基于生成对抗网络的二次屏柜交直流侵入信号检测方法  

AC-DC Intrusion Signal Detection Method Based on Generative Adversarial Network for Secondary Panel Cabinet

作　　者：尹清青 何涛 吴馨 金照盈 高筱婷 YIN Qingqing;HE Tao;WU Xin;JIN Zhaoying;GAO Xiaoting(Ma’anshan Electric Power Supply Company,State Grid Anhui Electric Power Co.,Ltd.,Ma’anshan Anhui 243000;College of Light Industry,Liaoning University,Shenyang Liaoning 110036;Key Laboratory of Symbolic Symbolic Computation and Knowledge Engineering of Ministry of Education,Changchun Jilin 130000)

机构地区：[1]国网安徽省电力有限公司马鞍山供电公司,安徽马鞍山243000 [2]辽宁大学轻型产业学院,辽宁沈阳110036 [3]吉林大学符号计算与知识工程教育部重点实验室,吉林长春130000

出　　处：《东北电力大学学报》2024年第4期21-27,共7页Journal of Northeast Electric Power University

基　　金：中央高校基本科研业务费-符号计算与知识工程教育部重点实验室(吉林大学)开放课题(93K172023K13);国网安徽马鞍山供电公司科技项目(SGTYHT/23-JS-003)。

摘　　要：变电站继电保护二次屏柜接线端子存在因人为操作失误或检测设备故障导致的交直流侵入情况,针对该情况特有的数据样本数量少、质量差的问题,文中提出基于生成对抗网络和主成分分析的侵入信号检测方法。采用高斯核平滑对接线端子电压、电流、频率等数据进行预处理,滤除干扰,再将清洗后的数据经过生成对抗网络进行扩增,以对交直流侵入信号进行主成分分析和故障检测识别。所提出方法识别准确率达到95%,实现了对二次屏柜接线端子小样本故障数据的精确检测。Based on the unpredictable AC-DC intrusion caused by human error or detection equipment failure in the terminal of secondary screen cabinet of substation relay protection,a new intrusion signal detection method based on generative adversarial network and principal component analysis is proposed to solve the problem of poor quality and small quantity of data samples.Gaussian kernel smoothing is used to preprocess the terminal data to reduce noise interference,and then the cleaned data is amplified through the generated adversarial network(GAN)to meet the subsequent principal component analysis processing and fault detection and identification of AC and DC intrusion signals.The identification accuracy of the proposed method reaches 95%,which realizes the accurate detection of the small sample fault data of the terminal of the secondary screen cabinet.

关 键 词：生成对抗网络 主成分分析 二次屏柜端子排 交直流侵入检测 高斯核平滑 

分 类 号：TM933.14[电气工程—电力电子与电力传动]