王坚俊,孙林涛,刘昌标,刘江明,周国伟,郭创新
WANG Jianjun,SUN Lintao,LIU Changbiao,LIU Jiangming,ZHOU Guowei,GUO Chuangxin

• 1:国网浙江省电力有限公司
• 2:浙江大学电气工程学院

摘要(Abstract)：

绝缘套管是变压器箱外主要的绝缘装置，对变压器安全稳定运行具有重要意义。为提高绝缘套管故障诊断效率，提出了一种基于改进Mask R-CNN的变压器绝缘套管故障智能诊断方法。在运维过程中收集含绝缘套管的红外图像并标注，完成数据集的初步搭建。为缓解平衡正负样本不均衡，使用旋转缩放、裁剪平移等方法扩增负样本数量，并引入了GHM Loss函数。采用MobileNetv3作为Mask R-CNN的主干网络，以满足实时检测的要求。对比实验表明，改进后的Mask R-CNN检测速度达到216 ms每帧，故障诊断率为89.72%，误报率为6.78%，可以准确实现变压器绝缘套管故障智能诊断，对自动巡检和建设智能电站有一定应用价值。
As a major insulation device outside the transformer box,insulating bushing is of great significance to the safe and stable operation of the transformer. In order to improve fault diagnosis efficiency of insulating bushing,an intelligent diagnosis method of transformer insulating bushing fault based on improved Mask R-CNN is proposed in this paper. Firstly,the infrared images including those of insulating bushings are collected and labeled,and a preliminary data set is established in the process of operation and maintenance. In order to remedy the imbalance between positive and negative samples,negative samples are increased by rotating,scaling,clipping and translation,and the GHM loss function is introduced. MobileNetv3 is used as the backbone network of mask R-CNN to meet the requirements of real-time detection. The comparative experiments show that the improved Mask R-CNN acquires a detection speed as high as 216 ms per frame,a fault diagnosis rate of 89.72% and a misreport rate of 6.78%;it can accurately realize the intelligent fault diagnosis of transformer insulating bushings and can be used for automatic inspection and construction of intelligent power stations.

关键词(KeyWords)： 变压器绝缘套管;红外图像;改进Mask R-CNN;故障诊断
transformer insulating bushing;infrared image;improved Mask R-CNN;fault diagnosis

Abstract:

Keywords:

基金项目(Foundation): 国网浙江省电力有限公司科技项目（5211MR20004T）

作者(Author): 王坚俊,孙林涛,刘昌标,刘江明,周国伟,郭创新
WANG Jianjun,SUN Lintao,LIU Changbiao,LIU Jiangming,ZHOU Guowei,GUO Chuangxin

参考文献(References)：

• [1] TENBOHLEN S,JAGERS J,BASTOS G,et al.Transformer reliability survey[R]. Paris:CIGRE Working Group,2015.
• [2]吴俊锋.局部放电信号经油纸绝缘套管向外传播特性及检测方法研究[D].重庆：重庆大学，2016.
• [3]李维江.基于介电频域特性的套管绝缘状态评估研究[D].大连：大连理工大学，2015.
• [4]张大宁，白帆，牛朝滨，等.不同受潮类型下油纸绝缘套管的频域介电谱特性[J].中国电机工程学报，2018,38(16):4942-4950.
• [5]魏元，杨华，程正东，等.红外小目标检测的滞后阈值分割法[J].激光与红外，2020,50(1):111-116.
• [6]唐启永，徐凯.基于小波变换和区域分割的红外可见光彩色图像融合算法[J].光学技术，2015,41(1):68-71.
• [7]朱文静，陈华，李林，等.基于红外热成像边缘检测算法的小麦叶锈病分级研究[J].农业机械学报，2019,50(4):36-41.
• [8]陈同凡，刘云鹏，裴少通.基于改进YOLOv3的电力设备红外诊断方法[J].广东电力，2021,34(6):21-29.
• [9]李文璞，谢可，廖逍，等.基于Faster RCNN变电设备红外图像缺陷识别方法[J].南方电网技术，2019,13(12):79-84.
• [10]刘国特，伍伟权，郭芳，等.基于改进级联Gentle Adaboost分类器的支柱绝缘子红外图像AI识别[J/OL].高电压技术：1-9[2022-01-16].DOI:10.13336/j.1003-6520.hve.20210231.
• [11]李文璞，毛颖科，廖逍，等.基于旋转目标检测的变电设备红外图像电压致热型缺陷智能诊断方法[J].高电压技术，2021,47(9):3246-3253.
• [12] HE K,GKIOXARI G,DOLLáR P，等. Mask R-CNN[J].arXiv:1703.06870[cs],2018.
• [13]任之俊，蔺素珍，李大威，等.基于改进特征金字塔的Mask R-CNN目标检测方法[J].激光与光电子学进展，2019,56(4):174-179.
• [14]何文其，李剑，赵文浩.基于改进FasterRCNN的绝缘子异常检测[J].浙江电力，2021,40(8):40-46.
• [15]蒋梦莹，林小竹，柯岩.基于优化分类的数据增广方法[J].计算机工程与设计，2018,39(11):3559-3563.
• [16]贺双龙，杨斌.基于局部结构和视觉显著特征的红外和可见光图像泊松重构融合算法[J].南华大学学报（自然科学版），2020,34(5):62-70.
• [17]杨智宏，贺石中，冯伟，等.基于Mask R-CNN网络的磨损颗粒智能识别与应用[J].摩擦学学报，2021,41(1):105-114.
• [18]邵伟平，王兴，曹昭睿，等.基于MobileNet与YOLOv3的轻量化卷积神经网络设计[J].计算机应用，2020,40(S1):8-13.
• [19] LIN T Y,GOYAL P,GIRSHICK R,et al.Focal loss for dense object detection[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence,2020,42(2):318-327.
• [20] LI B,LIU Y,WANG X.Gradient harmonized single-stage detector[J].Proceedings of the AAAI Conference on Artificial Intelligence,2019,33(1):8577-8584.
• [21] LI M,SOLTANOLKOTABI M,OYMAK S. Gradient descent with early stopping is provably robust to label noise for overparameterized neural networks[C]//Proceedings of the Twenty Third International Conference on Artificial Intelligence and Statistics,PMLR,2020:4313-4324.