Scholarlink

<p>The increasing integration of photovoltaic (PV) sources in DC microgrids introduces significant protection challenges due to power intermittency, converter switching dynamics, and frequent reconfiguration of distributed generation units. This paper presents a resilient hybrid protection framework combining the Hilbert–Huang Transform with Empirical Mode Decomposition (HHT–EMD) for adaptive time–frequency feature extraction and a Gray Wolf Optimized Long Short-Term Memory (GWO–LSTM) classifier for intelligent fault detection and discrimination. The optimization of LSTM hyperparameters, including the number of hidden units, learning rate, dropout rate, and batch size, through GWO ensures enhanced generalization and robustness. Real-time validation performed on an OPAL-RT digital simulator demonstrates that the proposed scheme accurately identifies PV array and DC line faults under various irradiance levels, converter outages, and reconfiguration events. The model achieves 99.07% overall classification accuracy, dependability of 99.04%, and fault detection latency below 5&#xa0;ms, outperforming conventional ensemble and threshold-based protection schemes. These results confirm the proposed strategy’s capability to deliver reliable, fast, and adaptive protection for PV-fed DC microgrids operating under dynamic renewable conditions.</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

Resilient Protection Strategy for PV-Fed DC Microgrid with Immunity to Intermittency and Outage of PVDG Using GWO-Tuned LSTM

  • 期刊论文

  • 发表日期：

    2026年2月9日

查看全文
  • Vishwanath Tiwari,
  • C. S. Rajeshwari

摘要

The increasing integration of photovoltaic (PV) sources in DC microgrids introduces significant protection challenges due to power intermittency, converter switching dynamics, and frequent reconfiguration of distributed generation units. This paper presents a resilient hybrid protection framework combining the Hilbert–Huang Transform with Empirical Mode Decomposition (HHT–EMD) for adaptive time–frequency feature extraction and a Gray Wolf Optimized Long Short-Term Memory (GWO–LSTM) classifier for intelligent fault detection and discrimination. The optimization of LSTM hyperparameters, including the number of hidden units, learning rate, dropout rate, and batch size, through GWO ensures enhanced generalization and robustness. Real-time validation performed on an OPAL-RT digital simulator demonstrates that the proposed scheme accurately identifies PV array and DC line faults under various irradiance levels, converter outages, and reconfiguration events. The model achieves 99.07% overall classification accuracy, dependability of 99.04%, and fault detection latency below 5 ms, outperforming conventional ensemble and threshold-based protection schemes. These results confirm the proposed strategy’s capability to deliver reliable, fast, and adaptive protection for PV-fed DC microgrids operating under dynamic renewable conditions.