Diagnosis of Commutation Failure in a High- Voltage Direct Current Transmission System Based on Fuzzy Entropy Feature Vectors and a PCNN-GRU

To enhance the diagnostic accuracy of commutation failure in weak receiving-end high-voltage direct current (HVDC) transmission systems, this study proposes a novel diagnostic model that integrates Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN)-based fuzzy entropy and a Parallel Convolutional Gated Recurrent Unit Neural Network (PCNN-GRU). The model initially decomposes the rectifier-side DC voltage signal through CEEMDAN to obtain the high- and low-frequency characteristic components of fault voltages, followed by fuzzy entropy calculation for each component to construct the fault diagnosis input matrix. Subsequently, the PCNN-GRU architecture performs deep feature extraction through two distinct mechanisms: the PCNN branch employs dual-path convolutional kernels of varying sizes for multidimensional feature mining, whereas the GRU network enhances temporal feature extraction capabilities. The simulation results demonstrate the model’s exceptional performance in discriminating DC line faults and commutation failures, achieving diagnostic accuracies of 98.75%, 99.17%, and 99.5833% under 20 dB, 30 dB, and noise-free environments, respectively, significantly outperforming conventional deep learning models. The proposed approach not only improves diagnostic precision, but also exhibits remarkable noise immunity, providing a novel methodology for HVDC commutation failure detection. This breakthrough research offers critical technical support for stable HVDC system operation and holds significant application value for smart grid maintenance.