基于卷积神经网络的变压器局部放电模式识别  

Partial Discharge Pattern Recognition in Power Transformers Based on Convolutional Neural Networks

作　　者：刘兵[1] 郑剑[2] 

机构地区：[1]湖南机电职业技术学院电气工程学院,长沙410151 [2]湖南大学电气与信息工程学院,长沙410082

出　　处：《高压电器》2017年第5期70-74,81,共6页High Voltage Apparatus

基　　金：湖南省科技厅科研项目(2011FJ4152);湖南省教育厅科研项目(11C0482;13C218)~~

摘　　要：对局部放电进行有效识别可以为评估变压器设备绝缘状况提供科学的参考依据,然而局部放电类型的识别往往需要人为地提取描述特征,适应性很差。针对此问题,提出一种基于卷积神经网络的智能识别新方法。根据视觉注意机制分割出放电信号图像,并将灰度化和双线性插值归一化处理的图像作为卷积神经网络的输入。该方法模拟人脑的机制来解释数据,可以直接对采集到的放电信号图像进行自动特征学习与模式识别。实验中对4种典型放电类型的识别率超过了94%,显著优于传统的方法。试验结果表明,该方法无需进行复杂的特征提取,有较高的准确率和很好的鲁棒性。The effective recognition for partial discharge insulation condition in power transformers. However, the will provide a scientific reference basis for evaluating the recognition of partial discharge type often needs to extract the description feature artificially, which has poor adaptability. Aiming at the current problem, an intelligent recognition method based on convolutional neural networks is proposed. According to visual attention mechanism, partial discharge signal images are slipped out. After the processes of graying, bilinear interpolation, and normal- izing, the images are taken as the input of the convolutional neural network. This method simulates the human brains to explain data, which can automatically learn features and identify patterns for discharge signals are col- lected directly. The experiments show that the recognition rate of four typical kinds of discharge type is above 94%, which is superior to that of traditional methods significantly. Results demonstrate that the proposed method requires no complex feature extraction and has high accuracy and good robustness.

关 键 词：电力变压器 局部放电 卷积神经网络 深度学习 模式识别 

分 类 号：TM855[电气工程—高电压与绝缘技术]