Research on high voltage circuit breaker fault diagnosis using multi-strategy improved dung beetle algorithm optimized CNN-BiGRU-attention model

Accurate fault diagnosis of high-voltage circuit breakers is essential for improving power-system reliability. However, coil-current signals are often affected by environmental interference and random fluctuations in the measurement circuit, which may reduce diagnostic accuracy and lead to unstable recognition results. To address this problem, this paper proposes a fault diagnosis method based on opening and closing coil-current signals of high-voltage circuit breakers.First, an improved wavelet-threshold variational mode decomposition method is used to suppress noise while preserving fault-sensitive waveform characteristics. Then, extrema, temporal features, and waveform statistical features are extracted from the denoised signals. A CNN-BiGRU-Attention model is constructed to learn local waveform features and temporal dependencies. The model parameters are further optimized using a multi-strategy improved dung beetle optimization algorithm to improve parameter search performance and reduce the risk of local optima.Experiments were conducted on a physical ZN63–12 high-voltage vacuum circuit breaker platform under controlled laboratory fault-simulation conditions. The proposed method achieved an overall diagnostic accuracy of 96.5% across five operating states. The results indicate that the method can effectively distinguish normal operation, inter-turn short circuit, poor contact, overvoltage, and undervoltage conditions within the tested experimental scope. This study provides a potential data-driven approach for coil-current-based fault diagnosis of high-voltage circuit breakers, while further validation on additional breaker types and field operating conditions is still required.