Sensor Fault Diagnosis of Control System Based on CMKPCA and BP Neural Network

Higher levels of automation and intelligence put higher demands on fault diagnosis. BP neural network has great advantages in fault identification and classification, but its efficiency and accuracy are restricted in practical application. In this paper, Class Mean Kernel Principal Component Analysis (CMKPCA) is applied to the optimization of BP neural network. By extracting the characteristic information of the fault samples, this method highlights the main components that cause faults and at the same time achieves lossless dimensionality reduction. The processed data will reduce the complexity of the neural network and further optimize the network parameters. The training result will be compared with that only using BP neural network. Simulation shows that this method can improve the recognition rate of fault diagnosis.