基于通道选择多尺度融合深度残差网络的电能质量扰动识别  

Power quality disturbance identification of multi-scale fusion depth residual network based on channel selection

作　　者：刘伟[1] 王凯 LIU Wei;WANG Kai(College of Electrical and Information Engineering,Northeast Petroleum University,Daqing,Heilongjiang 163318)

机构地区：[1]东北石油大学电气信息工程学院,黑龙江大庆163318

出　　处：《电气技术》2023年第5期11-15,22,共6页Electrical Engineering

基　　金：国家自然科学基金项目(No.N11372071);教育部产学合作协同育人项目(No.201902241011)。

摘　　要：针对传统电能质量扰动分类方法人工特征选择困难、准确率低的缺点,在传统卷积网络的基础上,借鉴Inception、残差的思想,结合混合池化和高效通道注意力机制,提出一种基于通道选择多尺度融合深度残差网络(CSSF-ResNet)的电能质量扰动识别方法。使用多尺度卷积提取不同尺度的特征,将全局混合池化与高效通道注意力机制相结合,在通道维度进行特征筛选,挖掘有效特征,并引入残差连接,构成CSSF-ResNet。仿真结果表明,所提方法具有分类准确率高、噪声鲁棒性强等优点。Aiming at the shortcomings of traditional power quality disturbance classification methods that are difficult to manually select features and low accuracy,on the basis of traditional convolutional networks,drawing on the ideas of inception and residuals,combined with hybrid pooling and efficient channel attention mechanism,a power quality disturbance identification method based on channel selection multi-scale fusion deep residual network(CSSF-ResNet)is proposed.Multi-scale convolution is used to extract features of different scales,and global mixed pooling is combined with efficient channel attention mechanism.Feature screening is carried out in the channel dimension.Effective features are mined,and residual connections are introduced to form a CSSF-ResNet.The simulation results show that the proposed method has the advantages of high classification accuracy and strong noise robustness.

关 键 词：电能质量扰动 多尺度卷积 残差 通道注意力 全局混合池化 

分 类 号：TM711[电气工程—电力系统及自动化] TP183[自动化与计算机技术—控制理论与控制工程]