Hyperparameter-tuned variational Bayesian Gaussian mixture model-based fault time detection method for distribution networks with distributed generation

The fault self-synchronization method enables cost-effective data synchronization for differential protection in distribution networks but suffers from synchronization errors under complex faults due to detection algorithm sensitivity. To address this, the paper analyzes the delay mechanism in active distribution network fault detection and factors affecting synchronization errors. A framework combining curvature analysis and mathematical morphology is proposed to enhance fault feature identification in current signals. Curvature analysis extracts and amplifies fault features, while mathematical morphology preprocesses data, improving signal quality and reducing distortion. Additionally, a fault time detection method based on Hyperparameter-Tuned Variational Bayesian Gaussian Mixture Models is introduced. This method eliminates redundant Gaussian components for optimal modeling and uses a sliding window for adaptive clustering, ensuring precise fault detection. Simulations confirm its effectiveness in rapid fault detection, maintaining synchronization accuracy, and exhibiting robustness to noise, initial fault phase angles, and sampling frequency variations.