An Improved DGA Feature Clustering-Based Method for Transformer Fault Diagnosis
The power transformer is the core equipment of a power system, and its reliable operation is crucial for maintaining the safety and stability of power grids. Dissolved gases in insulating oil are an important information source for analyzing transformer operating status and fault diagnosis. At prese...
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
China electric power research institute
2025-01-01
|
| Series: | CSEE Journal of Power and Energy Systems |
| Subjects: | |
| Online Access: | https://ieeexplore.ieee.org/document/9877998/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | The power transformer is the core equipment of a power system, and its reliable operation is crucial for maintaining the safety and stability of power grids. Dissolved gases in insulating oil are an important information source for analyzing transformer operating status and fault diagnosis. At present, intelligent fault diagnosis methods for power transformers are mostly based on classification algorithms, but the diagnosis models may be relatively complicated. Some models have poor generalization ability when training samples are scarce. Clustering algorithms can better deal with this problem. Fault diagnosis of transformers based on a clustering algorithm primarily utilizes the proportional data of dissolved gases as features, which have not considered abundant gas ratio features, and those clustering methods are prone to invalid clustering. In order to solve those problems, this paper uses more features as information sources of power transformer diagnosis based on clustering method. Different clustering spaces are considered for different fault types. Clustering centers are found on samples with the same fault type, which aims to expand the data distribution difference in different fault types. This paper also uses genetic algorithm (GA) to optimize multiple data clustering spaces and improve clustering effect. Based on multiple data sets, it is verified that the proposed method can effectively avoid the occurrence of invalid clustering, and the difference among different fault types based on multiple clustering spaces method is more obvious. |
|---|---|
| ISSN: | 2096-0042 |