摘要
随着大数据平台的建立,数据中心积累了大量现场检测存储的图像等非结构化数据。传统的局部放电模式识别方法一般针对结构化数据,无法直接应用于非结构化数据。为解决该问题,提出一种基于一维卷积神经网络的局部放电时域波形图像的模式识别方法。利用图像处理技术对输入图像进行预处理,获取数据一维特性并进行线性归一化。基于深度学习,利用网络直接进行模式识别。通过变电站现场带电检测和实验室模拟实验,建立了 5种局放缺陷类型的时域波形图像数据集,并进行了对比实验。实验结果表明,使用一维卷积神经网络对局放缺陷进行模式识别的正确率为88.9%,显著优于支持向量机、反向传播神经网络模型,且在相同时间复杂度情况下优于二维卷积神经网络。该方法通过网络自主学习特征,无需人工提取,实现了对时域波形图像类非结构化数据的直接识别,实验复杂度低,具有更高识别率和更好鲁棒性。
With established big data platforms, a large number of unstructured on-site data such as images are accumulated in data centers. Traditional partial discharge pattern recognition method is generally aimed at structured data and can not be directly applied to unstructured data. To solve this problem, a time-domain waveform pattern recognition method based on one-dimensional convolutional neural network is proposed. The image processing technology is used to preprocess the input images and one-dimensional characteristics of the waveform are obtained. Then the linearized function is used to normalize the data. Based on deep learning, the network is used for pattern recognition directly. Through on-site detection and lab simulated experiments, image data sets for five partial discharge defects are established and comparative experiments are conducted. Experimental results show that the recognition rate of partial discharges using one-dimensional convolutional neural network is 88.9%, significantly higher than that of support vector machine and back propagation neural network model. It also performs better than two-dimensional convolutional neural network under the same time complexity. The method autonomously learns features through the network and does not need to manually extract features. In conclusion, it has advantages of lower experimental complexity, higher recognition rate and better robustness.
作者
万晓琪
宋辉
罗林根
李喆
盛戈皞
江秀臣
WAN Xiaoqi;SONG Hui;LUO Lingen;LI Zhe;SHENG Gehao;JIANG Xiuchen(Department of Electrical Engineering,Shanghai Jiao Tong University,Minhang District,Shanghai 200240,China)
出处
《电网技术》
EI
CSCD
北大核心
2019年第6期2219-2226,共8页
Power System Technology
基金
国家重点研发计划项目(2017YFB0902705)
国家电网公司科技项目~~
关键词
卷积神经网络
局部放电
图像
模式识别
convolutional neural network (CNN)
partial discharge
image
pattern recognition
作者简介
万晓琪(1995),女,硕士研究生,研究方向为输变电设备状态监测与智能化,E-mail:sin043@sjtu.edu.cn;通信作者:宋辉(1987),男,博士,助理研究员,研究方向为输变电设备状态监测与智能化,E-mail:songeos@163.com。
