基于旋转分量和动态卷积的复杂电能质量扰动分类  

Classification of complex power quality disturbances based on rotational components and dynamic convolution

作　　者：何昊 董优丽 周寿湖 赵伟哲 李佳 HE Hao;DONG Youli;ZHOU Shouhu;ZHAO Weizhe;LI Jia(Electric Power Research Institute of State Grid Jiangxi Electric Power Co.,Ltd.,Nanchang 330096,China;Hubei Engineering Research Center of Safety Monitoring of New Energy and Power Grid Equipment,Hubei University of Technology,Wuhan 430068,China;State Grid Jiangxi Electric Power Co.,Ltd.,Nanchang 330096,China)

机构地区：[1]国网江西省电力有限公司电力科学研究院,江西南昌330096 [2]湖北工业大学新能源及电网装备安全监测湖北省工程研究中心,湖北武汉430068 [3]国网江西省电力有限公司,江西南昌330096

出　　处：《武汉大学学报（工学版）》2024年第11期1659-1668,共10页Engineering Journal of Wuhan University

基　　金：国家自然科学基金资助项目(编号:52077089);江西省电网公司科技项目(编号:20232ABC03A03)。

摘　　要：新能源发电功率的波动性及大量电力转换器、无功功率设备等的广泛应用,使得电网电能质量扰动波形复杂难辨,特别是所引发的频率偏差扰动、间谐波扰动严重影响了现有电能质量的分类。为此,提出了一种基于旋转坐标特征和动态卷积网络的分类方法。首先,通过dq变换将扰动信号变换到旋转坐标系,消除其正弦背景,获得具有辨识度的包括传统扰动、频率偏差扰动和间谐波扰动的畸变特征;接着,设计了一种多尺度特征提取网络,利用一维动态卷积模块的多尺度特性分别捕获原始信号和dq分量的短时特征和趋势特征;最后,将特征图合并,使用深度卷积模块进行特征提取,实现复杂扰动信号的多标签分类。实验结果表明,所提方法不仅能有效识别含频偏和间谐波的复合扰动,还能大幅提升对其他扰动信号的分类能力。The power fluctuation of new energy power generation and the widespread use of a large number of power converters and reactive power devices lead to complex and indistinguishable power quality disturbance waveforms in the power grid,especially the frequency deviation disturbances and inter-harmonic disturbances caused severely affects the classification of existing power quality.Therefore,a classification method based on rotational coordinate features and dynamic convolution networks is proposed.Firstly,the disturbance signals are transformed to a rotating coordinate system by dq transformation,which not only eliminates the sinusoidal background,but also obtains the distortion features of traditional disturbance,frequency deviation and interharmonic disturbances with identification.Then,a multi-scale feature extraction network is designed,and the multi-scale feature of one-dimensional dynamic convolution module is used to capture the short-time features and trend features of the original signal and the dq sequence components respectively.Finally,the feature maps are combined and the deep convolutional network is used for feature extraction to realize the multi-label classification of complex disturbance signals.The experimental results show that the proposed method can not only effectively identify the complex disturbances containing frequency deviation and inter-harmonics,but also greatly improve the classification ability of other disturbance signals.

关 键 词：电能质量扰动 动态卷积 dq变换 多标签分类 

分 类 号：TM761[电气工程—电力系统及自动化]