基于云/边协同的输电通道隐患目标自学习识别系统研发  

Research and Development of Self-Learning Identification System for Hidden Danger Object of Transmission Channel Based on Cloud-Edge Collaboration

作　　者：陶汉涛 李健 姜志博[1,2] 吴大伟 白冰洁[1,2] 张磊 TAO Hantao;LI Jian;JIANG Zhibo;WU Dawei;BAI Bingjie;ZHANG Lei(NARI Group Co.,Ltd.,Nanjing 211106,China;State Grid Electric Power Research Institute,Wuhan NARI Co.,Ltd.,Wuhan 430074,China)

机构地区：[1]南瑞集团有限公司,江苏南京211106 [2]国网电力科学研究院武汉南瑞有限责任公司,湖北武汉430074

出　　处：《电力信息与通信技术》2022年第10期36-43,共8页Electric Power Information and Communication Technology

基　　金：国家电网有限公司总部科技项目资助“特高压密集通道应对极端环境条件智能监测及防护能力提升技术研究”(524625210017)。

摘　　要：为解决现有输电线路通道隐患自动识别系统中海量图像数据导致的网络负载、云端计算量大以及图像识别算法缺乏自学习迭代更新机制的问题,提出基于人工智能计算平台的输电线路通道隐患自学习识别系统。首先,所提系统基于云/边协同计算策略,将推理计算前移至边缘计算设备端,模型训练在云端进行,更新后的算法再部署至边端,减少网络传输和云端计算压力;其次,采用自学习机制,通过对隐患目标进行识别与跟踪,应用插值计算自主发现并标注典型图像帧,传至云端迭代学习,实现云/边自学习更新;最后,所提系统在特高压输电线路智慧物联监控平台中进行应用。实验结果表明,基于云/边协同的智能识别算法架构有效降低网络传输负载和云端计算压力,实现实时性、自动化识别,采用自学习机制,目标识别算法准确率在96%以上。In order to solve the problem of network load and large amount of cloud computing caused by massive image data and the problem of lack of self-learning iterative update mechanism of image recognition algorithm in the existing automatic identification system of hidden danger of transmission line channel, this paper proposes a self-learning identification system of transmission line channel hidden danger based on artificial intelligence computing platform. First, based on the cloud-edge collaborative computing strategy, the proposed system moves forward the reasoning calculation to the edge computing device, conducts model training in the cloud, and deploys the updated algorithm to the edge, reducing the pressure of network transmission and cloud computing. Secondly, the self-learning mechanism is used to identify and track hidden danger objects, and the interpolation calculation is used to find and label typical image frames, which are transferred to the cloud for iterative learning to achieve cloud-edge self-learning update. Finally, the proposed system is applied in the intelligent Internet of things monitoring platform for UHV transmission lines. The experimental results show that the intelligent recognition algorithm architecture based on cloud-edge collaboration effectively reduces the network transmission load and cloud computing pressure, realizes real-time and automatic recognition, and adopts self-learning mechanism, and the accuracy of object recognition algorithm is more than 96%.

关 键 词：输电线路 通道隐患 云/边协同 目标检测算法 自学习机制 

分 类 号：TM74[电气工程—电力系统及自动化]