Research or Anomaly Detection Model for Distribution Grids Using Deep Learning-Based Object Detection

In the modern society, power infrastructure is an essential energy supply chain, and its stability and efficient management are emerging as very important tasks. Recently, as the number of cases of unauthorized communication installations in distribution workplaces has increased rapidly, it is recognized as a risk factor that threatens the structural stability of power facilities, and the current abnormality of power facilities mainly relies on visual inspection, making it difficult to efficiently maintain them. This study proposes a model for detecting abnormalities in distribution workplaces using computer vision and deep learning technology. In particular, in order to overcome the limitations of the existing object detection technology accurately identifying objects with a thin and complex structure like wires, we propose an abnormality detection system that combines YOLOv11l-seg and 2DCNN. This study improved the detection performance in various environments by applying segmentation-based labeling and data augmentation techniques, and verified the excellence through comparative experiments with various models such as YOLOv8l-seg, YOLOv9c-seg, YOLOv11l-seg, Mask R-CNN+ResNet50+FPN, and ConvNeXT. As a result of the study, the proposed YOLOv11-seg-based model effectively detects structural changes and abnormalities in power facilities and presents the practical possibility of deep learning-based technology that improves the efficiency and stability of power infrastructure management.