Research Of Diagnosis And Prediction For Boiler Fault

The software of diagnosis and prediction for boiler fault is developed by using fuzzy modular networks and recurrent composed networks, and the method of mixed knowledge representation and expert system technology etc are used in this paper. The new samples and the knowledge database of fault diagnosis and fault prediction are added in the process of software development. Knowledge acquirement module,diagnosis module,prediction module and interpretation module are established at the same time. Simulation tests of common boiler fault are carried out by using fault diagnosis and fault prediction software, which is programmed with Visual C++.The simulation tests result indicates that the speed and precision of sample training are increased because of sample clustering for fuzzy modular networks. And the problem of slow training speed and local minimum point are avoided when BP networks are applied in the fault diagnosis of complex boiler. Judging from multi output of the fuzzy modular networks, can decide not only whether the fault of boiler happens, but the severity degree of boiler fault also well. It is used to provide reference to operator of power plant. In recurrent composed BP networks, the relation of interior node is enhanced because the link weight of input layer and output layer are added, and the saturation of fault prediction is avoided by using the linear prompting function. The comparison of three models(single variable time series model,multivariable time series model and gray prediction model)shows that the multivariable time series model's prediction precision is the highest. It indicates that using recurrent composed BP networks can exactly predict the boiler fault in order to prevent the fault, and help operator of power plant to adjust the parameters in a permitting range.