Scholarlink

<p>Dissolved gas analysis (DGA) is widely used for transformer fault diagnosis, but often suffers from limited accuracy. To overcome this challenge, this study proposes a novel diagnostic framework combining sparse autoencoder-based dimensionality reduction with a Bayesian-optimized bidirectional long short-term memory network (SAE-BO-BiLSTM). Firstly, to improve the accuracy of transformer fault diagnosis and fully characterize the fault information, the fault data are paired two by two to generate the fault data for DGA diagnosis; Then, SAE is used to synthesize the nonlinear high-dimensional DGA gas data of the transformer so that synthesis of new data can still represent the fault information of the transformer and prevent the redundant information from interfering with the diagnostic accuracy; Secondly, a Bayes-BiLSTM based fault diagnosis model is established to obtain the optimal parameters of the BiLSTM using BO Algorithm; Finally, a comparative analysis of multiple transformer fault diagnosis models shows that the SAE-BO-BiLSTM network model has higher fault diagnosis accuracy compared to Characteristic Gas Method (CGM), Support Vector Machine (SVM), Random Forest Algorithm (RF), Convolutional Neural Network (CNN), and Particle Swarm Optimization-Bidirectional Long Short-Term Memory (PSO-BiLSTM) network models.</p>

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 Research Based on SAE-BO-BiLSTM

  • 期刊论文

  • 发表日期：

    2025年8月8日

查看全文
  • Shuxin Liu,
  • Simiao Yu,
  • Yankai Li

摘要

Dissolved gas analysis (DGA) is widely used for transformer fault diagnosis, but often suffers from limited accuracy. To overcome this challenge, this study proposes a novel diagnostic framework combining sparse autoencoder-based dimensionality reduction with a Bayesian-optimized bidirectional long short-term memory network (SAE-BO-BiLSTM). Firstly, to improve the accuracy of transformer fault diagnosis and fully characterize the fault information, the fault data are paired two by two to generate the fault data for DGA diagnosis; Then, SAE is used to synthesize the nonlinear high-dimensional DGA gas data of the transformer so that synthesis of new data can still represent the fault information of the transformer and prevent the redundant information from interfering with the diagnostic accuracy; Secondly, a Bayes-BiLSTM based fault diagnosis model is established to obtain the optimal parameters of the BiLSTM using BO Algorithm; Finally, a comparative analysis of multiple transformer fault diagnosis models shows that the SAE-BO-BiLSTM network model has higher fault diagnosis accuracy compared to Characteristic Gas Method (CGM), Support Vector Machine (SVM), Random Forest Algorithm (RF), Convolutional Neural Network (CNN), and Particle Swarm Optimization-Bidirectional Long Short-Term Memory (PSO-BiLSTM) network models.