基于特征迁移和深度学习的配电网故障定位  

Fault Location in Distribution Networks Based on Feature Transfer and Deep Learning

作　　者：齐振兴 张倩[1,3] 丁津津 李国丽 QI Zhen-xing;ZHANG Qian;DING Jin-jin;LI Guo-li(School of Electrical Engineering and Automation,Anhui University,Hefei 230601,China;Collaborative Innovation Center of Industrial Energy-saving and Power Quality Control,Anhui University,Hefei 230601,China;Engineering Research Center of Power Quality,Ministry of Education,Anhui University,Hefei 230601,China;State Grid Anhui Electric Power Co.,Ltd.,Electric Power Research Institute,Hefei 230601,China;Anhui Industrial Power Saving and Electricity Safety Laboratory(Anhui University),Hefei 230601,China)

机构地区：[1]安徽大学电气工程及自动化学院,合肥230601 [2]安徽大学工业节电与电能质量控制协同创新中心,合肥230601 [3]教育部电能质量工程研究中心(安徽大学),合肥230601 [4]国网安徽省电力有限公司科学研究院,合肥230601 [5]工业节电与用电安全安徽省重点实验室(安徽大学),合肥230601

出　　处：《科学技术与工程》2022年第33期14752-14758,共7页Science Technology and Engineering

基　　金：国家自然科学基金(52077001)。

摘　　要：故障诊断对电力系统的稳定运行至关重要。当配电网的拓扑结构发生较大变化时,难以获取大量带有标签的暂态数据,导致传统的故障预测模型精度难以提高。针对此问题,提出一种将特征迁移和深度学习相结合的配电网故障诊断新方法。首先,采集配电网不同线路的零序电流构造故障特征集;其次,引入加权半监督迁移成分分析(semi-supervised transfer component analysis,SSTCA)方法,利用混合核函数将不同拓扑结构下的特征样本映射到同一特征空间中,缩小数据间的分布差异性;最后,将映射后的源域样本输入卷积神经网络中进行分类训练,并测试映射后的目标域样本。通过Simulink仿真表明,在改变配电网拓扑结构的新场景下,所提的特征迁移方法与其他方法相比,对目标域故障定位精度最高,达到98%以上。Fault diagnosis is crucial to the stable operation of power systems.When the topology of the distribution network changes significantly,it is difficult to obtain a large amount of transient data with labels,which makes it difficult to improve the accuracy of traditional fault prediction models.To address this problem,a new method of distribution network fault diagnosis that combines feature transfer and deep learning was proposed.Firstly,the zero-sequence currents of different lines in the distribution network were collected to construct fault feature sets.Secondly,a weighted semi-supervised transfer component analysis(SSTCA)method was introduced to map the feature samples under different topologies into the same feature space by using the hybrid kernel function,so as to reduce the distribution difference between the data.Finally,the mapped source domain samples were input to the convolutional neural network for classification training,and the mapped target domain samples were tested.The simulation shows that the proposed feature transfer method has the highest target domain fault recognition accuracy of more than 98%compared with other methods under the new scenario of changing the distribution network topology.

关 键 词：配电网 故障诊断 特征迁移 半监督迁移成分分析 卷积神经网络 

分 类 号：TM773[电气工程—电力系统及自动化]