Abstract: In this study, a single-ended fault diagnosis method for flexible DC power grids based on Swin Transformer is proposed to address the problems of low precision, susceptibility to transition resistance, and requiring manual input to set the threshold in the existing fault detection methods of flexible DC power grids. First, we collect the transient voltage time-domain data at fault and convert it into a two-dimensional gramian angular field（GAF） image with a better recognition effect after data processing, which is used for offline training of the Swin Transformer; Second fault features are extracted using the moving window of the Swin Transformer, and different fault diagnoses are realized according to the training results. This method does not require manual setting of the threshold. Finally, after many simulations, it is proven that the method proposed in this study satisfies the quick action requirement, can accurately diagnose faults, and has strong transition resistance and anti-noise ability.