基于改进卷积神经网络的电气绝缘子放电无损检测  

Electrical Insulator Discharge Detection Based on Improved Convolutional Neural Network

作　　者：谭兴华 陈瑞斌 丁卫东[3] 张国锋 TAN Xing-hua;CHEN Rui-bin;DING Wei-dong;ZHANG Guo-feng(Henan Sida Testing Technology Co.,Ltd.,Xuchang 461500,China;Henan Sida Electric Power Equipment Co.,Ltd.,Xuchang 461503,China;State Key Laboratory of Electric Power Equipment and Electrical Insulation,Xi’an Jiaotong University,Xi’an 710049,China)

机构地区：[1]河南四达检测技术有限公司,许昌461500 [2]河南四达电力设备股份有限公司,许昌461503 [3]西安交通大学电力设备与电气绝缘国家重点实验室,西安710049

出　　处：《自动化与仪表》2023年第4期88-91,97,共5页Automation & Instrumentation

基　　金：河南四达检测技术有限公司科研项目(KYXM-202116JD01)。

摘　　要：为了提高电力设备的放电检测,该文根据绝缘子闪络试验得到放电特征数据库,设计了一种基于改进卷积神经网络(CNN)的电气绝缘子放电无损检测方法。改进CNN框架经多次非线性转换后增强卷积核提取特征的效果,降低了参数量,识别准确率获得明显提升。研究结果表明,正确率随训练次数的增加而升高,当训练率太大或太小时都会增加获得最高正确率的难度。在训练率为0.001时,可以使模型发生快速收敛,此时准确率达到98.3%。当训练率为0.001时,改进CNN比AlexNet算法达到最高准确率,获得了比BP算法更强收敛性能。该研究对准确定位电气绝缘子放电性能具有很好的实际指导意义。In order to improve the discharge detection of power equipment,a non-destructive detection method of electrical insulator discharge based on improved convolutional neural network(CNN)was designed.The improved CNN algorithm is used to judge the insulator deterioration state and compared with other algorithms.Improved CNN framework after multiple nonlinear transformation to enhance feature extraction effect of convolution kernel,reduce the number of parameters,recognition accuracy is significantly improved.The results show that the correct rate increases with the increase of training times,and when the training rate is too large or too small,it will increase the difficulty of obtaining the highest correct rate.When the training rate is 0.001,the model can be rapidly convergent,and the accuracy can reach 98.3%.When the training rate is 0.001,the improved CNN achieves the highest accuracy than AlexNet algorithm.Both the improved CNN algorithm and AlexNet algorithm achieve stronger convergence performance than BP algorithm.This research has a very good practical guiding significance for accurately locating the discharge performance of electrical insulators.

关 键 词：无损检测 卷积神经网络 放电检测 电气绝缘子 准确率 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]