基于深度学习的厂站一次接线图拓扑关系检测  

Topological Relationship Detection of Substation Primary Wiring Diagram Based on Deep Learning

作　　者：刘萍 吐松江·卡日[1] 阮佳阳 徐丽 LIU Ping;Kari·Tusongjiang;RUAN Jiayang;XU Li(School of Electrical Engineering,Xinjiang University,Urumqi Xinjiang 830046,China;Beijing Zhimeng I&T Tech Co.,Ltd.Beijing 100053,China;Hohhot Power Supply Company,Hohhot Inner Mongolia010050,China)

机构地区：[1]新疆大学电气工程学院,新疆乌鲁木齐830046 [2]北京智盟信通科技有限公司,北京100053 [3]呼和浩特市供电公司,内蒙古呼和浩特010050

出　　处：《计算机仿真》2024年第1期541-547,共7页Computer Simulation

基　　金：国家自然科学基金(52067021);新疆优秀青年科技人才培养项目(2019Q012);新疆维吾尔自治区自然科学基金(2022D01C35);新疆大学博士启动基金(BS190221)。

摘　　要：目前调度控制系统中厂站一次接线图的绘制采用人工绘制、录入的方式,由于图形样式复杂,设备类型众多,极易出现元件缺失、关联错误、连接线虚接等问题。针对这些问题,提出了一种基于深度学习和改进概率霍夫变换相结合的厂站一次接线图的识别算法。首先,利用基于改进Faster-RCNN对电气元件、文本框进行识别,并获取其位置信息。然后,利用改进概率霍夫变换检测母线与连接线。最后,根据检测到的元件、母线和连接线,设定距离阈值、构建图结构,确定各元素的关联关系。实验证明,本文所提方法的母线检测准确率高达100%;对于厂站一次接线图拓扑关系检测,其准确率、召回率、综合F1值分别达89.8%、88.6%、89.2%,与Faster-RCNN、YOLOv5等方法对比,准确率提升显著,能够满足厂站一次接线图自动识别的要求。At present,the substation wiring diagram in the dispatching and control system is drawn and input manually.Due to the complex graphics style and numerous equipment types,it is very easy to have problems such as missing components,association errors,virtual connection of connecting lines,etc.To address these problems,a recognition algorithm based on deep learning and improved progressive probabilistic Hough transform(PPHT)is proposed in this paper.Firstly,the object detection method based on improvedFaster-RCNN was used to identify the electrical components and text box to obtain their location information.Then,the improved PPHT was adopted to detect busbar and connecting line based.Finally,the mutual relationship of all elements was ascertained after determination of the distance threshold and graph structure.The experimental results show that the proposed method has achieved 100%accuracy in busbar detection.Besides,the accuracy,recall and comprehensive F1 values of the topology relationship detection of substation primary wiring diagram are 89.8%,88.6%and 89.2%,respectively.Compared with the Faster-RCNN and YOLOv5 methods,the detection performance is improved significantly,which indicates that the proposed method can satisfy the requirement of automatic identification of substation primary wiring diagram.

关 键 词：厂站一次接线图 深度学习 目标检测 概率霍夫变换 

分 类 号：TM734[电气工程—电力系统及自动化] TP391.41[自动化与计算机技术—计算机应用技术]