In order to absorb large-scale new energy into the power network

coal-fired power stations require frequent load adjustments

which brings some difficulties to reheat steam temperature control. Based on the historical operation data of a certain 1 000 MW ultra-supercritical coal-fired boiler

a prediction model of boiler reheat steam temperature based on multi-layer perceptron（MLP） neural network is established. The results show that the mean square error of the reheat steam temperature prediction model is only 0.71℃

which is superior to long and short term（LSTM）neural network and short vector machine（SVM）in mean relative error

correlation coefficient

training time and generalization effect. Introducing the prediction results of the model into the controller is helpful to improve the reheat steam temperature control quality.