Scholarlink

<p>Distributed generation (DG) enhances power system reliability by minimizing transmission losses, reducing reliance on centralized plants, lowering voltage fluctuations, and supporting renewable integration to reduce carbon emissions while improving grid resilience during extreme events or failures. Nevertheless, the extensive integration of DGs introduces challenges such as unintentional islanding, where energy segments are isolated from the grid, undermining system reliability and efficiency, especially as traditional detection methods fail in high-DG penetration scenarios. Unintentional island identification becomes challenging when local generation closely matches the load demand, requiring advanced solutions like phasor measurement units (PMUs) to monitor the phase angle, frequency, and signal magnitude. While PMUs improve detection accuracy, they necessitate vast amounts of data processing and automated methods because of the complexity of system events and potential islanding scenarios. Hence, stringent criteria exist for precise, rapid, and dependable identification of islanding in renewable and DG-based systems. Recently, intelligent techniques have garnered attention because of their exceptional characteristics and advantages compared to conventional approaches. Considering this, this study has proposed a PMU-based method for real-time detection of unintentional islanding via a classification-enabled machine learning algorithm. The extraction of parameters and data from the modified IEEE 30-bus standard test system is conducted in the PowerWorld simulator environment, while the training and testing of the proposed technique are performed in Python. The evaluation employs performance indicators such as 94.29% accuracy, 97% F1-score, 1 precision, 0.93 recall, 0.810 MCC score, 0.8 Cohen’s kappa score, and 1 ROC-AUC score, with supervised and classification-based machine learning algorithms exhibiting commendable performance.</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

Unintentional islanding detection in electrical power distribution systems via PMU and a classification-enabled machine learning algorithm

  • 期刊论文

  • 发表日期：

    2026年1月9日

查看全文
  • Md Siddikur Rahman,
  • Mohd Fakhizan Bin Romlie,
  • Khairul Nisak Binti Md Hasan,
  • Mohd Faris Bin Abdullah,
  • Taib Bin Ibrahim

摘要

Distributed generation (DG) enhances power system reliability by minimizing transmission losses, reducing reliance on centralized plants, lowering voltage fluctuations, and supporting renewable integration to reduce carbon emissions while improving grid resilience during extreme events or failures. Nevertheless, the extensive integration of DGs introduces challenges such as unintentional islanding, where energy segments are isolated from the grid, undermining system reliability and efficiency, especially as traditional detection methods fail in high-DG penetration scenarios. Unintentional island identification becomes challenging when local generation closely matches the load demand, requiring advanced solutions like phasor measurement units (PMUs) to monitor the phase angle, frequency, and signal magnitude. While PMUs improve detection accuracy, they necessitate vast amounts of data processing and automated methods because of the complexity of system events and potential islanding scenarios. Hence, stringent criteria exist for precise, rapid, and dependable identification of islanding in renewable and DG-based systems. Recently, intelligent techniques have garnered attention because of their exceptional characteristics and advantages compared to conventional approaches. Considering this, this study has proposed a PMU-based method for real-time detection of unintentional islanding via a classification-enabled machine learning algorithm. The extraction of parameters and data from the modified IEEE 30-bus standard test system is conducted in the PowerWorld simulator environment, while the training and testing of the proposed technique are performed in Python. The evaluation employs performance indicators such as 94.29% accuracy, 97% F1-score, 1 precision, 0.93 recall, 0.810 MCC score, 0.8 Cohen’s kappa score, and 1 ROC-AUC score, with supervised and classification-based machine learning algorithms exhibiting commendable performance.