GIS Partial Discharge Patterns Recognition with Spherical Convolutional Neural Network

The ubiquitous construction of the power Internet of Things provides a new idea for the real-time and accurate diagnosis of GIS partial discharge online monitoring fault diagnosis. However, the traditional partial discharge fault diagnosis method is difficult to solve the problem that the fault information of different online monitoring systems is different from the reference axis. In order to solve the problem that the fault information is difficult to identify in rotation and transformation, and improve the accuracy of fault diagnosis, this paper proposes a spherical convolutional neural network based on complex data sources. First, the PRPS picture transmitted to the ubiquitous power Internet of Things terminal is selected as the fault feature information. Secondly, a generalized Fourier algorithm (GFT) algorithm is used to construct a spherical convolution structure for PD pattern recognition. The algorithm can perform automatic feature extraction. Thirdly, the spherical convolutional neural network-based PD recognition method is applied to processing of the complex data sources with 84.88% average accuracy rate. It shows that the PRPS 3D map is one of effective way to avoid the complexity of artificial feature extraction for spherical CNN and in the meantime, it can also improve the accuracy of fault diagnosis.