基于图卷积整图分类的低压台区短路故障研判方法  

Research and Judgment Method of Short Circuit Faults in Low Voltage Substation Based on Graph Convolution and Whole Graph Classification

作　　者：余盛达 沈文科 梁杰 郑有武 安禹铮 杨银 Yu Shengda;Shen Wenke;Liang Jie;Zheng Youwu;An Yuzheng;Yang Yin(Wenchang Power Supply Bureau,Hainan Power Grid Co.,Ltd.,Wenchang,Hainan 571300,China;Guangzhou Power Electrical Technology Co.,Ltd.,Guangzhou 510700,China)

机构地区：[1]海南电网有限责任公司文昌供电局,海南文昌571300 [2]广州市奔流电力科技有限公司,广州510700

出　　处：《机电工程技术》2023年第9期215-219,共5页Mechanical & Electrical Engineering Technology

基　　金：海南电网有限责任公司科技项目(HNKJXM20200240)。

摘　　要：低压台区的故障识别不仅是台区运行管理的重要手段,同时台区的故障会对低压用户直接造成人身安全问题。现有的低压台区故障识别手段通常采用客户上报结合人工现场勘察,耗时费力。随着透明电网的建设,低压台区的运行数据为低压台区的故障识别提供了有力支撑。因此提出了基于图卷积神经网络整图分类的低压台区故障识别方法。首先,将台区用户作为节点,自动生成台区拓扑邻接矩阵;其次,通过智能设备终端采集台区发生故障时节点的故障特征数据并降维处理,与邻接矩阵形成成对的映射图数据。最后将整个台区的图数据作为整图输入至图卷积神经网络,通过平均池化图的特征数据得到整图的节点表示,最终输出对应的短路类型。最后仿真验证了本文方法在低压台区短路故障类型研判中准确率较高,达到90%。The fault identification of low-voltage substation area is not only an important means of substation area operation management,but also a direct cause of personal safety problems for low-voltage users.The existing low-voltage substation fault identification methods usually use customer reporting combined with manual on-site investigation,which is time-consuming and laborious.With the construction of transparent power grid,the operation data of low-voltage substation area provides a strong support for the fault identification of low-voltage substation area.Therefore,this paper proposes a method of low-voltage substation fault identification based on graph convolution neural network whole graph classification.First,the station users are regarded as nodes,and the station topology adjacency matrix is automatically generated;Secondly,by using the intelligent device terminal,the fault feature data of the node during a fault in a substation area is collected and the data dimension is reduced,a paired mapping data with the adjacency matrix is then formed.Finally,the graph data of the whole station area is input into the graph convolution neural network as the whole graph,and the node representation of the whole graph is obtained by averaging the characteristic data of pooling graph,and the corresponding short-circuit type is finally output.Finally,the simulation results show that the accuracy of this method in the research and judgment of low-voltage substation short-circuit fault types reaches 99%.

关 键 词：低压台区 图卷积神经网络 短路故障识别 

分 类 号：TM631.2[电气工程—电力系统及自动化]