基于深度残差网络的GIS局部放电PRPD谱图模式识别

Pattern Recognition of Partial Discharge PRPD Spectrum in GIS Based on Deep Residual Network

为了研究气体绝缘组合电器(gasinsulatedswitchgear,GIS)局部放电脉冲相位分布(phaseresolvedpartial discharge,PRPD)谱图的模式识别,解决传统的统计参数分析方法识别准确率低的问题,该文提出了一种基于深度残差网络的GIS局部放电PRPD谱图模式识别方法。首先,设计并搭建了GIS中4类典型局部放电缺陷的实验模型并采集实验数据;然后,利用条件生成对抗网络对PRPD谱图训练集进行数据扩充;最后,利用深度残差网络提取每类缺陷的PRPD谱图特征并将其分类。实验结果表明,该方法相较于普通卷积神经网络和统计参数分析方法,其识别准确率有明显提升,最高可达98.75%。研究结果表明所提方法能有效区分出GIS中4类典型的局部放电缺陷类型,在工程实际中有良好的应用前景。

In order to research the pattern recognition of phase resolved partial discharge(PRPD)spectrum of gas insulated switchgear(GIS)and solve the problem of low accuracy of traditional statistical parameter analysis methods,a method for pattern recognition of partial discharge PRPD spectrum in GIS based on deep residual network is proposed in this paper.Firstly,the experimental models of four typical types of partial discharge defects in GIS are set up and the experimental data are collected.Then,a conditional generative adversarial network is used to expand the data of PRPD spectrum training set.Finally,the characteristics of PRPD spectrum of each defect are extracted and classified by deep residual network.Experimental results show that the recognition accuracy of the proposed method is significantly improved compared with that of the convolutional neural network or the traditional statistical parameter analysis method.The recognition accuracy is up to 98.75%.The results of research show that the proposed method can be adopted to effectively distinguish four typical types of partial discharge defects in GIS and has a good application prospect in engineering practice.