发电技术 ›› 2019, Vol. 40 ›› Issue (6): 548-554.DOI: 10.12096/j.2096-4528.pgt.19103

• 能源互联网 • 上一篇    下一篇

基于深度学习的小电流接地系统故障选线方法

张国栋1(),蒲海涛1(),刘凯2()   

  1. 1 山东科技大学电气信息系, 山东省 济南市 253500
    2 洛阳供电公司, 河南省 洛阳市 471000
  • 收稿日期:2019-07-06 出版日期:2019-12-30 发布日期:2019-12-31
  • 作者简介:张国栋(1982),男,硕士,讲师,主要研究方向为电力系统运行控制, 459532455@qq.com|蒲海涛(1979),男,博士,副教授,主要研究方向为电力系统自动化, kdpht@163.com|刘凯(1981),男,硕士,高级工程师,主要研究方向为电网规划设计, 51781628@qq.com
  • 基金资助:
    教育部产学合作协同育人计划项目(201702064021);山东科技大学济南校区教研项目(JNJG2017203)

Fault Line Selection Method of Small Current Grounding System Based on Deep Learning

Guodong ZHANG1(),Haitao PU1(),Kai LIU2()   

  1. 1 Department of Electrical and Information, Shandong University of Science and Technology, Jinan 253500, Shandong Province, China
    2 Luoyang Power Supply Company, Luoyang 471000, Henan Province, China
  • Received:2019-07-06 Published:2019-12-30 Online:2019-12-31
  • Supported by:
    Ministry of Education's Cooperative Education Program Project(201702064021);Research Project of Jinan Campus of Shandong University of Science and Technology(JNJG2017203)

摘要:

小电流接地系统的单相接地故障选线问题目前仍没有完全解决。为了提高单相接地故障选线成功率,提出一种基于深度学习网络的选线方法。首先,利用PSCAD搭建了中性点不接地系统仿真模型,通过设置每条线路在不同接地电阻下的故障,得到基于各出线零序电流幅值和相角的样本数据,并将样本数据分为训练集、验证集和测试集3部分。其次,基于Keras搭建了深度学习神经网络,利用训练集和验证集数据对该网络进行训练。最后,利用测试集数据对训练好的模型进行测试。结果表明,该方法具有建模简单、成功率高及选线不受过渡电阻影响的特点。

关键词: 小电流接地系统, 单相接地故障, 故障选线, 深度学习

Abstract:

At present, the single-phase grounding fault line selection problem of small current grounding system has not been completely solved. In order to improve the success rate of single-phase grounding fault line selection, a new method based on deep learning network was proposed. Firstly, the simulation model of neutral un-grounded system was built by PSCAD. By setting the fault of each line under different grounding resistors, the sample data based on the zero sequence current value and phase angle of each line was obtained out. The sample data was divided into three parts:training set, verification set, and test set. Secondly, a deep learning neural network was constructed based on Keras, and the network was trained using the training set and validation set. Finally, the trained model was tested The results show that the method has the characteristics of simple modeling, high success rate and no influence of transition resistance.

Key words: small current grounding system, singlephase grounding fault, fault line selection, deep learning