Scholarlink

<p>This paper introduces a deep learning approach for detecting and localizing switch faults in photovoltaic (PV) systems, specifically targeting PV-fed cascaded H-bridge five-level inverters (CHB5LI). The proposed method is designed to identify single short-circuit faults and combinations of up to two open-circuit faults occurring concurrently across the inverter switches. The research utilizes a residual network (ResNet) architecture with residual connections to effectively identify and localize faults. By incorporating noise signals, the study addresses practical challenges in solar inverter operations and advances the methodology for detection of fault in PV systems. Extensive testing across 48 unique fault classes and one non-fault case demonstrated the model’s robustness, achieving an accuracy of 92% at −20 dB noise, approximately 94% at −10 dB and 0 dB, and around 95% at 10 dB and 20 dB. The model was trained using an NVIDIA A100 Graphics Processing Unit (GPU). This research highlights the development of a real-time fault detection system capable of operating under multiple modulation indices, ranging from 0.55 to 1, in the presence of both single and double switch faults. The results underscore the potential of proposed methodology to markedly improve upon the reliability and performance of renewable energy technologies, marking a progressive step in fault detection for solar energy systems.</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

Fault diagnosis and localization in photovoltaic-fed cascaded half-bridge five-level inverter using residual network architecture

  • 期刊论文

  • 发表日期：

    2025年6月5日

查看全文
  • Iqra Ashfaq,
  • Zaki Uddin,
  • Naeem Ul Islam,
  • Azhar ul Haq,
  • Muhammad Nouman Arif

摘要

This paper introduces a deep learning approach for detecting and localizing switch faults in photovoltaic (PV) systems, specifically targeting PV-fed cascaded H-bridge five-level inverters (CHB5LI). The proposed method is designed to identify single short-circuit faults and combinations of up to two open-circuit faults occurring concurrently across the inverter switches. The research utilizes a residual network (ResNet) architecture with residual connections to effectively identify and localize faults. By incorporating noise signals, the study addresses practical challenges in solar inverter operations and advances the methodology for detection of fault in PV systems. Extensive testing across 48 unique fault classes and one non-fault case demonstrated the model’s robustness, achieving an accuracy of 92% at −20 dB noise, approximately 94% at −10 dB and 0 dB, and around 95% at 10 dB and 20 dB. The model was trained using an NVIDIA A100 Graphics Processing Unit (GPU). This research highlights the development of a real-time fault detection system capable of operating under multiple modulation indices, ranging from 0.55 to 1, in the presence of both single and double switch faults. The results underscore the potential of proposed methodology to markedly improve upon the reliability and performance of renewable energy technologies, marking a progressive step in fault detection for solar energy systems.