摘要
变电站屏柜的二次接线复杂,易发生错接、漏接等情况。针对人工检查时存在耗时久、效率低且有漏检的可能性,提出了基于联合训练方法的变电站屏柜接线故障诊断技术,通过联合训练屏柜标签字符定位网络和字符识别网络,可有效提高屏柜接线信息识别的准确率和识别时间。开展验证测试,准确率达到97.9%。开发了变电站屏柜接线故障诊断应用平台和移动端平台,利用此平台,提高了诊断效率,减少了人员工作量,保证了电网安全可靠地运行。
The secondary wiring of the substation panel is complicated, which is easy to be connected to the wrong connection and missed. However, at present, only manual inspection means verify it, which is time-consuming, inefficient, and has the possibility of missed detection, affecting the safe and reliable operation of the power grid. Aiming at the above problems, a fault diagnosis technology for substation panel wiring based on joint training method is proposed. Through jointly training the screen label character positioning network and character recognition network, the accuracy and recognition time of the panel wiring information identification can be effectively improved. Conducted verification tests, the accuracy rate reached 97.9%. A substation screen cabinet wiring fault diagnosis application platform and a mobile terminal platform is developed. Using this platform, the diagnosis efficiency is improved, the workload of personnel is reduced, and the reliable operation of the power grid is guaranteed.
作者
王磊
黄力
张礼波
龙志
李岩
周金桥
WANG Lei;HUANG Li;ZHANG Libo;LONG Zhi;LI Yan;ZHOU Jinqiao(Liupanshui Power Supply Bureau of Guizhou Power Grid Company,Liupanshui 553000,China;Wuhan INRE Power Technology Co.,Ltd.,Wuhan 430000,China)
出处
《供用电》
2020年第5期85-90,共6页
Distribution & Utilization
基金
中国南方电网有限责任公司科技项目(060200KK52180018)。
关键词
联合训练
变电站屏柜
文本识别
文本定位
门递归神经网络
joint training
substation panel
text recognition
text localization
gate recurrent neural network
