Scholarlink

<p>Cast-resin transformers have been widely used in electrical power systems for decades because of their safety and reliability. However, once partial discharge (PD) occurs due to internal defects, it progressively deteriorates the insulation system, which eventually may lead to severe failures. This paper deals with machine learning (ML)-based PD recognition in the cast-resin transformers using an embedded phase-resolved partial discharge (PRPD) sensor, which can detect PD signals phase-synchronized with HV signals. Four typical insulation defect models, floating metal, particle on epoxy resin, surface discharge, and cracked epoxy resin, were prepared, and PD single pulses and PRPD patterns were measured at partial discharge inception voltage (PDIV) levels of each defect. From the measured PD signals, a total of 23 PD features were systematically extracted to construct PD datasets. Representative ML algorithms, including Random Forest (RF), Support Vector Machine (SVM), Artificial Neural Network (ANN), and <i>k</i>-Nearest Neighbor (<i>k</i>NN), were applied to classify the PD defects, and their performances were quantitatively evaluated using multiple statistical indices. The experimental results demonstrated that the RF algorithm model achieved the highest accuracy over 96%, closely followed by the ANN model. The findings confirm that the proposed detection and classification methods enable reliable and accurate PD recognition, providing advanced condition monitoring of cast-resin transformers.</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

Machine Learning-Based Partial Discharge Recognition in Cast-resin transformers Using Phase-Synchronized Partial Discharge Detection Technique

  • 期刊论文

  • 发表日期：

    2026年4月28日

查看全文
  • Gyeong-Yeol Lee,
  • Gyung-Suk Kil

摘要

Cast-resin transformers have been widely used in electrical power systems for decades because of their safety and reliability. However, once partial discharge (PD) occurs due to internal defects, it progressively deteriorates the insulation system, which eventually may lead to severe failures. This paper deals with machine learning (ML)-based PD recognition in the cast-resin transformers using an embedded phase-resolved partial discharge (PRPD) sensor, which can detect PD signals phase-synchronized with HV signals. Four typical insulation defect models, floating metal, particle on epoxy resin, surface discharge, and cracked epoxy resin, were prepared, and PD single pulses and PRPD patterns were measured at partial discharge inception voltage (PDIV) levels of each defect. From the measured PD signals, a total of 23 PD features were systematically extracted to construct PD datasets. Representative ML algorithms, including Random Forest (RF), Support Vector Machine (SVM), Artificial Neural Network (ANN), and k-Nearest Neighbor (kNN), were applied to classify the PD defects, and their performances were quantitatively evaluated using multiple statistical indices. The experimental results demonstrated that the RF algorithm model achieved the highest accuracy over 96%, closely followed by the ANN model. The findings confirm that the proposed detection and classification methods enable reliable and accurate PD recognition, providing advanced condition monitoring of cast-resin transformers.