Single Phase Grounding Fault Line Selection Method based on Improved Multivariate Variational Mode Decomposition with You Only Look Once Version 10

In recent era, the distribution network generally allows to operate the faults with one to two hours in a single-phase ground fault occurrences. As it leads to the further development of fault and threatens the power system's safe operations. In existing, the Residual Network (ResNet) is utilized for the fault diagnosis of single-phase grounding fault line selection. However, it heavily relies on the quality of data especially with zero-sequence current measurements, which led to the poor performance. Therefore, in this research an Improved Multivariate Variational Mode Decomposition (IMVMD) with You Only Look Once version 10 (YOLOv10) is proposed and named as IMVMD-YOLOv10. This procedure started with collecting data from Electric Fault Detection and Classification (EFDC) dataset. Then, the data is further decomposed with the help of IMVMD to decompose the signals with fault characteristics for better fault detection. After that, Convolutional Neural Network (CNN) is utilized for feature extraction from the decomposed signals. Finally, the YOLOv10 is employed for detecting single phase grounding fault line from the spatial and temporal features. From the results, the proposed IMVMD-YOLOv10 established better results including precision (96%), recall (95%), F1-score (95%) and the Running Time (RT) (0.001 sec) when compared to existing YOLOv3.