近日，昆明理工大学（Kunming University of Science and Technology）刘可真团队以《基于样本集成学习和SO-SVM的变压器故障诊断》为题在《电机与控制应用》上发表最新研究论文，第一作者为刘可真教授，通信作者为姚岳。
本文研究了变压器故障诊断方法，将蛇优化算法与支持向量机方法相结合，提出了一种基于样本集成学习和蛇优化算法（SO）优化支持向量机（SVM）的变压器故障诊断模型。
图1 故障诊断技术路线
该模型中，针对SVM模型中关键参数惩罚因子c和核函数系数g设置不合理而导致分类性能降低的问题，本文通过采用蛇优化算法（SO）对其进行优化，以优化参数后的SVM模型作为BalancedBaggingClassifier集成学习策略的基分类器，建立变压器故障诊断模型。同时，得益于蛇优化算法的引入，大大提高了对变压器运行状态分析的准确性，适用于实际工程的应用。
本文的主体内容主要分为一下三个部分：首先介绍了蛇优化算法及支持向量机的基本原理，其次提出了一种基于样本集成学习和蛇优化算法（SO）优化支持向量机（SVM）的变压器故障诊断模型，该模型先利用EasyEnsemble采样器对样本进行多次欠采样后生成类别平衡的多个子集；然后以Bagging策略训练SO优化关键参数后的SVM模型，综合各个分类器结果得到最终故障类型；最后通过仿真算例对本文所提模型有效性进行验证。
算例分析表明，本文所提模型相对已有研究故障识别能力更强，参数设置更合理，故障诊断能力更优，证明了提出方案的有效性。
Transformer Fault Diagnosis Based on Sample Integration Learning and SO-SVM
In this paper, the method of transformer fault diagnosis is studied. By combining snake optimization algorithm with support vector machine method, a transformer fault diagnosis model based on sample ensemble learning and snake optimization algorithm (SO) to optimize support vector machine (SVM) is proposed.
图2 Technical Route for Fault Diagnosis
In this model, in order to solve the problem of unreasonable setting of the key parameter penalty factor c and kernel function coefficient g in the SVM model, which leads to reduced classification performance, this paper optimizes it by using the snake optimization algorithm (SO) to optimize the parameters of the SVM model. As the base classifier of the BalancedBaggingClassifier integrated learning strategy, a transformer fault diagnosis model is established. At the same time, thanks to the introduction of the snake optimization algorithm, the accuracy of the analysis of the transformer operating status has been greatly improved, making it suitable for practical engineering applications.
The calculation example analysis shows that the model proposed in this paper has stronger fault identification ability, more reasonable parameter settings and better fault diagnosis ability than the existing research, which proves the effectiveness of the proposed solution.
支持基金：
云南省教育厅科学研究基金资助项目( 2022J1279) ；
云南电网有限责任公司科技项目( YNKJXM20180736)。
Funded by the Scientific Research Fund of Yunnan Provincial Department of Education (2022J1279); Yunnan Power Grid Co., Ltd., Science and Technology Project (YNKJXM20180736)
论文链接：
http://www.motor-abc.cn/djykzyy/article/abstract/20231203
推荐引用格式：
刘可真, 姚岳, 赵现平, 杨春昊, 盛戈皞, 王科. 基于样本集成学习和SO-SVM的变压器故障诊断[J]. 电机与控制应用, 2023, 50(12): 21-31.
LIU Kezhen, YAO Yue, ZHAO Xianping, YANG Chunhao, SHENG Gehao, WANG Ke. Transformer Fault Diagnosis Based on Sample Integration Learning and SO-SVM[J]. Electric Machines & Control Application, 2023, 50(12): 21-31.
刘可真，女，白族，1974.7月生，博士，教授，昆明理工大学教务处副处长，研究生导师。长期从事综合能源电力系统规划分析、电力设备状态监测和大数据分析方向的研究。主持和参与国家自然科学省部联合重点项目1项，面上项目1项，省部级科技项目和横向科技项目20余项，累计发表相关SCI/EI学术论文40余篇，授权发明专利20余项。获得省部级奖励3项。现任中国电工技术学会理事、云南省电力行业协会理事、云南省高等学校电气类专业教学指导委员会秘书长。
Liu Kezhen, female, Bai nationality, born in July 1974, Ph.D., professor, deputy director of the Academic Affairs Office of Kunming University of Science and Technology, and graduate tutor. She has long been engaged in research on comprehensive energy and power system planning and analysis, power equipment status monitoring and big data analysis. Hosted and participated in 1 joint key project of the Ministry of National Natural Sciences, 1 general project, more than 20 provincial and ministerial science and technology projects and horizontal science and technology projects, published more than 40 relevant SCI/EI academic papers, and authorized more than 20 invention patents item. Won 3 provincial and ministerial level awards. She is currently a director of the China Electrotechnical Society, a director of the Yunnan Electric Power Industry Association, and the secretary-general of the Electrical Major Teaching Steering Committee of Yunnan Higher Education Institutions.
