摘要
由高压电缆不同类型缺陷诱发的局部放电(PD)的识别难度较大,尤其是某些相似度较高的电缆绝缘缺陷类型难以区分。提出了一种基于卷积神经网络(CNN)的高压电缆PD模式识别方法,研究了不同网络层数、不同激活函数以及不同池化方式对识别效果的影响,并与传统的支持向量机(SVM)和反向传播神经网络(BPNN)算法进行了对比。结果表明,相比SVM和BPNN,CNN的总体识别精度分别提高了3.71%和4.06%,且能较好地识别具有高相似度的电缆缺陷类型。
The recognition of PD (Partial Discharge) caused by different types of defects in high-vokage cables is difficult,especially the recognition of PD caused by cable insulation defects with high similarity. The CNN( Convolu- tional Neural Network)-based PD pattern recognition method for high-voltage cables is presented. The influences of different network layers,activation functions and pooling methods of CNN on recognition effect are studied. The pro- posed method is also compared with SVM (Support Vector Machine) and BPNN (Back Propagation Neural Network) method,and the results show that the overall recognition accuracy of C NN is respectively 3.71% and 4.06% higher than that of SVM and BPNN. Furthermore, it is proved that the cable insulation defects with high similarity can be effectively recognized by CNN.
作者
杨帆
王干军
彭小圣
文劲宇
陈清江
杨光垚
李朝晖
YANG Fan;WANG Ganjun;PENG Xiaosheng;WEN Jinyu;CHEN Qingjiang;YANG Guangyao;LI Zhaohui(State Key Laboratory of Advanced Electromagnetic Engineering and Technology,School of Electrical and Electronic Engineering, Huazhong University of Science and Technology, Wuhan 430074, China;Zhongshan Power Supply Bureau of Guangdong Power Grid Corporation ,Zhongshan 528400, China)
出处
《电力自动化设备》
EI
CSCD
北大核心
2018年第5期123-128,共6页
Electric Power Automation Equipment
基金
国家自然科学基金资助项目(51541705)
湖北省自然科学基金资助项目(2016CFB536)
中国南方电网公司科技项目资助(GDKJXM20172769)~~
关键词
高压电缆
局部放电
卷积神经网络
模式识别
深度学习
high-voltage cables
partial discharge
convolutional neural network
pattern recognition
deep learning
作者简介
杨帆(1995-),男,陕西西安人,硕士研究生,主要研究方向为电力设备状态监测与故障诊断、新能源功率预测等;;彭小圣(1983-),男,湖北随州人,博士,IEEE会员,通信作者,主要研究方向为电力系统大数据理论与应用、电力系统主设备状态监测与故障诊断、局部放电信号提取与模式识别、新能源功率预测等(E-mail:xiaoshengpeng@hust.edu.cn)。