摘要
针对生弧材料弱化故障电弧特征和监督式学习方法依赖于大量标记数据的问题,提出了包络微分算子增强故障电弧特征,并采用半监督学习模型融合该特征辨识直流故障电弧和生弧材料的方法。首先依据UL 1699B搭建直流故障电弧实验平台,采集了铝、黄铜、不锈钢、石墨铸铁以及球墨铸铁常见电力设备材料下的故障电弧信号,再采用包络微分算子增强电弧小波特征,有效提升了多生弧材料条件下的故障电弧特征显著性。然后通过K⁃Means算法进行离群点检测,有效降低了同种生弧材料的特征波动。最后通过采用半监督学习算法MixMatch改进长短期记忆网络(long short⁃term memory,LSTM)模型,实验结果验证了该模型可在有限故障数据条件下获得比现有监督式学习模型更高的故障电弧和生弧材料辨识性能,为直流故障电弧精准运维提供了可行的技术手段。
Aiming at the problems that the arc generation material weakens arc fault features and the supervised learning method depends on a large number of labeled data,the method of envelope differential operator to enhance the arc fault feature is proposed in this paper.Meanwhile,a semi⁃supervised learning model is used to fuse the fea⁃tures to identify DC arc faults and arc generation material.Firstly,the DC arc fault experimental platform is set up in accordance with UL 1699B standard.Arc fault signals are sampled with common power equipment materials includ⁃ing aluminum,brass,stainless steel,graphite cast iron and nodular cast iron.Then,an envelope differential operator is used to enhance the arc fault wavelet feature,which effectively improves the significance of arc fault feature under the condition of multiple arc generation materials.After that,the K⁃Means algorithm is applied to detect outliers,which effectively reduces the fluctuation of arc fault features with the same arc generation material.Finally,the semi⁃supervised learning algorithm MixMatch is used to improve the long short⁃term memory(LSTM)model.The experi⁃mental results prove that the model can obtain higher identification performance of arc fault and arc generation mate⁃rial than existing supervised learning models in the case of limited fault data.This research provides a feasible techni⁃cal means for accurate maintenance of DC arc faults.
作者
陈思磊
王源丰
孟羽
高佳宇
伍李阳
CHEN Silei;WANG Yuanfeng;MENG Yu;GAO Jiayu;WU Liyang(School of Electrical Engneering,Xi’an University of Technology,Xi’an 710048,China;The State Key Laboratory of Electrical Insulation and Power Equipment,Xi’an Jiaotong University,Xi’an 710049,China)
出处
《高压电器》
CAS
CSCD
北大核心
2023年第7期145-155,共11页
High Voltage Apparatus
基金
陕西省自然科学基础研究计划(2021JQ⁃476)。
关键词
直流故障电弧
生弧材料
特征增强
离群点检测
半监督学习
DC arc fault
arc generation material
feature enhancement
outlier detection
semi⁃supervised learning
