Building an accurate prediction model of boiler main steam temperature is conducive to improving its control quality. The time delay characteristics of boiler operation parameters have a great impact on the prediction accuracy of main steam temperature. Long-short term memory(LSTM) neural network algorithm is adopted to establish a model to predict the change trend of boiler main steam temperature

and aiming at solving the problem of determining the time delay characteristics of boiler operation parameters

discrete particle swarm optimization algorithm is used to optimize the time delay of network model input variables. Moreover

based on the historical data of a 1 000 MW coal-fired boiler

the main steam temperature prediction model optimized by time delay characteristics is verified. The results show that

the prediction error of the optimized LSTM network model considering the time delay characteristics is 0.47 ℃

namely 6% smaller than that of the conventional LSTM model

indicating it has a higher prediction accuracy.