Research on Single Phase to Ground Fault Line Selection Based on Convolutional Neural Network and Recurrence Plot
Ye Tian, Geng Wang, Jinsheng Li, Hao Xu, Tao Zhang, Hao Zheng · 2024
When the convolutional neural network is used for line selection in resonant grounding systems, it not only requires much data sets to determine the parameters of the convolution filter, but also has high requirements on the computing performance of the hardware equipment. To solve the above problems, a single-phase ground fault line selection method based on the fusion of recurrence plot and convolutional neural network(RP-CNN) is proposed. Firstly, FIR filter is used to select the power frequency component of the zero sequence current of lines and generate a recurrence plot. After 90% of the data points in the recurrence plot are set to zero. Then the newly generated recurrence plot is used as the image data to train the model of convolutional neural networks. Finally, the trained model is used to find the fault line and verify the correctness of the lines selection. The simulation results show that because the input image matrix contains a large number of zero matrices, the convolutional neural network only needs a few epochs to complete the training, and it also reduces the requirements for the computational performance of hardware. Not only that, this method can solve the problem of intermittent arc ground fault difficulties, noise interference, and the impact of different data window sizes on the correctness of line selection.