Scholarlink

<p>Series arc faults in low-voltage residential circuits represent a significant fire hazard, yet they remain difficult to detect due to fault currents typically falling below the tripping thresholds of conventional protection devices. Existing detection methods often suffer from false alarms caused by non-fault arcs, alongside excessive computational complexity and deployment costs. To address these challenges, a real-time arc fault detection method based on a dual sliding window and support vector machine (DSW-SVM) framework is proposed. The method employs an event-driven mechanism in which a fluctuation detection window continuously monitors voltage signals and activates an arc detection window only upon identifying abnormal fluctuations. This structure effectively eliminates unnecessary computations and enhances detection responsiveness. Feature extraction is performed on the activated segments, incorporating three conventional time-domain features and a novel metric, zero duration time (ZDT), which is designed to improve the differentiation between fault and non-fault arcs. The sampling window length is optimized using the NSGA-II, balancing classification accuracy with detection latency. A linear SVM is then used to evaluate the severity of detected faults. Experimental results, conducted at a 10&#xa0;kHz sampling rate, demonstrate that the proposed method achieves 99.00% classification accuracy with only 570 sampling points and an average detection time of 3.19 ms when deployed on a Raspberry Pi. These results confirm that the DSW-SVM approach is both accurate and computationally efficient, rendering it highly suitable for real-time arc fault diagnostics in embedded 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

Series arc fault detection using DSW-SVM approach

  • 期刊论文

  • 发表日期：

    2026年2月26日

查看全文
  • Yuan Jiang,
  • Xianghua Meng,
  • Suliang Ma,
  • Qing Li

摘要

Series arc faults in low-voltage residential circuits represent a significant fire hazard, yet they remain difficult to detect due to fault currents typically falling below the tripping thresholds of conventional protection devices. Existing detection methods often suffer from false alarms caused by non-fault arcs, alongside excessive computational complexity and deployment costs. To address these challenges, a real-time arc fault detection method based on a dual sliding window and support vector machine (DSW-SVM) framework is proposed. The method employs an event-driven mechanism in which a fluctuation detection window continuously monitors voltage signals and activates an arc detection window only upon identifying abnormal fluctuations. This structure effectively eliminates unnecessary computations and enhances detection responsiveness. Feature extraction is performed on the activated segments, incorporating three conventional time-domain features and a novel metric, zero duration time (ZDT), which is designed to improve the differentiation between fault and non-fault arcs. The sampling window length is optimized using the NSGA-II, balancing classification accuracy with detection latency. A linear SVM is then used to evaluate the severity of detected faults. Experimental results, conducted at a 10 kHz sampling rate, demonstrate that the proposed method achieves 99.00% classification accuracy with only 570 sampling points and an average detection time of 3.19 ms when deployed on a Raspberry Pi. These results confirm that the DSW-SVM approach is both accurate and computationally efficient, rendering it highly suitable for real-time arc fault diagnostics in embedded systems.