电力物联网下基于卷积神经网络和迁移学习的GIS局部放电模式识别分类方法研究

Research on Partial Discharge Pattern Recognition and Classification in GIS Based on Convolutional Neural Network and Transfer Learning in Power Internet of Things

随着电力物联网建设的快速推进,对实时监测气体绝缘金属封闭开关设备(GIS)内部局部放电信号的特高频法提出了新的和更高的要求。在充分利用表征GIS局部放电信息构建模型以提升模式识别准确率的同时,将模型移植到嵌入式系统,进而构成以边缘计算为支撑的物联网智能终端,成为亟待解决的一个关键问题。为此,文中深入研究了基于卷积神经网络和迁移学习的GIS局部放电模式识别分类方法,在多种典型缺陷下对比了不同模型在训练时间、准确率、参数量和存储花销等方面的性能。结果表明,Mobilenet模型具有最小的参数量和存储花销以及较短的训练时间,在电力物联网下基于智能终端的GIS局部放电模式识别中具有明显优势。

With the rapid development of the power internet of things,new and higher requirements have been put forward for the ultra high frequency method for real-time monitoring of partial discharge signals in gas-insulated metalclosed switchgear(GIS).Make full use of GIS partial discharge information to build a model to improve the accuracy of pattern recognition,at the same time,it has become a key problem to be solved when the model is transplanted to an embedded system to form an intelligent terminal of the Internet of things supported by edge computing.For this reason,deep analysis and discussion have been made in this paper for GIS partial discharge pattern recognition and classification method based on convolution neural network and transfer learning,and the performance of different models have been compared in terms of training time,accuracy,parameter number and storage cost under various typical defects.The results show that the Mobilenet model has the minimum number of parameters,the lowest storage cost and the shorter training time,thus it has obvious advantages in GIS partial discharge pattern recognition under the power internet of things.