摘要
为发挥深度学习算法特征自学习及其在图像处理领域的优势,避免当前小电流接地系统单相接地故障选线中人工提取故障特征信息缺失的问题,提出了一种通过生成故障电流全信息空间域图像,再利用图像识别与分类算法实现故障选线的方法。所提方法首先使用三相电流构建三维空间域图像,并分别在3个平面上进行投影得到多幅投影图像,解决了获取二维图像的关键问题;然后对投影图像进行二次像素级图像融合,得到了相应的RGB彩色图像,最后使用深度学习算法对图像进行识别与分类从而实现故障选线。将所提方法与已有方法的故障选线结果进行对比,结果表明所提方法在多种因素影响下,均不损失故障信息,图像故障特征更明显、分类准确率更高,且具有抗噪声能力,证明其用于小电流接地系统单相接地故障选线具有可行性。
In order to take full use of feature self-learning advantage of deep learning algorithm and its advantages in the field of image processing,and to avoid the information loss problem of manual fault feature extraction in single-phase grounding fault line selection for small current grounding system,a method is proposed by generating a full-information space domain image of fault current and then using image recognition and classification algorithm to select the fault line.Firstly,three-phase current is used to construct a three-dimensional spatial domain image,which is respectively project on three planes to obtain multiple projection images,so the key problem of acquiring two-dimensional images is solved.Then,the secondary pixellevel images fusion of the projected images is carried out,by which the corresponding RGB color image is obtained.Finally,the deep learning algorithm is used to identify and classify the image to achieve fault line selection.The comparison of fault line selection results between the proposed method and the existing methods show that under the influence of various factors,the proposed method has no loss of fault information,more obvious image fault characteristics,higher classification accuracy and anti-noise ability,which proves its feasibility in single-phase grounding fault line selection for small current grounding system.
作者
程文傲
徐明
高金峰
CHENG Wenao;XU Ming;GAO Jinfeng(School of Electrical Engineering,Zhengzhou University,Zhengzhou 450001,China)
出处
《电力自动化设备》
EI
CSCD
北大核心
2021年第7期97-103,共7页
Electric Power Automation Equipment
基金
国家自然科学基金资助项目(51307152)。
关键词
小电流接地系统
故障选线
空间域
图像融合
深度学习
small current grounding system
fault line selection
spatial domain
image fusion
deep learning
