Data fusion is a method which suits for complex system fault diagnosis such as nuclear power plants, and is multi-source information processing technology. In this paper, the data fusion information hierarchical thinking used and the nuclear power plants fault diagnosis divided into three levels. In data level data mining method adopted to handle data and reduction attributes. In feature level three parallel neural networks used to deal with attributes reduction of data level and the outputs of three networks are as the basic probability assignment of Dempster-Shafer (D-S) evidence theory. The improved D-S evidence theory synthesizes the outputs of neural networks in decision level, which conquers the traditional D-S evidence theory limitation that cannot dispose conflict information. The diagnosis method is tested through using correlation data of document. The test results indicate that the data fusion diagnosis system can diagnose nuclear power plants faults accurately and the method which has a certain applicable value in use.

This content is only available via PDF.
You do not currently have access to this content.