlation coefficient between the actual value and the predicted value is over 0.979.