摘要
电缆终端缺陷类型一般与局部放电信号特征密切相关,因此可以通过对局部放电信号进行模式识别来实现缺陷分类。对15 kV XLPE电缆终端4种典型缺陷的放电脉冲波形和时频谱图特征进行分析处理,得到可用于识别的数据样本,然后分别采用Vision Transformer模型、LeNet5、AlexNet和支持向量机对数据进行训练,对比不同算法的识别准确率。结果显示在数据充足的情况下,Vision Transformer模型的识别精度高于其他识别算法。所提方法及结论可为电缆附件的绝缘评估提供可靠依据,具有一定的指导意义。
Cable terminal defect types are generally closely related to the characteristics of partial discharge signals,thus it is feasible to classify the defect types by means of pattern recognition of the partial discharge signals.This paper analyzes the discharge pulse waveforms and time spectrogram characteristics of four typical defects of 15 kV XLPE cable terminals and obtains data samples that can be used for identification.Then it uses the Vision Transformer(VIT)model,LeNet5,AlexNet and support vector machine to train the data to compare the recognition accuracy of different algorithms.The results show that the recognition accuracy of the VIT model is higher than that of other recognition algorithms as the data is sufficient.The methods and conclusions proposed can provide a reliable basis for the insulation evaluation of cable accessories and have certain guiding significance.
作者
唐庆华
方静
李旭
宋鹏先
孟庆霖
魏占朋
TANG Qinghua;FANG Jing;LI Xu;SONG Pengxian;MENG Qinglin;WEI Zhanpeng(State Grid Tianjin Electric Power Research Institute,Tianjin 300384,China;State Grid Tianjin Electric Power Company,Tianjin 300010,China;State Grid Tianjin Electric Power Company Cable Branch,Tianjin 300170,China)
出处
《广东电力》
2023年第11期138-145,共8页
Guangdong Electric Power
基金
国家自然科学基金面上项目(52277156)
国网天津市电力公司科技项目(电科-研发2023-37)。
