摘要
阐述配电网单相接地故障特征提取难点,分析现有选线方法、选线精度不高的问题,提出一种连续小波变换(Continuous wavelet transform,CWT)和CNN-LSTM的故障选线方法。首先对零序暂态电流进行连续小波变换获取对应的时频灰度图像,然后CNN自适应提取时频灰度图像的局部特征,LSTM层从CNN层学到的局部特征中学习上下文依赖关系,最后通过SoftMax层实现故障选线。仿真结果表明,所提方法的选线精度为99.65%,与CWT-CNN等方法相比,具有较强的鲁棒性。
This paper describes the problem of the difficulty in single-phase grounding fault feature extraction in distribution network and the low accuracy of existing line selection methods,proposes a fault line selection method based on continuous wavelet transform and CNN-LSTM.Firstly,the corresponding time-frequency gray image is obtained by continuous wavelet transform of zero-sequence transient current.Then,the local features of the time-frequency grayscale image are adaptively extracted by CNN,and the LSTM layer learns the context dependency from the local features learned by the CNN layer.Finally,the fault line selection is realized through the SoftMax layer.The simulation results show that the proposed method's line selection accuracy is 99.65%,which shows strong robustness compared to other methods,such as CWT-CNN.
作者
何银
何宇
聂祥论
HE Yin;HE Yu;NIE Xianglun(Power China Guizhou Electric Power Engineering Co.,Ltd.,Guizhou 550081,China;College of Electrical Engineering,Guizhou University,Guizhou 550025,China;Bijie Power Supply Bureau of Guizhou Power Grid Co.,Ltd.,Guizhou 551700,China)
出处
《集成电路应用》
2024年第1期418-421,共4页
Application of IC
基金
黔科合支撑项目([2022]一般014)。
关键词
故障选线
连续小波变换
卷积神经网络
长短期记忆神经网络
特征提取
fault line selection
continuous wavelet transform
convolutional neural network
long short-term memory neural network
feature extraction
