改进FasterR-CNN的变电站电气主接线图图元检测

打开文本图片集

关键词：变电站；接线图；图元检测；FasterR-CNN优化算法；深度学习

DOI：10. 15938/j. jhust. 2025.04.005

中图分类号：TP391.4;TP183;TM63 文献标志码：A 文章编号：1007-2683（2025）04-0039-09

Improved Faster R-CNN Forsubstation Electrical Main Wiring Diagram Element Detection

IG Bing'，DU Yuefan1， JIN Yao’， ZONG Xiangrui'， JIN Hua²，LIU Tanjing²，WANG Cong 3 （204 （1.High Voltage Branch， State Grid Tianjin Electric Power Company，Tianjin 3OO232，China; 2.School of Control and Computer Enginering，North China Electric Power University，Beijing 102206，China; 3.School of Electrical and Electronic Engineering，North China Electric Power University，Beijing 102206，China）

Abstract：Aimig atthe problems oflowaccuracyofelementdetectionandhighfalsedetectionrateandleakagerateofsmaltarget elements nelectricalmainwiringdiagramsofsubstations，amethodfordetecting elements intheelectricalmain wiringdiagrasof substationsbasedoimprovedFasterR-CalgoritisproposedFirst，adeepresiualnetworkstructureisintroducedtoeplace theoriginalfeatureextractionnetworkofFasterR-CNNtoenhancethefeatureextractioncapabilityofmulti-scalediagramelement targets.Then，afeaturepyramidnetworkisintroduced tofusetheshalowfeatureinformationwiththedepfeatureinforationto improvethedetectionperformaneoftheeenetworkforsalltargetsFinally，asedonthescaledistrbutioncharacteristsfsal targets，theparametervaluesofAnchorreresetintheregionalproposalnetworktofurtherimprovethedetectionperformancofsal targets.The experimental results show that the average detection accuracy of the improved algorithm reaches 88.9% ，which is 4.2% （2号 higher thanthatof theoriginal algorithm，and ithas higherdetectionaccuracyandlower falseand missddetectionrates.

Keywords：substation；wiring diagram；element detection；faster R-CNN optimization algorithm；deep learning

0引言

变电站电气主接线图是表示站内一次设备连接关系的电路图纸，主要用于站内电气设备的选型、安装、调试运行维护以及故障分析。（剩余13425字）