Scholarlink

<p>This paper presents a machine learning-based approach for fault location estimation, fault type classification, and fault section identification under restricted conditions, such as measurement errors and minimal input data. Data were collected using PSCAD/EMTDC and validated on a radial distribution system and an IEEE-34 node test feeder with distributed generators. Artificial neural networks, wavelet neural networks, support vector machines, and impedance-based approaches were compared for fault distance estimation in a radial system to assess the limitations of conventional methods. Decision trees, k-nearest neighbors, support vector machines, and ensemble approaches were employed for fault type and section classification. The radial distribution system utilized substation-side data, while the IEEE-34 node test feeder incorporated data from both distributed generator and substation sides for verification. Additionally, Gaussian noise levels of 1%, 5%, and 10% were introduced to assess error impact, leading to the proposal of a robust fault location estimation model.</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

Application and Comparison of Machine Learning-Based Fault Location Estimation Methods in a Power Distribution Network Under Restricted Conditions

  • 期刊论文

  • 发表日期：

    2026年1月7日

查看全文
  • Chul-Sang Hwang,
  • Shahid Hussain,
  • Ji-Yeon Kang,
  • Byuk-Keun Jo,
  • Tae-Jin Kim,
  • Yun-Su Kim

摘要

This paper presents a machine learning-based approach for fault location estimation, fault type classification, and fault section identification under restricted conditions, such as measurement errors and minimal input data. Data were collected using PSCAD/EMTDC and validated on a radial distribution system and an IEEE-34 node test feeder with distributed generators. Artificial neural networks, wavelet neural networks, support vector machines, and impedance-based approaches were compared for fault distance estimation in a radial system to assess the limitations of conventional methods. Decision trees, k-nearest neighbors, support vector machines, and ensemble approaches were employed for fault type and section classification. The radial distribution system utilized substation-side data, while the IEEE-34 node test feeder incorporated data from both distributed generator and substation sides for verification. Additionally, Gaussian noise levels of 1%, 5%, and 10% were introduced to assess error impact, leading to the proposal of a robust fault location estimation model.