多路视频流边缘智能识别设备设计
2022,30(4):263-266
摘要：变电站是电网运行的重要环节，其运维巡检质量关系变电站的安全稳定，变电设备数量增长和人员相对短缺的矛盾导致人工巡检往智能巡检方式转变。变电站视频设备数量众多，视频数据多通过光纤统一传输到后台服务器分析处理，为了解决传输通道带宽压力大，应用服务端计算性能要求高等问题，提高图像识别效率，基于人工智能与边缘计算技术，研制了变电站多路视频流边缘智能识别设备，分别从总体架构、主控软件、以及算法模型等方面开展了设计与开发，使用变电图像数据库，测试了缺陷智能识别算法的精度，并面向变电站智能运维场景开展试点应用验证，模型精度测试与实际应用结果表明，该设备智能识别精度可满足变电智能巡视业务需求，能够在边缘侧实现变电多路视频流的实时采集、在线智能识别、优化推流等功能，可显著提升变电站智能运维水平。
关键词：多路视频流; 人工智能; 缺陷识别; 边缘计算; YOLOv4
Design of Edge Intelligent Recognition Equipment for Multiple Video Streams in Transformer Station
Abstract：Substations are an important part of the power grid. The quality of substations’ maintenance is related to the safety and stability of the power grid. The number of devices in substations increases rapidly, but the number of personnel for maintenance is relatively shortage, which leads to the transformation of substation maintenance from manual inspection to intelligent inspection. There are a large number of video cameras in substation, and the video data is generally transmitted to the back-end server for analysis and processing through optical fibers. In order to solve the problem of high bandwidth pressure of transmission channel, high computational performance requirements of application server, and to improve the image recognition efficiency, the edge intelligent recognition equipment for multiple video streams in substations was developed based on artificial intelligence and edge computing technology. The design and development were carried out from the aspects of overall architecture, main control software, algorithm model, and so on. The accuracy of the defect intelligent identification algorithm was tested by using the substation image database, and the pilot application verification was carried out for the intelligent operation and maintenance scene of the substation. The results of algorithm model accuracy test and the practical application showed that the intelligent identification accuracy of the equipment can meet the requirements of the intelligent patrol business of the substation. The equipment can realized the real-time collection, online intelligent recognition, and optimized streaming of multiple video streams in substations at the edge side. The intelligent level of substation’s maintenance can be significantly improved.
Key words：multiple video streams; artificial intelligence; defect recognition; edge computing; YOLOv4
收稿日期：2021-11-01
基金项目：
