基于CBAM-CNN的直流线路雷击干扰与短路识别模型  

Identification Model of Lightning Interference and Short Circuit at HVDC Transmission Line Based on CBAM-CNN

作　　者：廖志伟[1] 庄竞 王博文 郑广昱 谢汛恺 LIAO Zhiwei;ZHUANG Jing;WANG Bowen;ZHENG Guangyu;XIE Xunkai(School of Electric Power,South China University of Technology,Guangzhou 510641,Guangdong,China)

机构地区：[1]华南理工大学电力学院,广东广州510641

出　　处：《电气传动》2023年第9期88-96,共9页Electric Drive

基　　金：国家自然科学基金(52077082)。

摘　　要：针对直流线路保护和故障测距易受非故障性雷击干扰、传统基于时域和频域特征构造的输电线路雷击干扰识别方法存在阈值难以整定和噪声鲁棒性较差的问题,提出使用深度学习方法实现雷击干扰与短路行波特征自动提取与分类,相模解耦和小波包分解后得到的电流、电压行波分量作为不同通道输入至一维卷积模块注意力模块卷积神经网络(CBAM-CNN)分类模型。通过仿真和算例分析验证了所提模型相比传统方法具有更高的识别正确率,CBAM能有效提升CNN分类模型的噪声鲁棒性,同时验证了4层小波包分解与所提CBAM-CNN模型的结合具有最佳的性能。Aiming at the problems that high voltage direct current(HVDC)transmission line protection and fault location are vulnerable to lightning interference,and the traditional lightning interference identification methods of transmission line based on time-domain and frequency-domain features exist the problems of difficult threshold setting and poor noise robustness,a deep learning method was proposed to extract the characteristics of lightning interference and short-circuit traveling wave and classify automatically.After phase mode decoupling and wavelet packet decomposition,the current and voltage traveling wave components were input into the one-dimensional convolutional block attention module convolutional neural network(CBAM-CNN)classification model as different channels.Through simulation and example analysis,it is verified that the proposed model shows higher recognition accuracy than the traditional methods,and the CBAM can effectively improve the noise robustness of CNN classification model.At the same time,it is verified that the combination of four-layer wavelet packet decomposition and the proposed CBAM-CNN model has the best performance.

关 键 词：卷积神经网络 卷积模块注意力模块 小波包分解 直流输电线路 雷击干扰 时频分析 

分 类 号：TM28[一般工业技术—材料科学与工程]