摘要
为了提升电力变压器的故障诊断的准确性,提出一种基于领域知识图谱(domain knowledge graph, DKG)和改进图卷积网络(improved graph convolutional network, IGCN)的电力变压器故障诊断方法。首先,利用DKG构建由节点与边构成的电力变压器知识图谱,并建立知识图谱与故障样本之间的映射关系,形成变压器故障知识图谱,为后续故障诊断提供数据支撑;其次,在谱域GCN原理的基础上,引入基于稀疏注意力的自适应图池化,对谱域GCN网络进行改进;最后,使用不同诊断方法对采集的3 000个变压器故障样本进行故障诊断实验,验证所提方法的优越性。实验结果表明,所提方法能够有效诊断变压器故障类型,较卷积神经网络、深度置信网络和深度神经网络等常规故障诊断方法:在2 400个样本下,识别率分别提升0.7%、4.7%和9.2%;随着样本数量的减小识别优势更为明显,在400个样本数量下,识别率分别提升4.2%、7.4%和14.3%。由此可知所提方法对变压器故障样本信息利用更为全面,诊断效果更优,且识别优势随着样本数量的减小更为明显。
In order to improve the accuracy of fault diagnosis for power transformers,this paper proposes a fault diagnosis method for power transformers based on domain specific knowledge graph(DKG)and improved graph convolutional neural network(IGCN).Firstly,it uses DKG to construct a knowledge graph of power transformers composed of nodes and edges,and establishes a mapping relationship between the knowledge graph and fault samples to form a transformer fault knowledge graph,providing data support for subsequent fault diagnosis.Secondly,on the basis of the principle of spectral domain GCN,it introduces the adaptive graph pooling based on sparse attention to improve the spectral domain GCN network.Finally,by using different diagnostic methods,it carries out fault diagnosis experiments on 3000 transformer fault samples collected to verify the superiority of the proposed method.The experimental results show that the proposed method can effectively diagnose transformer fault types.Compared with traditional fault diagnosis methods such as CNN,DBN,and DNN,the recognition rate increases by 0.7%,4.7%,and 9.2%for 2400 sample sizes respectively.As the number of samples decreases,the recognition advantage becomes more apparent.At 400 sample sizes,the recognition rates increase by 4.2%,7.4%,and 14.3%respectively.It can be seen that the proposed method can utilize more comprehensive information on transformer fault samples,has better diagnostic performance,and the recognition advantage becomes more obvious as the number of samples decreases.
作者
黄欣
郇嘉嘉
赵敏彤
吴伟杰
张伊宁
HUANG Xin;HUAN Jiajia;ZHAO Mintong;WU Weijie;ZHANG Yining(Grid Planning&Research Center,Guangdong Power Grid Co.,Ltd.,Guangzhou,Guangdong 510080,China)
出处
《广东电力》
2023年第11期146-156,共11页
Guangdong Electric Power
基金
中国南方电网有限责任公司科技项目(GDKJXM20190387)。
关键词
电力变压器
领域知识图谱
改进图卷积网络
稀疏注意力
自适应池化
power transformer
domain knowledge graph(DKG)
improved graph convolutional network(IGCN)
sparse attention
adaptive pooling
