摘要
当室内配电系统发生串联电弧故障时,电弧燃烧温度可高达数千摄氏度,从而导致电气火灾的发生。而低压配电网中负载类型复杂,利用一般的电流信号时频分析,很难对串联电弧故障进行有效识别。针对这一问题,文章利用深度学习强大的计算机视觉能力,提出了一种基于注意力机制和深度残差收缩网络(attention mechanism and deep residual shrinkage network,Attention-DRSN)的故障检测方法。首先,使用连续小波变换提取电流信号特征信息,并转化为图像特征。其次,对提取到的图像特征进行数据增强和灰度化处理,并利用主成分分析方法(principal component analysis,PCA)对特征图像进行了重构。最后,构建了Attention-DRSN电弧故障检测模型,并采用K-折交叉验证方法对数据集进行划分,验证了所提方法的有效性。实验结果表明,该检测方法对串联电弧故障具有较高的检测精度,平均检测准确率为98.52%,对未来电弧故障检测装置设计具有重要的借鉴意义。
When a series arc fault occurs in the indoor power distribution system,the arc burning temperature may reach up to as high as thousands of Celsius,which leads to the occurrence of electrical fires.The complex load types in the low-voltage distribution network makes it difficult to effectively identify the series arcing faults using a general time-frequency analysis of the current signals.To address this problem,this paper proposes a fault detection based on the attention mechanism and the deep residual shrinkage network(Attention-DRSN)by using the powerful computer vision capability of deep learning.First,the current signal feature information was extracted by using a continuous wavelet transform,and transformed into image features.Second,the extracted image features were enhanced and grayed out,and the feature images were reconstructed using the principal component analysis(PCA).Finally,the Attention-DRSN arc fault detection model was constructed,and the K-fold cross-validation was used to divide the data set,verifying the effectiveness of the proposed method.The experimental results show that the detection method has a high detection accuracy for series arc faults with an average detection accuracy of 98.52%,which is an important reference for the design of arc fault detection devices in the future.
作者
胡从强
曲娜
张帅
冮震
HU Congqiang;QU Na;ZHANG Shuai;GANG Zhen(School of Safety Engineering,Shenyang Aerospace University,Shenyang 110136,Liaoning Province,China)
出处
《电网技术》
EI
CSCD
北大核心
2023年第5期1897-1904,共8页
Power System Technology
基金
国家自然科学基金项目(61901283)。
关键词
注意力机制
深度残差收缩网络
连续小波变换
PCA特征提取重构
串联电弧故障
attention mechanism
deep residual shrinkage network
continuous wavelet transform
extraction and reconstruction of features with PCA
series arc fault
