基于半监督学习的配网架空线局部放电诊断研究
Research on partial discharge diagnosis of distribution network overhead lines based on semi-supervised learning
金建华 1孙超 1肖睿 1邓军 2武雍烨3
作者信息
- 1. 国网浙江省电力有限公司杭州市钱塘区供电公司,浙江 杭州 310007
- 2. 重庆大学输配电装备及系统安全与新技术国家重点实验室,重庆 400000
- 3. 国网四川省电力公司成都供电公司,四川 成都 610000
- 折叠
摘要
针对配电网设备现场运维过程中数据标注率不足的问题,文中设计了一种基于半监督学习策略的混合框架模型.在保留电压序列关键信息的同时将一维信号转换成多特征图输入形式,通过部分标注数据标签信息和数据重构误差进行训练,并结合软投票法进行多特征决策融合.实验测试结果表明,在数据集标注率为 30%、60%、70%和 90%的条件下,平均识别准确率分别为91.296 0%、95.564 3%、96.726 3%和96.991 8%,相较基于有监督学习的ResNet、VGG等模型,半监督混合框架模型提高了约5%的准确率,为配网架空线路局部放电初期诊断提供了一种新的模型方法,能够提高架空线路的维护和管理水平.
Abstract
In response to the problem of insufficient data annotation rate during the on-site operation and maintenance of distribution network equipment,a hybrid framework model based on semi supervised learning strategy is designed in this paper.While retaining the key information of the voltage sequence,the one-dimensional signal is converted into a multi feature map input form,trained through partially annotated data label information and data reconstruction error,and combined with soft voting method for multi feature decision fusion.The experimental test results show that under the conditions of dataset annotation rates of 30%,60%,70% and 90%,the average recognition accuracy is 91.2960%,95.564 3%,96.726 3%,and 96.991 8%,respectively.Compared with models such as ResNet and VGG based on supervised learning,the semi supervised hybrid framework model improves the accuracy by about 5%,providing a new model method for early diagnosis of partial discharge in overhead transmission lines,it can improve the maintenance and management level of overhead lines.
关键词
配网架空线/半监督学习/局部放电/故障诊断/深度学习/信号处理Key words
distribution network overhead lines/semi-supervised learning/partial discharge/fault dia-gnosis/deep learning/signal processing引用本文复制引用
出版年
2025
NETL
NSTL
万方数据