摘要
针对现有变电站视频监控系统人工巡视模式效率低下的问题,本文基于AI边缘计算技术建立变电站视频深度学习识别模型,构建了适用于变电站现场的电力专用视频智能识别计算单元,采用卷积神经网络算法实现了“烟火”“安全帽”“异物”“画面质量”四大电力定制场景的融合识别,形成“现场视频分析装置+主站云分析平台”的“云-边”协同视频智能分析系统。在云端构服务器实现对现场上传识别结果的二次深度分析,并通过主站样本库积累及反向传输机制使装置具备持续学习的能力。试运行阶段的实际数据表明,四大场景平均识别准确达到90%以上,系统实现了变电站视频画面内容的自动识别,极大提高变电站视频监控的质量和效率,具有很强的实用价值和推广价值。
Aiming at the low efficiency of the manual patrol mode of the existing substation video monitoring system,this paper establishes a substation video deep learning recognition model based on AI edge computing technology,and constructs a power dedicated video intelligent recognition calculation unit suitable for substation site,and uses convolutional neural network algorithm to realize the fusion recognition of the four power customization scenarios of“fireworks”“safety helmet”“foreign objects”and“picture quality”,forming a“cloud-side”collaborative video intelligence of“on-site video analysis device+master station cloud analysis platform”.The analysis system implements the secondary in-depth analysis of the on-site upload recognition results on the cloud server,and enables the device to have the ability to continue learning through the accumulation of the master station sample library and the reverse transmission mechanism.The actual data in the trial operation stage shows that the average recognition accuracy of the four major scenes is more than 90%,and the system realizes the automatic recognition of the substation video screen content,which greatly improves the quality and efficiency of the substation video monitoring,and has strong practical value and promotion value.
作者
王浩
王功臣
娄德章
刘永
张乐
付娟娟
WANG Hao;WANG Gongchen;LOU Dezhang;LIU Yongi;ZHANG Le;FU Juanjuan(Xuzhou Power Supply Branch of State Grid Jiangsu Electric Power Co.,Ltd.,Xuzhou 221000 Jiangsu,China)
出处
《电力大数据》
2021年第11期1-8,共8页
Power Systems and Big Data
关键词
视频监控
边缘计算
卷积神经网络
电力场景
智能识别
video surveillance
edge computing
convolutional neural network
power scene
intelligent recognition
