[1]彭 刚,唐松平,张作刚,等.基于改进多分类概率SVM模型的变压器故障诊断[J].机械与电子,2018,(04):42-47.
　PENG Gang,TANG Songping,ZHANG Zuogang,et al.Fault Diagnosis for Power Transformer Based on Improved Multi-Classification Probabilistic Support Vector Machine[J].Machinery & Electronics,2018,(04):42-47.
点击复制

基于改进多分类概率SVM模型的变压器故障诊断
分享到：

《机械与电子》[ISSN:1001-2257/CN:52-1052/TH]

卷:
期数:
2018年04期
页码:
42-47
栏目:
自动控制与检测
出版日期:
2018-04-24

文章信息/Info

Title:
Fault Diagnosis for Power Transformer Based on Improved Multi-Classification Probabilistic Support Vector Machine
文章编号:
1001-2257(2018)04-0042-06
作者:
彭 刚1; 唐松平1; 张作刚1; 彭 杰1; 张彦斌2
(1.广东电网有限责任公司惠州供电局,广东 惠州 516000; 2.西安交通大学电气学院,陕西 西安 710049)
Author(s):
PENG Gang1; TANG Songping1;  ZHANG Zuogang1;  PENG Jie1; ZHANG Yanbin2
(Huizhou Power Supply Bureau of Guangdong Power Grid Co., Ltd., Huizhou 516000, China; 2. School of Electrical Engineering, Xi'an Jiaotong University, Xi'an 10049, China)
关键词:
改进多分类;  电力变压器;  故障诊断;  概率支持向量机
Keywords:
improved multi classification;  power transformer;  fault diagnosis;  probabilistic support vector machine(PSVM)
分类号:
TM411
文献标志码:
A
摘要:
针对电力变压器内部结构复杂,故障类型繁多,难以实现故障准确有效诊断的问题。提出k近邻及改进多分类概率支持向量机对电力变压器进行多分类故障诊断的方法。首先,采用有向无环图的形式对变压器各种故障进行归类,进而利用k近邻算法对故障大类进行预分类,缩小故障所属类别,降低了后续多分类模型的构建复杂度。然后,以预分类后的类别样本数据作为输入,训练OVO-SVMs分类器,以概率的形式输出隶属各类的概率矩阵,并用改进OVR-SVMs分类器的概率输出作为概率矩阵中各元素的权重系数,对概率矩阵进行更新、修正,提高故障诊断正确率及可靠性。实际诊断结果显示,所提出的方法与IEC三比值法和传统的多分类支持向量机相比,在故障诊断范围、故障诊断正确率和故障诊断效率上均有所提高。
Abstract:
It is difficult to accurately diagnose the faults of power transformer due to its complex internal structure. Hence, this paper proposes a new method based on the k-nearest neighbor algorithm and the improved multi classification method to diagnose the faults of power transformer. Firstly, directed acyclic graph was used to define the fault categories. In order to reduce the complexity of the construction of multi classification models, the k-nearest neighbor algorithm was applied as a pre-classification method to filter a part of fault types. Next, the one-versus-one support vector machines(OVO-SVMs)trained by the sample data was used to calculate the probability of the fault types so as to form a probability matrix. Importantly, the probability gained from the OVR-SVMs was used as the weight coefficient of each element, based on which the probability matrix was updated as well as amended. As a result, the accuracy and reliability of fault diagnosis was improved. The results of actual diagnosis show the new method proposed in this paper, compared with the IEC three-ratio method and the traditional multi class support vector machines, expands the scope of fault diagnosis and improves the diagnostic accuracy and the efficiency.

参考文献/References:

[1] 张东波,徐瑜,王耀南.主动差异学习神经网络集成方法在变压器DGA故障诊断中的应用[J].中国电机工程学报,2010,30(22):64-70.
[2] Yang Q, Su P, Chen Y.Comparison of impulse wave and sweep frequency response analysis methods for diagnosis of transformer winding faults [J].Energies,2017,10(4):431.
[3] 潘翀,陈伟根,云玉新,等.基于遗传算法进化小波神经网络的电力变压器故障诊断[J].电力系统自动化,2007,31(13):88-92.
[4] 郭创新,朱承治,张琳,等.应用多分类多核学习支持向量机的变压器故障诊断方法[J].中国电机工程学报,2010,30(13):128-134.
[5] 郑蕊蕊,赵继印,赵婷婷,等.基于遗传支持向量机和灰色人工免疫算法的电力变压器故障诊断[J].中国电机工程学报,2011,31(7):56-63.
[6] 胡青,孙才新,杜林,等.核主成分分析与随机森林相结合的变压器故障诊断方法[J].高电压技术,2010,36(7):1725-1729.
[7] 唐勇波,桂卫华,彭涛,等.PCA和KICA特征提取的变压器故障诊断模型[J].高电压技术,2014,40(2):557-563.
[8] 毕建权,鹿鸣明,郭创新,等.一种基于多分类概率输出的变压器故障诊断方法[J].电力系统自动化,2015,39(5):88-93.
[9] Platt J.Probabilistic outputs for support vector machines and comparisons to regularized likelihood methods [J].Advances in Large Margin Classifiers,1999,10(3):61-74.
[10] Chmielnicki W, Stapor K. Using the one–versus–rest strategy with samples balancing to improve pairwise coupling classification[J]. International Journal of Applied Mathematics and Computer Science, 2016,26(1):191-201.
[11] 刘凡,张昀,姚晓,等.基于K近邻算法的换流变压器局部放电模式识别[J].电力自动化设备,2013,33(5):89-93.
[12] Galar M, Fernández A, Barrenechea E, et al.Dynamic classifier selection for one-vs-one strategy: avoiding non-competent classifiers[J]. Pattern Recognition, 2013, 46(12): 3412-3424.
[13] 田质广,张慧芬.基于遗传聚类算法的油中溶解气体分析电力变压器故障诊断[J].电力自动化设备,2008,28(2):15-18.

备注/Memo

备注/Memo:
收稿日期:2017-12-28
基金项目:南网重大科技项目(031300KK52150019); 国家自然科学基金项目(61304118)
作者简介:彭 刚(1981-),男,广东惠州人,硕士,高级工程师,主要从事输变电设备试验、监测与管理工作; 唐松平(1984-),男,广东龙川人,硕士,高级工程师,主要从事电力系统高压设备管理工作; 张作刚(1970-),男,山东安丘人,硕士,高级工程师,主要从事安全生产管理工作; 彭 杰(1989-),男,湖南宁乡人,工程师,主要从事输变电设备电
更新日期/Last Update: 2018-04-24