Abstract

For a single neural network the lack of fault diagnosis method, the diagnosis of multiple neural networks combined with the voting fusion method to study and vote on the integration of multiple neural networks to nuclear power plant fault diagnosis. In this method, several different types of neural network training for nuclear power plant fault diagnosis. Select the security of nuclear power plant operating parameters have a major impact as the neural network input variables, neural network output is nuclear power unit failure mode. Integration with the voting method of diagnosis of different neural network fusion, nuclear power plant to get the final results of fault diagnosis. A typical use of nuclear power plant operating mode to verify the proposed method of diagnosis results. Results showed that a single neural network, the method can improve the results of nuclear power plant fault diagnosis accuracy and reliability. A new fault diagnosis method based on multiple neural networks and voting fusion for nuclear power plants was proposed in view of the shortcoming of single neural network fault diagnosis method. In this method, multiple neural networks that the types of neural networks were different were trained for the fault diagnosis of NPP. The operation parameters of NPP, which have important affect on the safety of NPP, were selected as the input variable of neural networks. The output of neural networks is fault patterns of NPP. The last results of diagnosis for NPP were obtained by fusing the diagnosing results of different neural networks by voting fusion. The typical operation patterns of NPP were diagnosed to demonstrate the effect of the proposed method. The results show that employing the proposed diagnosing method can improve the precision and reliability of fault diagnosis results of NPPs