基于改进生成对抗网络的电压暂降事件类型辨识研究  

Research on Voltage Sag Event Type Identification Based on Improved Generative Adversarial Networks

作　　者：沙浩源 梅飞[2] 李丹奇 李轩[1] 张宸宇 史明明 郑建勇[1] SHA Haoyuan;MEI Fei;LI Danqi;LI Xuan;ZHANG Chenyu;SHI Mingming;ZHENG Jianyong(School of Electrical Engineering,Southeast University,Nanjing 210096,Jiangsu province,China;College of Energy and Electrical Engineering,Hohai University,Nanjing 211100,Jiangsu Province,China;State Grid Jiangsu Electric Power CO.,LTD.Research Institute,Nanjing 211103,Jiangsu Province,China)

机构地区：[1]东南大学电气工程学院,江苏省南京市210096 [2]河海大学能源与电气学院,江苏省南京市211100 [3]国网江苏省电力有限公司电力科学研究院,江苏省南京市211103

出　　处：《中国电机工程学报》2021年第22期7648-7659,共12页Proceedings of the CSEE

基　　金：国家重点研发计划项目(2018YFB1500800);智能电网保护和运行控制国家重点实验室项目(519054212)。

摘　　要：为缓解特征自提取模型对电压暂降样本数据量的依赖,提高模型的特征抓取能力,该文提出基于改进辅助分类生成对抗网络(auxiliary classifier generative adversarial networks,AC-GAN)的暂降事件类型辨识算法。首先,将暂降三相电压数据转换为基于空间矢量(space phasor model,SPM)的二维轨迹曲线,以此作为智能模型的输入。然后,对AC-GAN进行改进,通过在判别器内融合卷积注意力模块(convolutional block attention module,CBAM)来改善判断模型的特征自提取能力,从而提高整个AC-GAN网络的性能。利用所生成的与真实样本特性及分布一致的数据,来实现数据增强,以解决非平衡样本条件下特征学习不充分的问题。最后,利用江苏地区实际数据场景验证了所提算法在不同数据条件下准确而稳定的暂降类型辨识能力。In order to alleviate the dependence of feature Self Extraction Model on voltage sag sample data and improve the ability of feature capture, a voltage sag event type identification algorithm based on improved auxiliary classifier generative adversarial networks(AC-GAN) was proposed in this paper. Firstly, the sag three-phase voltage data was transformed into a two-dimensional trajectory curve based on space phasor model(SPM), and the SPM trajectory image was used as the input of the intelligent model. The AC-GAN was improved by fusing the convolutional block attention module(CBAM) in the discriminator to improve the feature self extraction ability of the judgment model, so as to improve the performance of the whole AC-GAN network. In order to solve the problem of insufficient feature learning under unbalanced samples, the generated data, which are consistent with the real sample characteristics and distribution, were used to enhance the data. Finally, the real data scenarios in Jiangsu Province were used to verify the accuracy and stability of the proposed algorithm in identifying the type of sag under different data conditions.

关 键 词：辅助分类生成对抗网络 空间矢量 卷积注意力机制 暂降事件 类型辨识 

分 类 号：TM73[电气工程—电力系统及自动化]