电力变压器内部故障的递进分层诊断方法  

Progressive Hierarchical Diagnosis of Internal Faults in Power Transformers

作　　者：咸日常[1] 李云淏 刘焕国 王昭璇 张海强 胡玉耀 王玮[1] XIAN Richang;LI Yunhao;LIU Huanguo;WANG Zhaoxuan;ZHANG Haiqiang;HU Yuyao;WANG Wei(College of Electrical and Electronic Engineering,Shandong University of Technology,Zibo 255049,Shandong Province,China;Shandong Huineng Electric Company,Zibo 255000,Shandong Province,China;Zibo Power Supply Company,State Grid Shandong Electric Power Company,Zibo 255000,Shandong Province,China)

机构地区：[1]山东理工大学电气与电子工程学院,山东省淄博市255049 [2]山东汇能电气有限公司,山东省淄博市255000 [3]国网山东省电力公司淄博供电公司,山东省淄博市255000

出　　处：《电网技术》2025年第4期1726-1734,I0079,I0080,共11页Power System Technology

基　　金：国家自然科学基金项目(52077221)。

摘　　要：电力变压器内部故障成因复杂、种类繁多,精确诊断难度大,现有诊断技术大多滞留于故障定性阶段。为实现多类型故障的精准定位,该文通过建立多状态量与故障特征之间的递进映射关系,提出一种改进灰狼算法与最小二乘支持向量机耦合的电力变压器故障递进分层诊断方法。首先介绍改进灰狼算法与最小二乘支持向量机的原理,建立电力变压器故障递进分层、自动诊断及定位模型;其次基于300组电力变压器的状态量,利用核主成分分析法进行降维处理,选取线性无关的特征状态量,依据DL/T 1685—2017《油浸式变压器状态评价导则》进行离散化处理,借助算法模型递进分层、自动诊断:第一层诊断故障回路、第二层确定故障部位、第三层明确故障原因,得到各分类器的诊断准确率及惩罚系数和核函数参数的最优组合解,并与其他算法模型的故障诊断结果进行分析对比;最后以实际故障案例验证方法的有效性。结果表明:该文所提诊断模型比其他方法拥有更高准确率和更快的运算速度。The causes of power transformer internal faults are complex and varied,making accurate diagnosis difficult,and most of the existing diagnostic techniques remain at fault characterization.To realize the accurate localization of multiple types of faults,this paper proposes a progressive hierarchical diagnosis method for power transformer faults with improved gray wolf algorithm(IGWO)coupled with Least Squares Support Vector Machines(LSSVM),based on the advantage of high binary classification accuracy of LSSVM,and the establishment of a recursive mapping relationship between multistate quantities and fault characteristics by increasing the number of classification layers of the model and reducing the number of classifications in each layer.Firstly,the principles of IGWO and LSSVM are introduced to establish the progressive hierarchical,automatic diagnosis and localization model for power transformer faults.Secondly,based on the state quantities of 300 groups of power transformers,the kernel principal component analysis is used for dimensionality reduction,and the linearly independent eigenstate quantities are selected,discretization according to DL/T 1685-2017 Guidelines for Condition Evaluation of Oil-immersed Transformers.Progressive stratification and automatic diagnosis with the help of algorithmic models:the first layer diagnoses the faulty circuit,the second layer determines the faulty part,and the third layer clarifies the cause of the fault.Obtain the diagnostic accuracy of each classifier and the optimal combination of penalty coefficients and kernel function parameters,and analyze and compare the fault diagnosis results with other algorithmic models.Finally,the validity of the methodology is verified with actual failure cases.The results show that the diagnostic model proposed in this paper possesses higher accuracy and faster computing speed than other methods.

关 键 词：电力变压器 改进灰狼算法 最小二乘支持向量机 多状态量 内部故障 递进分层诊断 

分 类 号：TM721[电气工程—电力系统及自动化]