Fault Line Selection Method for Non-effectively Grounded Distribution Network

In order to deal with the issue of low accuracy of the single-phase ground fault line selection method in conventional non-effectively grounded distribution network, this paper proposes a Gramian Angular Field and convolutional neural network (CNN) method. First, the Gramian Angular Field is used to convert the one-dimensional zero-sequence current signal into a two-dimensional image to form a single-phase ground fault data set. This method retains the original data characteristics and makes the fault characteristics more prominent and facilitates CNN processing. Then, the two-dimensional CNN is trained by inputting a training set, and then input into a test set for testing. Finally, the parameters are optimized to obtain the ideal results by repeated training iterations. The experimental data show that the algorithm proposed in this paper has a higher accuracy comparing to other methods in distribution network.