CHPOA-DBN Transformer Fault Diagnosis Method Considering Sample Within-Class Imbalance

In recent years, deep belief network (DBN) based transformer fault diagnosis methods have been developed. However, they share two prominent drawbacks, which are the low accuracy issue caused by the within-class imbalance of transformer faults samples and the artificial determination of the network p...

Full description

Saved in:
Bibliographic Details
Main Authors: Shuang WANG, Qian LUO, Bo TANG, Lan JIANG, Jin LI
Format: Article
Language:zho
Published: State Grid Energy Research Institute 2023-08-01
Series:Zhongguo dianli
Subjects:
Online Access:https://www.electricpower.com.cn/CN/10.11930/j.issn.1004-9649.202305039
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1850070280583512064
author Shuang WANG
Qian LUO
Bo TANG
Lan JIANG
Jin LI
author_facet Shuang WANG
Qian LUO
Bo TANG
Lan JIANG
Jin LI
author_sort Shuang WANG
collection DOAJ
description In recent years, deep belief network (DBN) based transformer fault diagnosis methods have been developed. However, they share two prominent drawbacks, which are the low accuracy issue caused by the within-class imbalance of transformer faults samples and the artificial determination of the network parameters of deep belief network (DBN). In this paper, a transformer fault diagnosis method based on sample balance processing and improved DBN is proposed. Firstly, an improved K-means (IK-means) synthesis minority oversampling technique (SMOTE) algorithm is proposed to obtain within-class and between-class balanced fault samples. Then, the Tent chaotic map embedded chaotic hybrid pelican optimization algorithm (CHPOA) is developed to optimize the number of hidden layer nodes and reverse fine-tuning learning rate of DBN, and the CHPOA-DBN transformer fault diagnosis model is constructed. Finally, the classical oversampling algorithm, the classical fault diagnosis model and the proposed method are compared and analyzed, based on the experimental data, respectively. The results show that the fault diagnosis accuracy of the proposed method reaches 96.25 %, which provide an important reference for intelligent fault diagnosis under imbalanced fault samples of transformers.
format Article
id doaj-art-9573b40d147e485cb20e4f60501ea924
institution DOAJ
issn 1004-9649
language zho
publishDate 2023-08-01
publisher State Grid Energy Research Institute
record_format Article
series Zhongguo dianli
spelling doaj-art-9573b40d147e485cb20e4f60501ea9242025-08-20T02:47:35ZzhoState Grid Energy Research InstituteZhongguo dianli1004-96492023-08-01561013314410.11930/j.issn.1004-9649.202305039zgdl-56-09-wangshuangCHPOA-DBN Transformer Fault Diagnosis Method Considering Sample Within-Class ImbalanceShuang WANG0Qian LUO1Bo TANG2Lan JIANG3Jin LI4School of Electrical Engineering and New Energy, China Three Gorges University, Yichang 443002, ChinaSchool of Electrical Engineering and New Energy, China Three Gorges University, Yichang 443002, ChinaSchool of Electrical Engineering and New Energy, China Three Gorges University, Yichang 443002, ChinaSchool of Electrical Engineering and New Energy, China Three Gorges University, Yichang 443002, ChinaState Grid Hubei Extra High Voltage Company, Wuhan 430051, ChinaIn recent years, deep belief network (DBN) based transformer fault diagnosis methods have been developed. However, they share two prominent drawbacks, which are the low accuracy issue caused by the within-class imbalance of transformer faults samples and the artificial determination of the network parameters of deep belief network (DBN). In this paper, a transformer fault diagnosis method based on sample balance processing and improved DBN is proposed. Firstly, an improved K-means (IK-means) synthesis minority oversampling technique (SMOTE) algorithm is proposed to obtain within-class and between-class balanced fault samples. Then, the Tent chaotic map embedded chaotic hybrid pelican optimization algorithm (CHPOA) is developed to optimize the number of hidden layer nodes and reverse fine-tuning learning rate of DBN, and the CHPOA-DBN transformer fault diagnosis model is constructed. Finally, the classical oversampling algorithm, the classical fault diagnosis model and the proposed method are compared and analyzed, based on the experimental data, respectively. The results show that the fault diagnosis accuracy of the proposed method reaches 96.25 %, which provide an important reference for intelligent fault diagnosis under imbalanced fault samples of transformers.https://www.electricpower.com.cn/CN/10.11930/j.issn.1004-9649.202305039transformer fault diagnosiswithin-class imbalancesample balance processingtent chaotic mapdbn network parameters optimization
spellingShingle Shuang WANG
Qian LUO
Bo TANG
Lan JIANG
Jin LI
CHPOA-DBN Transformer Fault Diagnosis Method Considering Sample Within-Class Imbalance
Zhongguo dianli
transformer fault diagnosis
within-class imbalance
sample balance processing
tent chaotic map
dbn network parameters optimization
title CHPOA-DBN Transformer Fault Diagnosis Method Considering Sample Within-Class Imbalance
title_full CHPOA-DBN Transformer Fault Diagnosis Method Considering Sample Within-Class Imbalance
title_fullStr CHPOA-DBN Transformer Fault Diagnosis Method Considering Sample Within-Class Imbalance
title_full_unstemmed CHPOA-DBN Transformer Fault Diagnosis Method Considering Sample Within-Class Imbalance
title_short CHPOA-DBN Transformer Fault Diagnosis Method Considering Sample Within-Class Imbalance
title_sort chpoa dbn transformer fault diagnosis method considering sample within class imbalance
topic transformer fault diagnosis
within-class imbalance
sample balance processing
tent chaotic map
dbn network parameters optimization
url https://www.electricpower.com.cn/CN/10.11930/j.issn.1004-9649.202305039
work_keys_str_mv AT shuangwang chpoadbntransformerfaultdiagnosismethodconsideringsamplewithinclassimbalance
AT qianluo chpoadbntransformerfaultdiagnosismethodconsideringsamplewithinclassimbalance
AT botang chpoadbntransformerfaultdiagnosismethodconsideringsamplewithinclassimbalance
AT lanjiang chpoadbntransformerfaultdiagnosismethodconsideringsamplewithinclassimbalance
AT jinli chpoadbntransformerfaultdiagnosismethodconsideringsamplewithinclassimbalance