Faulty feeder detection based on image recognition of current waveform superposition in distribution networks

Faulty feeder detection is essential for maintaining the security and stability of energy supply in distribution networks. However, it is rather difficult to identify a specific faulty feeder owing to small fault currents and complex fault transients. To improve the detection accuracy, this study proposes a faulty-feeder detection method based on image recognition of superimposed zero-sequence currents. A convolutional neural network (CNN) is utilized to recognize the superimposed currents in the same plot, rather than a raw single current, which can realize correlation comparisons between the currents. In addition, the zero-sequence currents of different feeders are superimposed according to a specific sequence, and the CNN can adapt to the changing topologies of distribution networks while conducting correlation comparisons. Because zero-sequence currents decay rapidly over time, an attention learning block is embedded into the CNN to enhance the discriminative capability. A total of 14,718 sets of experimental data obtained from simulations and practical distribution networks were collected to verify the effectiveness of the proposed method. Comparisons with other traditional methods and learning-based methods adopted in previous studies justify the superiority of the proposed method in terms of detection accuracy and detection efficiency. Therefore, the proposed method can be implemented in real distribution networks for faulty feeder detection.(c) 2022 Elsevier B.V. All rights reserved.