A Novel Sudden Fault Prediction Method Based on Hierarchical Structure with GRU Neural Network for X-ray High-voltage Power Supply. (June 2020)

• Record Type: Journal Article
• Authors: Zhang, Jianlong; Li, Qiao; Wang, Bin; Cui, Mengying
• Is Part Of: Journal of physics. Volume 1576(2020)
• Publication Date: 2020-06
• Subjects: Physics -- Congresses; 530.5
• DOI: 10.1088/1742-6596/1576/1/012020
• Languages: English
• ISSNs: 1742-6588

Abstract

Abstract: Considering the sudden fault feature of X-ray high-voltage power supply are mostly reflected in the high-frequency components, a hierarchical structure with time sequential neural network is proposed to predict sudden fault of X-ray high-voltage power supply in this paper. Gate recurrent unit is the basic unit of time sequential neural network. Firstly, multi-wavelet transform is used to decompose the signal of the X-ray high-voltage power signal to obtain the coefficients of each frequency band. Secondly, signal waveform reconstructed for each layer of wavelet coefficients is as the input of the gate recurrent unit network to predict signals at different frequencies. Finally, each gate recurrent unit performs multi-step prediction on the reconstructed waveform in this frequency band for sudden fault and all band prediction outputs are composed to obtain the final prediction results. The simulation experiment shows that our method has better performance than the Kalman filter, recurrent neural network and long-short term memory prediction methods, and can accurately predict sudden fault of X-ray high-voltage power supplies.