摘要
电力设备发生局部放电为小概率事件,现场检测出的局部放电脉冲相位分析图谱样本具有数量稀少、样本类别比例不平衡的问题,导致难以训练出能够准确判断局部放电类型的分类器。为了扩充并平衡不同种类局部放电样本的数量,提高局部放电模式识别准确率,提出一种基于BAGAN-CNN的局部放电模式识别方法。首先,构建BAGAN(Balancing GAN)网络模型,解决原始ACGAN损失函数矛盾的问题,使BAGAN生成器能够生成样本种类稀少的局部放电数据。然后,将原始样本和增强样本作为分类器输入,构造卷积神经网络,自动提取局部放电特征,并通过Softmax层进行分类。实验表明,通过BAGAN生成的数据相比于其他数据扩充方法,能够生成更高质量局部放电样本;相较于传统分类器,CNN分类器的识别准确率更高。
Partial discharge in power equipment is a rare event.The Phase Resolve Partial Discharge(PRPD)samples obtained from on-site measurements suffer from scarcity and imbalanced class proportions,making it difficult to train a classifier that can accurately identify the different types of partial discharge.To address the challenges associated with generating a large number of corresponding samples of rare partial discharge events in electrical equipment and accurately classifying them,a novel partial discharge pattern recognition method based on BAGAN-CNN is proposed.The first step involves constructing a BAGAN(Balancing GAN)network model that resolves the conflicting loss function of the original ACGAN,thus enabling the BAGAN generator to generate highquality samples of rare types of partial discharge data.In the second step,a convolutional neural network(CNN)is constructed with both original and augmented samples serving as inputs.The CNN automatically extracts partial discharge features using its nonlinear encoder and classifies them through a softmax layer.The experimental results demonstrate the superiority of the proposed method,as the data generated by BAGAN produces higher quality partial discharge samples compared to other data augmentation methods.Additionally,the recognition accuracy of the CNN classifier outperforms that of traditional classifiers.
作者
周云海
靳广伟
于高缘
黄伟
迟婉求
黄南天
ZHOU Yunhai;JIN Guangwei;YU Gaoyuan;HUANG Wei;CHI Wanqiu;HUANG Nantian(Anhui jidian New Energy Co.,Ltd.,Hefei 230000 China;School of Electrical Engineering,Northeast Electric Power University,Jilin 132012,China;Shandong Jidian New Energy Co.,Ltd.,Weifang 261000,China)
出处
《电气应用》
2023年第7期25-33,共9页
Electrotechnical Application
基金
国家重点研发计划(2022YFB2404002)
淮南市潘阳光伏发电有限公司科技项目(HNPY-HT/FW-2022-005)。
关键词
局部放电
生成对抗网络
模式识别
局部放电脉冲相位分析图谱
卷积神经网络
partial discharge
generative adversarial networks
pattern recognition
phase resolved partial discharge spectrum
convolutional neural networks
