Directing against the problems of too large size and poor generalization capability of the neural metwork structure due to too many input factors in establishing model for predicting NOx emission from coal-fired boilers in power plants

the BP neural network has been modified through principal component analysis and Bayesian regularization

the network structure being optimized

thereby

enhancing its generalization capability.Taking test data of boiler for one 300 MW unit under many operating conditions of hot state as example

through comparison of the prediction method using modified neural network with the method of traditional nuural network

the generalization capability has been remakably enhanced

and the covergence of network being stable

the effectiveness of practical prediction being good.