An Intelligent Fault Location Algorithm of High Voltage Lines Using Cascading Deep Network

Abstract

Due to ever-increasing power equipments and the distances of power transmission lines, insulator inspection presents a valuable but challenging issue. As a common insulator defect, missing-cap defects affect the structural strength of power insulators and cause irreparable harm to power supply security. Therefore, insulator defect detection is a basic and critical task for power line inspection. Most detection methods are mainly based on machine learning algorithms. Shallow learning methods rely on handcrafted image features and are always aimed at specific scenarios or prior knowledge. The unbalanced data sets of insulators affect the detection performance of deep learning algorithms. To address the above problems regarding insulator defect detection, a novel detection algorithm based on a cascading deep architecture is proposed for unmanned aerial vehicle (UAV) inspection. Combined with the strong detection performance of deep architecture, an insulator location algorithm based on the improved YOLOV3 model is proposed to remove complex backgrounds such as ”region of interest (ROI) extraction”. On this basis, a novel semantic segmentation algorithm is proposed to realize defect segmentation for small missing-cap defects. Experiments show that the proposed algorithm can satisfactorily meet the precision and robustness requirements of power line inspection compared with other related detection models.