引用本文: | 江军,李威,李波,马径坦,张潮海.电流频谱Hurst指数在串联电弧故障检测中的应用[J].控制理论与应用,2022,39(3):561~569.[点击复制] |
JIANG Jun,LI Wei,LI Bo,MA Jing-tan,ZHANG Chao-hai.Application of the hurst index of current frequency spectrum in series arc fault detection[J].Control Theory and Technology,2022,39(3):561~569.[点击复制] |
|
电流频谱Hurst指数在串联电弧故障检测中的应用 |
Application of the hurst index of current frequency spectrum in series arc fault detection |
摘要点击 1561 全文点击 633 投稿时间:2021-07-26 修订日期:2022-03-09 |
查看全文 查看/发表评论 下载PDF阅读器 |
DOI编号 10.7641/CTA.2021.10670 |
2022,39(3):561-569 |
中文关键词 串联电弧故障 负载建模 改进型Mayr模型 奇异值(SVD)分解 Hurst指数 BP神经网络 |
英文关键词 series arc fault load modeling improved Mayr model singular value decomposition Hurst index BP neural networks |
基金项目 江苏省自然科学基金项目(BK20211189), 中央高校基本科研业务费专项资金项目(NT2021012)资助. |
|
中文摘要 |
串联电弧故障因故障电流小、随机性强, 难以被普通保护电器所检测和切除. 本文基于已有的Mayr电弧模
型增加了弧柱电感、间隙电容和压控随机开关以模拟电弧高频特性, 从而构建了不同类别典型负载的精细化电路
模型. 实验证实所建负载模型和电弧模型的仿真数据较好拟合了相应的实验结果, 在此基础上将仿真波形重构为二
维矩阵, 提取矩阵奇异值之比作为时域特征; 提取02500 Hz内归一化谐波幅值波形的Hurst指数作为频域特征; 选
取隐层节点数为17的单隐层BP网络实现了串联电弧故障诊断, 总体准确率达到93.9%. 检测方案在STM32F767平台
上运行时间仅19.85 ms, 符合UL–1699中实时性指标要求. 本文提出的检测方案具有计算量相对较小、特征简洁、准
确率高等优势, 有助于串联电弧检测和有效辨识的工业应用. |
英文摘要 |
The series arc fault is impossible to be detected and removed by normal breakers for its stochasticity and
low fault current amplitude. An arc model was proposed in this paper, and it consisted by the classical Mayr model, the
accessional arc column inductance, the distributed capacitance between electrodes and a voltage-controlled random switch
to simulate the high frequency oscillation triggered by cycle of arcing and extinguishing. Thus six simulation circuits with
corresponding type of load were built precisely. These circuit models were calibrated with experiments and proved to
be reliable. Each simulation waveform was reshaped as a 2-D matrix which was decomposed to acquire singular values.
Later the ratios of neighboring singular values were calculated as the time domain feature. Then, the harmonic component
under 2500 Hz were normalized and applied to calculate Hurst index as the frequency domain feature. Finally, a BP neural
network with a 17 nodes hidden layer was selected to be trained to accomplish arc detection. The off-line test proved that
this network achieved a total accuracy of 93.9%. The whole detection program was tested on STM32F767 and reached a
running time of 19.85 ms, which matches the real-time requirement in standard UL–1699. The algorithm in this paper has
concise features with relatively low calculation, and it contributes to the application of series arc fault detection in industrial
applications. |
|
|
|
|
|