Three-Core Cable Fault Line Identification Based on Ground Wire Current and BiGRU-ResNet-MA

Faults in three-core armored cable lines in distribution networks are frequent, and using only zero-sequence current time-frequency data to identify fault lines has its limitations. To increase the accuracy of cable fault identification, this paper utilizes the cable ground wire current and proposes a new fault line identification method that integrates a Bidirectional Gated Recurrent Unit (BiGRU), a Residual Network (ResNet), and a Multi-head Attention Mechanism (MA). Firstly, the Northern Goshawk Algorithm optimizes Variational Mode Decomposition (NGO-VMD) for denoising the ground wire current signal, reducing interference from the complex external environment. Secondly, the processed current signal is used to generate two-dimensional images using the Markov Transition Field (MTF), expanding signal features. Finally, BiGRU-ResNet-MA is utilized to deeply extract, enhance, and identify the multi-dimensional features of the ground wire current signal. The BiGRU-ResNet-MA model combines the two-dimensional image spatial characteristic extraction capability of ResNet with the temporal characteristic extraction capability of BiGRU, and further enhances feature weighting through MA, improving the fault line identification capability. The simulation findings show that this identification method can effectively address the issue of traditional methods relying too heavily on subjective selection. It accurately identifies fault lines even in the presence of strong noise, unstable faults, and asynchronous sampling, showcasing its strong robustness.