Abstract: In order to solve the problem of difficult location after a single-phase grounding fault in the wind farm collector line, a fault section location method based on the Variational Mode Decomposition-Continuous Wavelet Transform (VMD-CWT) time frequency spectrum combined with siamese Vision Transformer (sViT) is proposed. It is found that the fault section is closely related to the VMD-CWT spectrum of the fault voltage of the collector, the fault section location of the collector line can be realized by mining the relationship between the spectral line and the fault section by using the deep learning algorithm. With the help of PSCAD/EMTDC software to build the collector line model, collect the data of various fault conditions, and generate the time frequency spectrum of VMD-CWT transformation; the optimal recognition parameters of the sViT network will be found on the training set, and the branch of this network will be used for test set recognition.The simulation shows that the method has good adaptability to multi-branch collector lines and mixed short lines, and the positioning is less affected by transition resistance, noise and fault phase angle.

    Key words: siamese vision transformer network; variational mode decomposition-continuous wavelet transform; wind farm collector line;fault section location