Scholarlink

<p>As a key component of the smart grid, ensuring the operational reliability of converter transformers is essential. Machine learning techniques such as support vector machines (SVM), genetic algorithms, and neural networks have shown effectiveness in partial discharge (PD) recognition. In this study, Rogowski coils were used to detect PD signals, and a new pattern recognition method was proposed. Four artificial insulation defect models were designed to generate PD signals under combined AC-DC voltages. Statistical PD histograms were employed for feature extraction, and their parameters were further processed using linear discriminant analysis (LDA) to reduce dimensionality. An adaptive particle swarm optimized support vector machine (APSO-SVM) was applied for defect classification. Comparative experiments with Genetic Algorithm optimized SVM (GA-SVM) and PSO-SVM were conducted, with each recognition test repeated 100 times in MATLAB 2021a to verify robustness. Results demonstrate that APSO-SVM achieved a recognition accuracy of 94.87%, which is 9.21% and 8.1% higher than GA-SVM and PSO-SVM, respectively. The proposed feature extraction and classification method is effective for PD pattern recognition and can also be extended to other transient signals, enhancing intelligent perception and fault management in smart grids.</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

APSO-SVM Based Approach for Partial Discharge Pattern Recognition in Converter Transformer

  • 期刊论文

  • 发表日期：

    2025年9月24日

查看全文
  • Tianyan Jiang,
  • Lin Chen,
  • Haoxiang Yuan,
  • Sirun Tan,
  • Haonan Xie,
  • Maoqiang Bi,
  • Xi Chen

摘要

As a key component of the smart grid, ensuring the operational reliability of converter transformers is essential. Machine learning techniques such as support vector machines (SVM), genetic algorithms, and neural networks have shown effectiveness in partial discharge (PD) recognition. In this study, Rogowski coils were used to detect PD signals, and a new pattern recognition method was proposed. Four artificial insulation defect models were designed to generate PD signals under combined AC-DC voltages. Statistical PD histograms were employed for feature extraction, and their parameters were further processed using linear discriminant analysis (LDA) to reduce dimensionality. An adaptive particle swarm optimized support vector machine (APSO-SVM) was applied for defect classification. Comparative experiments with Genetic Algorithm optimized SVM (GA-SVM) and PSO-SVM were conducted, with each recognition test repeated 100 times in MATLAB 2021a to verify robustness. Results demonstrate that APSO-SVM achieved a recognition accuracy of 94.87%, which is 9.21% and 8.1% higher than GA-SVM and PSO-SVM, respectively. The proposed feature extraction and classification method is effective for PD pattern recognition and can also be extended to other transient signals, enhancing intelligent perception and fault management in smart grids.