Scholarlink

As the key to voltage conversion, it is very important to realize its online monitoring and fault diagnosis of transformer. At present, Artificial Intelligence (AI) algorithm is typically utilized to identify transformer faults. However, the operation results of AI algorithms are often biased towards the majority of samples, and the transformer oil chromatographic fault data is precisely an unbalanced data set, which will have a great impact on the final diagnosis results. Therefore, this study first uses the optimized Synthetic Minority Oversampling Technique (SMOTE) to balance the transformer oil chromatographic fault data, and then uses the Random Forest (RF) algorithm to diagnose the transformer fault. To confirm the effectiveness of the optimized SMOTE in expanding unbalanced data, the data processed with SMOTE and the data without SMOTE are used to compare the diagnostic results. In order to confirm the superiority of the selected RF algorithm, two popular AI algorithm models, Support Vector Machine (SVM) and Naive Bayesian Model (NBM), are constructed as the control group of diagnosis results. At the same time, the actual substation data and sample data are used to verify the final transformer fault diagnosis model based on SMOTE and RF to verify the above conclusions.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Transformer Fault Diagnosis Method Based on the Combination of Optimized Synthetic Minority Oversampling Technology and Random Forest Algorithm

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Xiangchun Wen,
  • Li Zhang,
  • N. N. Kurbonov,
  • I. U. Rakhmonov,
  • Guan Wang

摘要

As the key to voltage conversion, it is very important to realize its online monitoring and fault diagnosis of transformer. At present, Artificial Intelligence (AI) algorithm is typically utilized to identify transformer faults. However, the operation results of AI algorithms are often biased towards the majority of samples, and the transformer oil chromatographic fault data is precisely an unbalanced data set, which will have a great impact on the final diagnosis results. Therefore, this study first uses the optimized Synthetic Minority Oversampling Technique (SMOTE) to balance the transformer oil chromatographic fault data, and then uses the Random Forest (RF) algorithm to diagnose the transformer fault. To confirm the effectiveness of the optimized SMOTE in expanding unbalanced data, the data processed with SMOTE and the data without SMOTE are used to compare the diagnostic results. In order to confirm the superiority of the selected RF algorithm, two popular AI algorithm models, Support Vector Machine (SVM) and Naive Bayesian Model (NBM), are constructed as the control group of diagnosis results. At the same time, the actual substation data and sample data are used to verify the final transformer fault diagnosis model based on SMOTE and RF to verify the above conclusions.