发电技术 ›› 2025, Vol. 46 ›› Issue (3): 532-540.DOI: 10.12096/j.2096-4528.pgt.24230

• AI在新型电力系统中的应用 • 上一篇    

基于无人机巡检的输电线路绝缘子及其异物检测算法

于子涵1, 王赫鸣1, 王建凯1, 朱胜强1, 孟祥忠2   

  1. 1.山东黄金电力有限公司,山东省 烟台市 266040
    2.青岛科技大学自动化与电子工程学院,山东省 青岛市 266061
  • 收稿日期:2024-10-29 修回日期:2024-11-29 出版日期:2025-06-30 发布日期:2025-06-16
  • 通讯作者: 孟祥忠
  • 作者简介:于子涵(1986),女,工程师,研究方向为智能变电站建设、电力智能巡检、电力系统大数据监控,690470882@qq.com
    王赫鸣(1989),男,硕士,工程师,研究方向为电力智能巡检、智能变电站建设、电力无人机巡检与控制,13465505668@139.com
    王建凯(1989),男,工程师,研究方向为智能变电站建设、电力智能巡检、电力系统开发与设计,807643738@qq.com
    朱胜强(1973),男,工程师,研究方向为智能变电站建设、电力智能巡检、电力系统监控与开发,595221206@qq.com
    孟祥忠(1964),男,博士,教授,研究方向电力电子技术、电力系统自动化技术、智能电力巡检,本文通信作者,woaiyishanyishui@126.com
  • 基金资助:
    山东省自然科学基金项目(ZR2022ME194)

Detection Algorithm for Insulators and Foreign Objects on Transmission Lines Based on Unmanned Aerial Vehicle Inspection

Zihan YU1, Heming WANG1, Jiankai WANG1, Shengqiang ZHU1, Xiangzhong MENG2   

  1. 1.Shandong Golden Electric Power Limited Company, Yantai 266040, Shandong Province, China
    2.College of Automation and Electronic Engineering, Qingdao University of Science & Technology, Qingdao 266061, Shandong Province, China
  • Received:2024-10-29 Revised:2024-11-29 Published:2025-06-30 Online:2025-06-16
  • Contact: Xiangzhong MENG
  • Supported by:
    Shandong Provincial Natural Science Foundation(ZR2022ME194)

摘要:

目的 传统的电网巡检方式存在劳动强度大、效率低等问题,为此,以山东黄金电网为研究对象,提出一种基于轻量化深度学习网络YOLOv5-Mv3的无人机巡检电网绝缘子及其异物检测算法。 方法 首先,通过无人机巡检电网拍摄图片来构建数据集,对数据集进行训练;其次,针对电网绝缘子及其异物,采用Mobilenetv3来替代CSPDarknet53作为特征提取网络,对YOLOv5-Mv3进行轻量化改进,减少模型参数和计算量,在保证准确率的同时,满足实时检测的要求。 结果 该检测算法的均值平均精度达到84.7%,检测速度达到56.6帧/s,改进的YOLOv5-Mv3与Faster RCNN、SSD、YOLOv4模型相比,具有更高的检测精度和更快的检测速度。 结论 该算法提高了无人机巡检电网效率,实现了轻量高效的效果,符合电网智能巡检的要求。

关键词: 电力系统, 人工智能(AI), 电网, 无人机, 目标检测, 输电线路, 电力巡检, 机器视觉, 图像处理

Abstract:

Objectives Traditional power grid inspection methods suffer from high labor intensity and low efficiency. Taking Shandong Golden Power Grid as the research object, this study proposes an inspection algorithm using unmanned aerial vehicle (UAV) based on lightweight deep learning network YOLOv5-Mv3 for detecting grid insulators and foreign objects. Methods Firstly, a dataset is constructed using images captured by UAVs during power grid inspection and is trained. Then, for the grid insulators and foreign objects, Mobilenetv3 is used to replace CSPDarknet53 as the feature extraction network in order to lighten the YOLOv5-Mv3 model, reducing parameters and computational cost while maintaining accuracy and enabling real-time detection. Results The proposed detection algorithm achieves a mean Average Precision of 84.7% and 56.6 frames per second. Compared to Faster RCNN, SSD, and YOLOv4 models, the improved YOLOv5-Mv3 demonstrates higher detection accuracy and faster performance. Conclusions The proposed algorithm improves the efficiency of UAV-based power grid inspection and achieves lightweight and high-efficiency effect, fully meeting the requirements for intelligent power grid inspection.

Key words: power system, artifical intelligence (AI), power grid, unmanned aerial vehicle, target detection, transmission line, power inspection, machine vision, image processing

中图分类号: