基于视觉深度学习算法的变电站智能巡检应用研究  

Application Research of Substation Intelligent Inspection Based on Visual Depth Learning Algorithm

作　　者：赵振喜 王朔 刘春生 郭玉福 王朝辉 武桐 范恩洪 侯林江 ZHAO Zhenxi;WANG Shuo;LIU Chunsheng;GUO Yufu;WANG Chaohui;WU Tong;FAN Enhong;HOU Linjiang(State Grid Jilin Electric Power Company Limited,Changchun 130028,China;State Grid Jilin Electric Power Company Limited Construction Branch,Changchun 130021,China;Weicheng Intelligent Power Technology(Hangzhou)Company Limited,Hangzhou 310000,China)

机构地区：[1]国网吉林省电力有限公司,长春130028 [2]国网吉林省电力有限公司建设分公司,长春130021 [3]炜呈智能电力科技(杭州)有限公司,杭州310000

出　　处：《吉林电力》2022年第4期1-4,共4页Jilin Electric Power

基　　金：国网吉林省电力有限公司科技项目(522371210003)。

摘　　要：针对变电站智能巡检,在变电站传统辅助系统及前期布设或配备的视频监控装置、机器人巡检、单兵作业装备等设备接入层基础上,通过边缘物联代理,以及人工智能(artificial intelligence, AI)分析服务器内置集成AI视觉分析算法和模型,构建分析服务层,以实现变电站智能巡检设备状态实时监测、站端表计智能识别、环境智能实时监测、人员安全智能评估、安全作业监控、电子虚拟围栏等业务场景;同时,介绍了卷积神经网络主要技术路线,着重介绍了以站内烟火作为实例对视觉学习算法构建成熟实用的视觉学习模型,验证了其技术先进性、可行性,减少了变电站巡检人员工作量,提高了站内事故隐患分析和处理效率,从而较大地提高了变电站安全管理水平。For intelligent inspection of substation, based on the traditional auxiliary system of substation and the equipment access layer such as video monitoring device, robot patrol inspection and individual operation equipment deployed or equipped in the early stage, this paper constructs the analysis service layer through the edge IoT agent and AI intelligent analysis server with built-in integrated artificial intelligence visual analysis algorithm and model, so as to realize the real-time monitoring of the status of intelligent patrol inspection equipment in substation Intelligent identification of station meters, intelligent real-time monitoring of environment, intelligent assessment of personnel safety, safe operation monitoring, electronic virtual fence and other business scenarios. At the same time, this paper introduces the convolution neural network technology as the main technical route, especially taking the station fireworks as an example to introduce a visual learning algorithm how to build a mature and practical visual learning model, and verify its advanced technology and feasibility. Through the research topic of this paper, the workload of substation patrol inspection personnel is greatly reduced, the analysis and treatment efficiency of accident hidden dangers in the station are improved, and the safety management level of substation is greatly improved.

关 键 词：边缘物联代理 卷积神经网络 深度学习 人工智能 智能巡检 图像标注 

分 类 号：TM63[电气工程—电力系统及自动化] TP277[自动化与计算机技术—检测技术与自动化装置]