递归量化分析的变压器健康评估与故障诊断  

Recurrence quantification analysis of transformer health assessment and fault diagnosis

作　　者：严侣 谢丽蓉[1] 吐松江·卡日[1] 孟杰[2] 张馨月 YAN Lü;XIE Lirong;TUSONGJIANG·Kari;MENG Jie;ZHANG Xinyue(School of Electrical Engineering,Xinjiang University,Urumqi 830047,China;TBEA Xinjiang Transformer Co.,Ltd.,Changji 831100,China)

机构地区：[1]新疆大学电气工程学院,新疆乌鲁木齐830047 [2]特变电工股份有限公司新疆变压器厂,新疆昌吉831100

出　　处：《现代电子技术》2023年第22期102-108,共7页Modern Electronics Technique

基　　金：国家自然科学基金项目(52067021);新疆维吾尔自治区自然科学基金项目(2022D01C35)。

摘　　要：针对传统时域、频域分析方法忽略信号的部分频率信息的问题,提出一种基于递归定量分析(RQA)和鸡群算法(CSO)优化支持向量机(SVM)的变压器健康分析与机械故障诊断方法。首先利用相空间重构技术构建变压器振动信号的递归图;然后进行递归定量分析,提取故障特征并结合马氏距离(MD),以置信度-故障程度关系式进行变压器健康评估;最后利用CSO优化SVM参数,进行故障识别。实验结果表明,提取的递归定量特征参数可以有效地评估变压器的健康程度,综合识别率为94.07%,较传统时域、频域特征高约15.4%、8.96%;相比于传统模型,采用CSO-SVM故障诊断模型可将诊断精度提高至99.26%。In allusion to the problem that traditional time domain and frequency domain analysis methods ignore part of the frequency information of the signals,a transformer health analysis and mechanical fault diagnosis method based on recurrence quantification analysis(RQA)and chicken swarm optimizer-support vector machine(CSO-SVM)is proposed.The phase space reconstruction technology is used to construct a recursive graph for the transformer vibration signal.Then,RQA is carried out,and fault characteristics are extracted and combined with Mahalanobis distance(MD)to perform transformer health analysis in a confidence-degree of failure relationship.CSO is used to optimize SVM parameters for the fault identification.The experimental results show that the extracted recursive quantitative feature parameters can effectively evaluate the health of transformers,and its comprehensive recognition rate is 94.07%,which is about 15.4%and 8.96%higher than that of traditional time domain and frequency domain features.In combination with the traditional model,the CSO-SVM fault diagnosis model can improve the diagnostic accuracy to 99.26%.

关 键 词：变压器故障诊断 健康评估 递归定量分析 故障特征提取 鸡群优化算法 支持向量机 

分 类 号：TN919-34[电子电信—通信与信息系统] TN624-34[电子电信—信息与通信工程]