Scholarlink

To address DC arc faults in photovoltaic systems—a major cause of electrical fires—this study proposes a multi-scale feature fusion method. A UL1699B-compliant experimental platform was built to simulate arc faults under varying illumination, electrode spacing (0–5 mm), and arc velocities (0–5 mm/s), generating a 45 k-sample fault database. Wavelet decomposition (db4, 5 scales) extracted 126 features from current signals. Random Forest selected the top 5 discriminative features (e.g., power spectrum entropy, wavelet Level_1 mean), which were fused via Kernel PCA into a diagnostic indicator. The XGBoost model achieved 99.93% fault detection accuracy, outperforming single-scale methods by 4.86%. This method enables full-time arc diagnosis, resolves threshold adaptability issues, and reduces feature redundancy by 95.2%.

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

Arc Fault Diagnosis Method of Photovoltaic System Based on Multi-scale Feature Fusion

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Jiawei Cai,
  • Chang Hong Weng,
  • Long Yang Zhu,
  • Chuan Jie Lin,
  • Feng Cao

摘要

To address DC arc faults in photovoltaic systems—a major cause of electrical fires—this study proposes a multi-scale feature fusion method. A UL1699B-compliant experimental platform was built to simulate arc faults under varying illumination, electrode spacing (0–5 mm), and arc velocities (0–5 mm/s), generating a 45 k-sample fault database. Wavelet decomposition (db4, 5 scales) extracted 126 features from current signals. Random Forest selected the top 5 discriminative features (e.g., power spectrum entropy, wavelet Level_1 mean), which were fused via Kernel PCA into a diagnostic indicator. The XGBoost model achieved 99.93% fault detection accuracy, outperforming single-scale methods by 4.86%. This method enables full-time arc diagnosis, resolves threshold adaptability issues, and reduces feature redundancy by 95.2%.