A real-time data neural netwerk prediction model used in thermal power plants established on BP neural network and based on Levenberg-Marquardt(abbrevated as L-M)algorithm has been put forward

to reduce the training times and enhance the training accuracy.Through training and checking inlet temperature of the high-pressure heater for 300 MW unit in one power plant

the importance of data preprocessing has been analysed.Results of emulation show that the reasonable prediction result at future moment can be obtained by using the said model and the model can be used for supplement of shortcoming data and for real-time data checking

to enhance the reliability of data

and being applicable to on-line evaluate the future condition

providing with reference basis for condition based repair.