基于深度学习的电容式电压互感器故障诊断  

Fault Diagnosis of Capacitor Voltage Transformer Based on Deep Learning

作　　者：黄桂平 吴杰 夏岩 熊兴中 张蕊 HUANG Guiping;WU Jie;XIA Yan;XIONG Xingzhong;ZHANG Rui(School of Automation and Information Engineering,Sichuan University of Science&Engineering,Yibin 644000,China;Intelligent Electric Power Grid Key Laboratory of Sichuan Province,Sichuan University,State Grid Sichuan Electric Power Company,Chengdu 610000,China;Electric Power Research Institute of Sichuan Electric Power Company,Chengdu 610000,China;Artificial Intelligence Key Laboratory of Sichuan Province,Yibin 644000,China)

机构地区：[1]四川轻化工大学自动化与信息工程学院,四川宜宾644000 [2]智能电网四川省重点实验室,四川大学、国网四川省电力公司,成都610000 [3]国网四川省电力公司电力科学研究院,成都610000 [4]人工智能四川省重点实验室,四川宜宾644000

出　　处：《四川轻化工大学学报（自然科学版）》2023年第5期76-84,共9页Journal of Sichuan University of Science & Engineering(Natural Science Edition)

基　　金：四川省科技计划项目(S-2023-000009);四川省人工智能重点实验室项目(2020RZY03);智能电网四川省重点实验室项目(2022-IEPGKLSP-KFYB05)。

摘　　要：电容式电压互感器(CVT)是应用于变电站的关键设备,能够实现长途通信、远方测量、选择性线路高频保护等功能,为提高变电站的可调控能力提供有力保障。本文基于CVT的结构形态特性与故障类型,提出一种改进的卷积神经网络(CNN)预测方法,将其应用于CVT故障在线检定。该方法在传统CNN模型中加入平均池化层,实现信号降采样并保留信号的特征信息,使用支持向量机(SVM)代替传统的softmax函数。对所提模型进行仿真实验,本文模型在187μs的检测时间内能够实现100%检测精度,检测精度与检测时间均优于传统CNN模型;同时,将某500 kV变电站CVT实测电压数据作为数据集,用于本文模型的仿真实验,仿真结果表明本文模型在实际工程案例中能迅速检出CVT早期故障并发出故障预警信息,故障诊断效果较好,对于变电站稳定运行具有重要意义。Capacitive voltage transformer(CVT)is a key equipment used in substations,which can realize long-distance communication,remote measurement,selective line high-frequency protection and other functions,providing a strong guarantee for improving the regulability of substations.Based on the structural characteristics and fault types of CVT,an improved convolutional neural network(CNN)prediction method has been proposed in this paper,which is applied to the online verification of CVT faults.This method is established by adding an average pooling layer to the traditional CNN model to achieve signal downsampling and retain the characteristic information of the signal.Support vector machine(SVM)is used to replace the traditional softmax function.The simulation experiment of the proposed model shows that the model proposed in this paper exhibits 100%detection accuracy within the detection time of 187μs,and the detection accuracy and detection time are better than the traditional CNN model.At the same time,the measured voltage data of a 500 kV substation CVT is used as the data set for the simulation experiment of the model in this paper.The simulation results show that the model in this paper can quickly detect CVT early faults and send out fault warning information in actual engineering cases,and the fault diagnosis effect is good,which is of great significance for the stable operation of the substation.

关 键 词：电容式电压互感器 全局平均池化 卷积神经网络 在线检定 

分 类 号：TM451[电气工程—电器] TP391.41[自动化与计算机技术—计算机应用技术]