湖南电力 ›› 2024, Vol. 44 ›› Issue (1): 32-37.doi: 10.3969/j.issn.1008-0198.2024.01.005

• 研究与试验 • 上一篇    下一篇

基于特征混叠分析与贝叶斯-随机森林的触电辨识方法研究

吴聪1, 刘谋海1, 周灿1, 黄瑞1, 仝海昕2, 鲁进2   

  1. 1.国网湖南省电力有限公司,湖南 长沙 410004;
    2.长沙理工大学电气与信息工程学院,湖南 长沙 410114
  • 收稿日期:2023-11-01 修回日期:2023-11-20 出版日期:2024-02-25 发布日期:2024-03-11
  • 通信作者: 刘谋海(1990),男,硕士,工程师,研究方向为智能电网分析。
  • 作者简介:吴聪(1993),男,博士,工程师,研究方向为智能电网分析。
  • 基金资助:
    国家电网有限公司科技项目(5216A8220011)

Research on Electric Shock Identification Based on Feature Overlap Analysis and Bayes-Random Forest

WU Cong1, LIU Mouhai1, ZHOU Can1, HUANG Rui1, TONG Haixin2, LU Jin2   

  1. 1. State Grid Hunan Electric Power Company Limited, Changsha 410004, China;
    2. School of Electrical and Information Engineering, Changsha University of Science and Technology, Changsha 410114, China
  • Received:2023-11-01 Revised:2023-11-20 Online:2024-02-25 Published:2024-03-11

摘要: 针对低压配电系统中特征混叠导致生物触电辨识困难的问题,提出基于贝叶斯-随机森林的生物触电辨识方法。首先,对包括非生命体对照组在内五种触电类型的电流波形进行特征分析,研究相同触电特征在不同类别中的混叠情况。其次,通过贝叶斯优化算法对随机森林模型超参数进行寻优,采用bagging方法、使用真实样本对随机森林进行数据拟合,描述了随机森林集成辨识机制的运作方式,以及方法的具体实施流程。再次,在136 802个包含五种类型的测试样本中进行方法验证,总体准确率为96.276%。最后,与现有方法进行对比,验证了所提方法在触电辨识方面的优越性。

关键词: 生物触电辨识, 随机森林, 特征混叠

Abstract: In response to the difficulty in identifying biological electric shocks caused by feature aliasing in low-voltage distribution systems, a Bayesian-random forest based method for identifying electric shock caused by characteristic overlap in low-voltage distribution systems is proposed. Firstly, the current waveforms of five types of electric shocks, including a non-living organism control group, are analyzed to study the overlap of the same shock characteristics in different categories. Secondly, the Bayesian optimization algorithm is used to optimize the hyper-parameters of the random forest model. The random forest is fitted with real samples using bagging, describing the operation of the ensemble identification mechanism and the specific implementation process of the method. Furthermore, the method is validated on 136 802 test samples containing five types, achieving an overall accuracy of 96.276%. Finally, a comparison with existing methods demonstrates the superiority of the proposed method in electric shock identification.

Key words: biological electric shock identification, random forest, feature overlap

中图分类号: