| 引用本文: | 栗磊,王廷涛,赫嘉楠,牛健,梁亚波,苗世洪.考虑过采样器与分类器参数优化的变压器故障诊断策略[J].电力自动化设备,2023,43(1): |
| LI Lei,WANG Tingtao,HE Jianan,NIU Jian,LIANG Yabo,MIAO Shihong.Transformer fault diagnosis strategy considering parameter optimization of oversampler and classifier[J].Electric Power Automation Equipment,2023,43(1): |
|
| 摘要: |
| 变压器故障样本的不平衡性使得故障诊断分类准确率低，且容易弱化少数类故障样本的分类效果。对此，采用过采样方法实现故障样本的均衡化，并提出一种考虑过采样器与分类器参数优化的变压器故障诊断策略。首先，搭建变压器故障诊断模型的整体结构，阐述故障诊断的实现过程。在此基础上，提出诊断模型中过采样器、分类器、参数优化器3种主要环节的算法实现：针对过采样器，提出一种基于近邻分布特性的改进合成少数过采样算法实现故障样本的均衡化；针对分类器，采用层次式有向无环图支持向量机算法实现故障样本的多标签分类；针对参数优化器，提出一种双层参数优化方法，上层采用层次搜索算法对过采样倍率寻优，下层采用改进哈里斯鹰算法对支持向量机参数寻优。最后，对所提策略进行算例分析，结果表明，所提策略能够合成质量更高的少数类故障样本，实现故障样本的准确分类。 |
| 关键词: 电力变压器 故障诊断 不平衡样本 过采样 基于近邻分布特性的改进合成少数过采样 层次搜索-改进哈里斯鹰算法 |
| DOI：10.16081/j.epae.202206011 |
| 分类号:TM41 |
| 基金项目:国网宁夏电力有限公司科技项目(5229DK20004Q) |
|
| Transformer fault diagnosis strategy considering parameter optimization of oversampler and classifier |
|
LI Lei1, WANG Tingtao2, HE Jianan1, NIU Jian1, LIANG Yabo1, MIAO Shihong2
|
|
1.Electric Power Research Institute, State Grid Ningxia Electric Power Co.,Ltd.,Yinchuan 750002, China;2.School of Electrical and Electronic Engineering, Huazhong University of Science and Technology, Wuhan 430074, China
|
| Abstract: |
| The imbalance of transformer fault samples makes the accuracy of fault diagnosis and classification low, and it is easy to weaken the classification effect of a few types of fault samples. Therefore, the oversampling method is used to realize the equalization of fault samples, and a transformer fault diagnosis strategy considering the parameter optimization of oversampler and classifier is proposed. Firstly, the overall structure of transformer fault diagnosis model is built, and the implementation process of fault diagnosis is described. On this basis, the algorithm implementation of three main links of oversampler, classifier and parameter optimizer in the diagnosis model is proposed. For the oversampler, an improved synthetic minority oversampling technique based on nearest neighbor distribution(SMOTE-NND) algorithm is proposed to realize the equalization of fault samples. For the classifier, the hierarchical directed acyclic graph support vector machine(HDAG-SVM) algorithm is used to realize the multi-label classification of fault samples. For the parameter optimizer, a double-layer parameter optimization method is proposed. The upper layer uses the hierarchical search(HS) algorithm to optimize the oversampling ratio, and the lower layer uses modified harris hawks optimization(MHHO) algorithm to optimize the parameters of support vector machine. Finally, an example is given to analyze the proposed strategy. The results show that the proposed strategy can synthesize a few fault samples with higher quality and realize the accurate classification of fault samples. |
| Key words: power transformers fault diagnosis unbalanced samples oversample SMOTE-NND HS-MHHO algorithm |
