Abstract: To address the issues of insufficient fault diagnosis samples and low diagnostic accuracy in traditional circuit breakers, this paper proposes a fault diagnosis method based on Wasserstein generative adversarial network with gradient penalty (WGAN-GP) and bidirectional gated recurrent unit (BiGRU). This method uses WGAN-GP to augment the circuit breaker data, which improves the instability of the original model training process and obtains high-quality sample data. Then, combined with the augmented dataset, the BiGRU model is used for fault diagnosis classification training. The experimental results show that, under the same test set, the proposed diagnostic method can effectively enhance the model’s ability to identify few-shot faults and improve the accuracy of fault diagnosis.

    Key words: circuit breaker fault diagnosis; generative adversarial network; bidirectional gate recurrent unit; data augmentation; gradient penalty