关键词: 故障选线, 格拉姆角差场：卷积注意力机制, Swin Transformer

Abstract: Due to the weak fault characteristics of single-phase grounding fault in the small current system of distribution network, the existing fault line selection methods often have problems such as low accuracy and weak robustness. Therefore, this study proposes a fault line selection method for distribution network based on BMF-GADF and improved Swin Transformer. Firstly, the zero-sequence current is converted into a feature-enhanced Gram angle difference field image by using the combination of Butterworth Mean Filtering and Gramian Angle Difference Field. Then, the image samples are sent to the improved Swin Transformer model for feature extraction. The improved Swin Transformer innovatively introduces the module parallel Convolution Block Attention Module on the basis of the original architecture, which realizes more accurate feature adaptive selection and effectively improves the accuracy of the model. Finally, the line selection results are output by Softmax. The experimental results show that the accuracy of the proposed method is 98.75 %. Compared with other fault line selection methods, the proposed method has higher line selection accuracy and noise robustness, and provides a new scheme for fault line selection in distribution network.

Key words: Fault line selection, GADF, CBAM, Swin Transformer