基于NNTR-SMOTE与GA-XGBoost的变压器故障诊断方法研究

doi:10.3969/j.issn.2097-0706.2024.01.010

摘要/Abstract

摘要：

针对变压器故障诊断中故障样本数量少且分布不均衡导致诊断率低的问题，提出了一种基于最近邻三角区域合成少数类过采样（NNTR-SMOTE）与利用遗传算法（GA）优化极端梯度提升（XGBoost）模型的变压器故障诊断方法。首先，将采集到的变压器故障样本数据进行标准化处理，使用NNTR-SMOTE方法得到平衡数据；其次，采用无编码比值法构造油中溶解气体的特征，得到特征数据集并对特征数据集采用多维尺度分析（MDS）方法进行特征融合；最后，利用GA对XGBoost模型的参数进行优化，构建变压器故障诊断模型。试验结果表明：基于NNTR-SMOTE与GA-XGBoost的变压器故障诊断方法诊断准确率达95.97%，不仅解决了诊断模型对多数类的偏向问题，还将模型的诊断精度进一步提高，适用于变压器非均衡数据集的多分类故障诊断。

关键词: 变压器, 故障诊断, 不平衡小样本, 极端梯度提升, 遗传算法

Abstract:

To address the low accuracy of transformer fault diagnosis caused by the insufficient number and uneven distribution of fault samples， a transformer fault diagnosis method based on nearest neighbour triangle regions synthetic minority oversampling technique （NNTR-SMOTE ）and genetic algorithm optimized extreme gradient boosting（GA-XGBoost）is proposed. Firstly， the transformer fault sample data are collected and standardized， and then balanced data are obtained by NNTR-SMOTE. Secondly， the feature data of dissolved gas are categorized by non-coding ratio method， and then fused by multi-dimensional scaling （MDS） method. Finally， a new transformer fault diagnosis model is constructed based on the XGBoost model optimized by GA. The experimental results show that the diagnostic accuracy of the proposed diagnosis method based on NNTR-SMOTE and GA-XGBoost reaches as high as 95.97%. This method not only solves the bias towards the majority class during diagnosis modelling， but also improves the diagnostic accuracy of the model， making it suitable for multi-classification fault diagnosis for transformers with unbalanced data.

Key words: transformer, fault diagnosis, unbalanced small sample, extreme gradient boosting, genetic algorithm

中图分类号:

TM411

汪李忠, 池建飞, 丁叶强, 姚海燕, 唐志鹏, 吴同宇. 基于NNTR-SMOTE与GA-XGBoost的变压器故障诊断方法研究[J]. 综合智慧能源, 2024, 46(1): 84-93.

WANG Lizhong, CHI Jianfei, DING Yeqiang, YAO Haiyan, TANG Zhipeng, WU Tongyu. Transformer fault diagnosis method based on NNTR-SMOTE and GA-XGBoost[J]. Integrated Intelligent Energy, 2024, 46(1): 84-93.

图/表 16

图1

图2

表1

混淆矩阵

项目	预测正类	预测负类
实际正类	$n T P$	$n F N$
实际负类	$n F P$	$n T N$

表1

表2

表3

表4

图3

表5

不同维数降维指标

降维指标	6维	5维	4维	3维	2维	1维
降维时间/s	2.46	2.48	2.53	2.57	2.59	2.64
$S s t r e s s$	0.014 7	0.015 3	0.016 4	0.017 8	0.019 1	0.021 6
Q_L	0.825 2	0.832 6	0.844 3	0.857 2	0.862 5	0.871 4
Q_G	0.403 2	0.416 4	0.422 0	0.430 2	0.438 5	0.449 2
Q_LG	0.614 2	0.624 5	0.633 2	0.643 7	0.650 5	0.660 3
准确率/%	91.84	92.36	93.52	95.45	95.97	90.63

表5

表6

图4

表7

表8

图5

表9

图6

表10

参考文献 32

[1]	YAN P C, ZHANG C Y, MEI K F, et al. Research on fault diagnosis of transformer based on laser induced fluorescence technology[J]. Journal of Molecular Structure, 2022, 1258: 132645. doi: 10.1016/j.molstruc.2022.132645
[2]	廖伟涵, 郭创新, 金宇, 等. 基于四阶段预处理与GBDT 的油浸式变压器故障诊断方法[J]. 电网技术, 2019, 43(6): 2195-2203.
	LIAO Weihan, GUO Chuangxin, JIN Yu, et al. Oil immersed transformer fault diagnosis method based on four-stage preprocessing and GBDT[J]. Power System Technology, 2019, 43(6):2195-2203.
[3]	MERVE D, HALUK G, CENGIZ M T. Improvement of power transformer fault diagnosis by using sequential Kalman filter sensor fusion[J]. International Journal of Electrical Power and Energy Systems, 2023, 149:109038. doi: 10.1016/j.ijepes.2023.109038
[4]	肖宏磊, 留毅, 夏红军, 等. 基于PCA与SSA-LightGBM的油浸式变压器故障诊断方法[J]. 综合智慧能源, 2023, 45(3): 9-16. doi: 10.3969/j.issn.2097-0706.2023.03.002
	XIAO Honglei, LIU Yi, XIA Hongjun, et al. Oil-immersed transformer fault diagnosis method based on PCA and SSA-LightGBM[J]. Integrated Intelligent Energy, 2023, 45(3): 9-16. doi: 10.3969/j.issn.2097-0706.2023.03.002
[5]	Mineral oil-impregnated electrical equipment in service-guide to the interpretation of dissolved and free gases analysis: IEC 60599—2007[S].
[6]	DAI J J, SONG H, SHENG G H, et al. Dissolved gas analysis of insulating oil for power transformer fault diagnosis with deep belief network[J]. IEEE Transactions on Dielectrics and Electrical Insulation, 2017, 24(5): 2828-2835. doi: 10.1109/TDEI.2017.006727
[7]	ROGERS R R. IEEE and IEC codes to interpret incipient faults in transformers,using gas in oil analysis[J]. IEEE Transactions on Electrical Insulation, 1978, 13 (5): 349-354.
[8]	薛华, 张宏鹏. 模糊无编码比值在变压器油中气体检测的应用[J]. 科学技术与工程, 2012, 12(33):9033-9037.
	XUE Hua, ZHANG Hongpeng. Fuzzy no coding ratio in transformer oil gas detection in the application[J]. Science Technology and Engineering, 2012, 12(33) :9033-9037.
[9]	QIN J, YANG D S, WANG N, et al. Convolutional sparse filter with data and mechanism fusion: A few-shot fault diagnosis method for power transformer[J]. Engineering Applications of Artificial Intelligence, 2023, 124:106606. doi: 10.1016/j.engappai.2023.106606
[10]	龙思成, 黄志鸿. 基于多尺度极限融合网络的电力变压器故障诊断方法研究[J]. 综合智慧能源, 2022, 44(9): 78-83. doi: 10.3969/j.issn.2097-0706.2022.09.011
	LONG Sicheng, HUANG Zhihong. Research on fault diagnosis method of power transformers based on multi-scale extreme fusion network[J]. Integrated Intelligent Energy, 2022, 44(9): 78-83. doi: 10.3969/j.issn.2097-0706.2022.09.011
[11]	朱永利, 吴立增, 李雪玉. 贝叶斯分类器与粗糙集相结合的变压器综合故障诊断[J]. 中国电机工程学报, 2005, 25(10):159-165.
	ZHU Yongli, WU Lizeng, LI Xueyu. Synthesized diagnosis on transformer faults based on Bayesian classifier and rough set[J]. Proceedings of the Csee, 2005, 25 (10): 159-165.
[12]	姜寅, 任荭葳, 陈凯, 等. 基于BP神经网络的风电系统控制器I/O硬件故障自诊断方法[J]. 华电技术, 2020, 42(5): 55-60.
	JIANG Yin, REN Hongwei, CHEN Kai, et al. Self-detecting method for faults in wind turbine controller I/O hardware based on BP neural network algorithm[J]. Huadian Technology, 2020, 42(5): 55-60.
[13]	李金阔, 王秀平. 基于循环BP模型的变压器状态数据清洗方法[J]. 电力学报, 2023, 38(2):101-109.
	LI Jinkuo, WANG Xiuping. Online detection method of transformer capacity based on cyclic BP model[J]. Journal of Electric Power, 2023, 38(2):101-109.
[14]	李勇琦, 雷旗开, 王浩, 等. 基于BP神经网络的梯次利用电池健康状态诊断[J]. 华电技术, 2021, 43(7): 42-46.
	LI Yongqi, LEI Qikai, WANG Hao, et al. State of health estimation for echelon-used batteries based on BP neural network[J]. Huadian Technology, 2021, 43(7): 42-46.
[15]	王永强, 律方成, 李和明. 基于粗糙集理论和贝叶斯网络的电力变压器故障诊断方法[J]. 中国电机工程学报, 2006, 26(8):137-141.
	WANG Yongqiang, LV Fangcheng, LI Heming. Synthetic fault diagnosis method of power transformer based on rough set theory and Bayesian network[J]. Proceedings of the CSEE, 2006, 26 (8): 137-141.
[16]	陈欢, 彭辉, 舒乃秋, 等. 基于蝙蝠算法优化最小二乘双支持向量机的变压器故障诊断[J]. 高电压技术, 2018, 44(11):3664-3671.
	CHEN Huan, PENG Hui, SHU Naiqiu, et al. Fault diagnosis of transformer based on LS-TSVM optimized by bat algorithm[J]. High Voltage Engineering, 2018, 44 (11): 3664-3671.
[17]	魏伟, 高赐威, 宋梦, 等. 基于HHO-SVM的电力线路故障检测分类系统[J]. 综合智慧能源, 2023, 45(3): 1-8. doi: 10.3969/j.issn.2097-0706.2023.03.001
	WEI Wei, GAO Ciwei, SONG Meng, et al. Power line fault detection and classification system based on HHO-SVM[J]. Integrated Intelligent Energy, 2023, 45(3):1-8. doi: 10.3969/j.issn.2097-0706.2023.03.001
[18]	李赢, 舒乃秋. 基于模糊聚类和完全二叉树支持向量机的变压器故障诊断[J]. 电工技术学报, 2016, 31(4): 64-70.
	LI Ying, SHU Naiqiu. Transformer fault diagnosis based on fuzzy clustering and complete binary tree support vector machine[J]. Transactions of China Electrotechnical Society, 2016, 31(4): 64-70.
[19]	ZHOU Y, LIU Y Y, WANG N, et al. Partial discharge ultrasonic signals pattern recognition in transformer using BSO-SVM based on microfiber coupler sensor[J]. Measurement, 2022, 201:111737. doi: 10.1016/j.measurement.2022.111737
[20]	CHEN T, TONG H. Higgs boson discovery with boosted trees[C]// International Conference on High-energy Physics & Machine Learning, 2015:69-80.
[21]	苌瑶, 武建文, 马速良, 等. 基于多特征评估与XGBoost的高压断路器故障诊断[J]. 高压电器, 2023, 59(4):1-9.
	CHANG Yao, WU Jianwen, MA Suliang, et al. Fault diagnosis of high voltage circuit breaker based on multi-feature assessment and XGBoost[J]. High Voltage Apparatus, 2023, 59 (4): 1-9.
[22]	谭雨欣, 田义超, 黄卓梅, 等. 北部湾茅尾海无瓣海桑红树林地上生物量反演——基于XGBoost机器学习算法[J]. 生态学报, 2023, 43(11): 4674-4688.
	TAN Yuxin, TIAN Yichao, HUANG Zhuomei, et al. Aboveground biomass of Sonneratia apetala mangroves in Maowei Sea of Beibu Gulf based on XGBoost machine learning algorithm[J]. Acta Ecologica Sinica, 2023, 43 (11): 4674-4688.
[23]	商俊燕, 丁辉, 胡学龙. 基于XGBoost的无线传感器网络冗余数据检测算法[J]. 传感技术学报, 2022, 35(11): 1568-1572.
	SHANG Junyan, DING Hui, HU Xuelong. Redundant data detection algorithm for wireless sensor network based on XGBoost[J]. Chinese Journal of Sensors and Actuators, 2022, 35 (11): 1568-1572.
[24]	TAN X M, QI J, GAN J Q, et al. Multi-filter semi-supervised transformer model for fault diagnosis[J]. Engineering Applications of Artificial Intelligence, 2023, 124:106498. doi: 10.1016/j.engappai.2023.106498
[25]	CHAWLA N V, BOWYER K W, HALL L O, et al. Smote: Synthetic minority over-sampling technique[J]. Journal of Artificial Intelligence Research, 2002, 16: 321-357. doi: 10.1613/jair.953
[26]	DEBABRATA D, TAMÁS S. Multidimensional stationary time series: Dimension reduction and prediction[M]. CRC Press: 2021.
[27]	马永杰, 云文霞. 遗传算法研究进展[J]. 计算机应用研究, 2012, 29(4):1201-1206,1210.
	MA Yongjie, YUN Wenxia. Research progress of genetic algorithm[J]. Application Research of Computers, 2012, 29(4):1201-1206,1210.
[28]	汪可, 李金忠, 张书琦, 等. 变压器故障诊断用油中溶解气体新特征参量[J]. 中国电机工程学报, 2016, 36(23):6570-6578.
	WANG Ke, LI Jinzhong, ZHANG Shuqi, et al. New features derived from dissolved gas analysis for fault diagnosis of power transformers[J]. Proceedings of the CSEE, 2016, 36(23): 6570-6578.
[29]	MAATEN L, HINTON G. Visualizing data using t-SNE[J]. Journal of Machine Learning Research, 2008, 9(2): 2579-2605.
[30]	樊清川, 于飞, 宣敏. 基于优化Bi-LSTM模型的电力变压器故障诊断[J]. 计算机仿真, 2022, 39(11):136-140.
	FAN Qingchuan, YU Fei, XUAN Min. Power transformer fault diagnosis based on optimized Bi-LSTM model[J]. Computer Simulation, 2022, 39(11): 136-140.
[31]	康佳宇, 张沈习, 张庆平, 等. 基于ANOVA和BO-SVM的变压器故障诊断方法[J]. 高电压技术, 2023, 49(5): 1882-1891.
	KANG Jiayu, ZHANG Shenxi, ZHANG Qingping, et al. Fault diagnosis method of transformer based on ANOVA and BO-SVM[J]. High Voltage Engineering, 2023, 49 (5): 1882-1891
[32]	安国庆, 史哲文, 马世峰, 等. 基于RF特征优选的WOA-SVM变压器故障诊断[J]. 高压电器, 2022, 58(2):171-178.
	AN Guoqing, SHI Zhewen, MA Shifeng, et al. Fault diagnosis of WOA-SVM transformer based on RF feature optimization[J]. High Voltage Apparatus, 2022, 58(2): 171-178.

状态类型	样本数量	占比/%	类别标签
正常状态	91	21.41	1
中温过热	46	10.82	2
中低温过热	33	7.77	3
高温过热	53	12.47	4
放电兼过热	17	4.00	5
局部放电	29	6.82	6
低能放电	72	16.95	7
高能放电	84	19.76	8

特征编码	特征量	特征编码	特征量
1	CH₄/H₂	10	C₂H₄/THC
2	C₂H₂/H₂	11	C₂H₆/THC
3	C₂H₂/C₂H₄	12	C₂H₂/THC
4	C₂H₄/C₂H₆	13	(CH₄+C₂H₄)/THC
5	C₂H₆/CH₄	14	H₂/ALL
6	C₂H₂/CH₄	15	CH₄/ALL
7	C₂H₄/CH₄	16	C₂H₂/ALL
8	H₂/THC	17	C₂H₄/ALL
9	CH₄/THC	18	C₂H₆/ALL

运行状态	查全率/%	查准率/%	误报率/%	漏报率/%	Kappa系数	准确率/%
正常状态	94.4	89.5	5.6	10.5	0.954 2	95.97
中温过热	94.4	100.0	5.6	0
中低温过热	100.0	100.0	0	0
高温过热	94.7	100.0	5.3	0
放电兼过热	89.5	89.5	10.5	10.5
局部放电	100.0	100.0	0	0
低能放电	94.7	90.0	5.3	10.0
高能放电	100.0	100.0	0	0

采样方法	准确率/%	Kappa系数
NNTR-SMOTE	95.97	0.954 2
SMOTE-ENN	93.26	0.939 5
ADASYN	91.75	0.913 3
SMOTE	90.56	0.896 1
随机过采样	89.32	0.884 9
非均衡数据集	86.32	0.854 2

评价指标	MDS融合特征	PCA融合特征	LDA融合特征	LLE融合特征	18维联合特征
诊断准确率/%	95.97	92.02	91.30	89.13	87.68
Kappa系数	0.954 2	0.937 3	0.912 6	0.900 2	0.875 4