Abstract: The fault identification accuracy in transmission lines is crucial for improving the reliability of power supply in the power system. To solve the problem that existing methods are difficult to effectively identify complex external factors such as tree discharge, wind deviation, lightning strikes, bird damage, wildfires, external damage, and foreign objects caused by power lines, a fault identification method for transmission lines based on frequency features and an improved KAN network is proposed. Firstly, the three-phase data under different faults such as external force damage, wildfire, foreign objects, etc. are analyzed, and a fractional frequency transformation method is introduced to extract the depth features of the faults. Then, a self-attention convolutional Kolmogorov-Arnold network (SCKAN) incorporating self-attention and convolution module is proposed, and the improved wavelet basis function is used for network weight initialization. Finally, the proposed methods effectiveness is verified through experiments based on the collected real data of transmission lines. The results show that the proposed method greatly enhances the ability to identify transmission line faults.

Key words: frequency feature, attention mechanism, Kolmogorov Arnold network, transmission line fault identification