针对传统局部放电模式识别方法识别准确率较低,特别在小样本数据集情况下识别效果较差的问题,提出了一种基于一维卷积神经网络的小样本数据集识别方法,试验采集了4类油中典型缺陷下的放电样本各200份,分别采用此方法与传统聚类法、SVM法、BP神经网络法、CNN图像法对上述样本构成的数据集进行了识别分类。结果表明:该方法在小样本数据集情况下相较传统聚类法、SVM法、BP神经网络法和CNN图像法具有更好的识别效果,在训练样本数为50时,其识别准确率分别高于SVM法和BP神经网络法2.00%和6.25%;同时,相对传统识别方法,该方法表现出较强的特征提取与学习能力,其迭代过程收敛较快,单次训练耗时远低于SVM法,与BP神经网络法相近;最后,在针对卷积网络各结构参数对该方法识别准确率的影响研究过程中发现,卷积核尺寸对其识别准确率的影响最大。
Aiming at the problem of the traditional partial discharge pattern recognition method has low recognition accuracy in the case of small sample data sets
we propose a recognition method based on one-dimensional convolutional neural network(CNN). We collect 200 discharge samples under four typical defects
and identify and classify the data sets composed of the above samples by this method
as well as by traditional clustering method
support vector machine(SVM) method
BP neural network method
and CNN image method. The results show that this method has better recognition effect than traditional clustering method
SVM method
BP neural network method and CNN image method in the case of small sample data set. Specifically
when the number of training samples is 50
the recognition accuracy is higher than that of SVM method and BP neural network method by 2.00% and 6.25% respectively. Furthermore
compared with the traditional recognition method
this method shows strong feature extraction and learning ability
with a faster converging iterative process. The time required for a single training time is significantly lower than that of the SVM method and comparable to the BP neural network method. Finally
in investigating the impact of various structural parameters of the convolutional network on the recognition accuracy of the method
it is found that the size of the convolution kernel has the greatest influence on the recognition accuracy.
Bachlechner M.,Birkenfeld T.,Soldin P.,Stahl A.,Wiebusch C..Partition pooling for convolutional graph network applications in particle physics[J].Journal of Instrumentation,2022.
Stančić Adam,Vyroubal Vedran,Slijepčević Vedran.Classification Efficiency of Pre-Trained Deep CNN Models on Camera Trap Images[J].Journal of Imaging,2022.
Anan Zhang,Jiahui He,Yu Lin,Qian Li,Wei Yang,Guanglong Qu.Recognition of partial discharge of cable accessories based on convolutional neural network with small data set[J].COMPEL - The international journal for computation and mathematics in electrical and electronic engineering,2020.
Asifullah Khan,Anabia Sohail,Umme Zahoora,Aqsa Saeed Qureshi.A survey of the recent architectures of deep convolutional neural networks[J].Artificial Intelligence Review,2020.
Duan Lian,Hu Jun,Zhao Gen,Chen Kunjin,He Jinliang,Wang Shan X..Identification of Partial Discharge Defects Based on Deep Learning Method[J].IEEE Transactions on Power Delivery,2019.
Jürgen Schmidhuber.Deep learning in neural networks: An overview[J].Neural Networks,2015.
Feng Yu,Xiaozhong Xu.A short-term load forecasting model of natural gas based on optimized genetic algorithm and improved BP neural network[J].Applied Energy,2014.
翟永杰,王璐瑶,郭聪彬.基于Faster R-CNN的复杂背景下绝缘子目标检测[J].电子测量技术,2023(16).
李智皓,鲁殿君,吕荣水,刘伟.基于振动图像和卷积神经网络的滚动轴承故障诊断方法研究[J].工程机械,2023(05).
陈起磊,蒋亦悦,唐瑶,张晓飞,王朝红.基于时频图与改进图卷积神经网络的异步电机故障诊断方法[J].振动与冲击,2022(24).
戴永东,姚建光,李勇,毛锋,文志科,曹世鹏.基于定点化自适应选择卷积神经网络的电力缺陷识别方法[J].高电压技术,2021(11).
戴志辉,耿宏贤,韩健硕,李金铄,方伟.基于矩阵算法和BP神经网络的智能站二次系统故障定位方法[J].华北电力大学学报(自然科学版),2022(06).
周秀,朱洪波,马云龙,刘威峰,高博,田天,罗艳,李秀广,何宁辉.基于深度学习的变压器局部放电模式识别研究[J].高压电器,2019(12).
郑艳艳,朱永利,高佳程.基于支持向量回归的VPMCD方法及其在局部放电模式识别中的应用[J].华北电力大学学报(自然科学版),2019(02).
杨帆,王干军,彭小圣,文劲宇,陈清江,杨光垚,李朝晖.基于卷积神经网络的高压电缆局部放电模式识别[J].电力自动化设备,2018(05).
谢红玲,刘新元,李燕青.基于决策树支持向量机模型的变压器故障诊断法[J].华北电力大学学报(自然科学版),2010(06).
苑津莎,李中.基于形状相似距离的K-means聚类算法[J].华北电力大学学报(自然科学版),2009(06).
耿兰芹,王芳,赵文清.SVM回归与朴素贝叶斯分类相结合的变压器故障诊断[J].华北电力大学学报,2006(06).
律方成,刘云鹏,李燕青.电力变压器局部放电检测与诊断方法评述[J].华北电力大学学报,2003(06).
王国利,郝艳捧,李彦明.油中典型局部放电模型放电波形特征参数的提取[J].电网技术,2001(10).
0
浏览量
333
下载量
0
CSCD
关联资源
相关文章
相关作者
相关机构
京公网安备11010802024621