Wall temperature overheating is considered to be one of the main causes for tube burst in supercritical(ultra-supercritical) boilers

and it has become a key problem that puzzles the operation of supercritical(ultra-supercritical) boilers. On the basis of analyzing the existing wall temperature prediction technology

this study applies NARX dynamic neural network to the advanced prediction of wall temperature of platen superheater. Firstly

manual screening and grey relational analysis method are employed to ascertain the key factors influencing the wall temperature change. Then

the input combinations of different influencing factors and the setting of model parameters are analyzed and optimized

and the NARX prediction model of wall temperature of platen superheater is established. Moreover

combined with the historical data of the actual unit

the proposed model is applied to predict the superheaters' tube wall temperature

and the results are compared with that of the NAR neural network model. The results show that

the NARX neural network model considering the key factors influencing the wall temperature has better prediction performance

which can keep higher prediction accuracy for a long time and forecast the wall temperature by 1 min in advance. The conclusion has certain theoretical and engineering application value for relevant researches and solving the problem of wall temperature overtemperature for supercritical(ultra-supercritical) boilers.