Partial Discharge Pattern Recognition in High Voltage Transmission Cable Project

This paper aims to improve the reliability and safety of cables, prolong the service life of cables and reduce the operating cost of power systems. The paper examines in detail how partial discharge (PD) patterns are identified in high-voltage transmission cable projects. This paper tests the performance of several layers of a convolutional neural network (CNN) to increase recognition efficiency and accuracy. During the experiment, unified network parameters are set, including the use of maximum pooling mode, ReLU activation function, and 32 33-size convolution kernels in each layer. By contrasting the recognition accuracy of 1-layer, 2-layer, 3-layer, 4-layer, 6-layer, and 9-layer CNN, the paper’s findings indicate that: (1) CNN’s recognition accuracy peaks at 90.77% when there are two network layers. However, as the number of layers increases, the accuracy of recognition gradually declines. This result reveals that moderately increasing the network depth is helpful to improve the performance in PD pattern recognition of high voltage transmission cables, but too deep network may lead to over-fitting. (2) CNN is stable in the detection of most defect types, while Support Vector Machine (SVM) is the best in some specific defects. Back propagation neural network (BPNN) performs well in some cases, but its overall performance fluctuates greatly. These results are influenced by many factors, such as defect characteristics, model parameters and data set size. Therefore, in practical application, it is necessary to select a suitable model for defect detection according to specific scenarios and requirements. This paper provides a useful reference for the follow-up research in high voltage transmission cable project.