| 引用本文: | 杨耿杰,王 康,高 伟.基于相空间重构和迁移学习的配电网高阻接地故障检测[J].电力系统保护与控制,2022,50(13):151-162.[点击复制] |
| YANG Gengjie,WANG Kang,GAO Wei.High impedance fault detection in a distribution network based on phase spacereconstruction and transfer learning[J].Power System Protection and Control,2022,50(13):151-162[点击复制] |
|
| 摘要: |
| 配电网中高阻接地故障(High Impedance Fault, HIF)时常发生，其故障特征微弱而难以检测，严重情况下可能导致火灾或人身事故。提出了一种基于相空间重构和迁移学习的故障识别方法，实现对谐振接地系统中HIF的辨识。首先，使用基于综合策略的小波阈值降噪方法对零序电流信号进行处理，以降低噪声的影响。随后，对降噪后的仿真信号及实测信号进行相空间重构，获取重构轨迹图，以此作为故障识别的特征量。最后，在辨识模型构建上，先使用仿真信号的重构轨迹图训练GoogLeNet模型，再使用实测信号对模型进行微调，实现迁移学习。所提算法的优点是使用相空间重构进行了信号转换，故障信号与干扰信号的重构轨迹图差异明显，且实测信号与仿真信号的重构轨迹图相似度较高。在进行迁移学习后，实现了对实测小样本数据较为准确的检测。实验结果表明，无论是故障实测数据还是故障仿真数据，识别准确率均达到95%以上。此外，在强噪声干扰、采样数据点缺失及故障回路间歇性导通情况下，所提算法也取得了较好的结果。 |
| 关键词: 配电网 高阻接地故障 相空间重构 小波阈值降噪 迁移学习 |
| DOI：DOI: 10.19783/j.cnki.pspc.211282 |
| 投稿时间：2021-09-17修订日期：2022-03-04 |
| 基金项目:福建省自然科学基金项目资助(2021J01633) |
|
| High impedance fault detection in a distribution network based on phase spacereconstruction and transfer learning |
| YANG Gengjie,WANG Kang,GAO Wei |
| (College of Electrical Engineering and Automation, Fuzhou University, Fuzhou 350116, China) |
| Abstract: |
| High impedance faults (HIFs) occur frequently in a distribution network, and their fault characteristics are weak and difficult to detect. In serious cases, they may lead to fires or accidents. A fault identification method based on phase space reconstruction and transfer learning is proposed to identify an HIF in a resonant grounding system. First, the wavelet threshold denoising method based on a comprehensive strategy is used to process the zero sequence current signal to reduce the influence of noise. Then, the simulated signal and the measured signal after noise reduction are reconstructed in phase space, and the reconstructed trajectory is obtained as the characteristic quantity of fault identification. Finally, in the construction of an identification model, the reconstructed trajectories of simulation signals are investigated to train a GoogLeNet model, and then the measured signals are adopted to fine tune the model to realize transfer learning. The advantages of the proposed algorithm are that the phase space reconstruction is used for signal conversion, the reconstructed trajectories of fault signal and interference signal are obviously different, and the reconstructed trajectories of measured signal and simulated signal are highly similar; after the transfer learning, more accurate detection of the measured small sample data is realized. The experimental results show that the recognition accuracy of both fault measured data and fault simulation data is more than 95%. The proposed algorithm also achieves good results in the case of strong noise interference, missing sampling data points and intermittent conduction of the fault circuit.
This work is supported by the Natural Science Foundation of Fujian Province (No. 2021J01633). |
| Key words: distribution network high impedance fault phase space reconstruction wavelet threshold denoising transfer learning |
