基于对抗性持续学习模型的输电线路部件缺陷分类

赵振兵; 蒋志钢; 熊静; 聂礼强; 吕雪纯

doi:10.11999/JEIT220200

基于对抗性持续学习模型的输电线路部件缺陷分类

doi: 10.11999/JEIT220200

赵振兵^{1, 2, 3, ,},
蒋志钢¹,
熊静¹,
聂礼强⁴,
吕雪纯¹

1.
华北电力大学电子与通信工程系保定 071003
2.
华北电力大学河北省电力物联网技术重点实验室保定 071003
3.
复杂能源系统智能计算教育部工程研究中心保定 071003
4.
山东大学计算机科学与技术学院青岛 266237

基金项目: 国家自然科学基金(61871182, U21A20486)，河北省自然科学基金(F2020502009, F2021502008, F2021502013)

详细信息

作者简介:
赵振兵：男，教授，研究方向为电力视觉技术

蒋志钢：男，硕士生，研究方向为电力设备缺陷识别

熊静：女，硕士生，研究方向为电力设备视觉检测

聂礼强：男，教授，研究方向为多媒体计算和信息检索

吕雪纯：女，硕士生，研究方向为电力设备缺陷识别

通讯作者:
赵振兵　zhaozhenbing@ncepu.edu.cn

中图分类号: TM726.3; TP391.4
计量
- 文章访问数: 1208
- HTML全文浏览量: 496
- PDF下载量: 118
- 被引次数: 0
出版历程
- 收稿日期: 2022-03-01
- 修回日期: 2022-05-26
- 录用日期: 2022-06-08
- 网络出版日期: 2022-06-13
- 刊出日期: 2022-11-14

Fault Classification of Transmission Line Components Based on the Adversarial Continual Learning Model

1.
Department of Electronic and Communication Engineering, North China Electric Power University, Baoding 071003, China
2.
Hebei Key Laboratory of Power Internet of Things Technology, North China Electric Power University, Baoding 071003, China
3.
Engineering Research Center of Intelligent Computing for Complex Energy Systems, Ministry of Education, Baoding 071003, China
4.
School of Computer Science and Technology, Shandong University, Qingdao 266237, China

Funds: The National Natural Science Foundation of China (61871182, U21A20486), The Natural Science Foundation of Hebei Province (F2020502009, F2021502008, F2021502013)

摘要

摘要: 输电线路金具巡检是电网安全态势感知中不可或缺的一部分，线路的定期巡检关系着电力系统是否能安稳运行。针对目前的输电线路部件缺陷分类模型无法处理现实情况中无限数据流的问题，该文提出一种基于对抗性持续学习的输电线路部件及其缺陷分类方法。将持续学习技术引入到输电线路部件缺陷分类任务中，使得分类模型在保证分类准确率的同时，可以从无限增长的数据流中不断学习新的分类任务，并且减少时间资源消耗。通过融入注意力机制，增强了模型对细微特征提取能力，解决了分类任务类间差异过小的问题，提高分类准确率。针对持续学习任务中的排序不可知性问题，提出基于离散度进行排序的方法，以实现持续学习分类模型的最优利用。最后，在CIFAR-100公共数据集和自建数据集上进行实验验证，并对模型的各种性能进行分析与比较。结果表明该文提出的方法实现了部件及其缺陷分类任务的可持续学习，缓解了灾难性遗忘的问题；融入注意力机制和使用L₃损失函数使分类准确率分别提高了1.43%和2.25%；实现了持续学习分类模型在已获取数据集上的最优利用，为电网安全态势感知打下了坚实的基础。
- 输电线路态势感知 /
- 缺陷分类 /
- 持续学习 /
- 注意力机制 /
- 排序不可知性
Abstract: The inspection of transmission line fittings is an indispensable part of power grid security situation awareness. Focusing on the fact that the current transmission line component defect classification model cannot handle the problem of unlimited data flow in real situations, a transmission line component and its defect classification method based on adversarial continuous learning is proposed. In this paper, continuous learning technology is introduced into the task of transmission line component defect classification, so that the classification model can continuously learn new classification tasks from the infinite growth of data stream while ensuring the classification accuracy, and reduce the consumption of time and resources. By integrating attention mechanism, the ability of the model to extract subtle features is enhanced, the problem of small difference between classification tasks is solved, and the classification accuracy is improved. Focusing on the problem of sorting unknowability in continual learning tasks, a method of sorting based on discrete degree is proposed to achieve the optimal utilization of continual learning classification model. Finally, experiments are carried out on CIFAR-100 public data set and self built data set, and various performances of the model are analyzed and compared. The results show that the proposed method realizes the sustainable learning of component and defect classification task, and alleviates the problem of catastrophic forgetting. The accuracy of classification is improved by 1.43% and 2.25% respectively by integrating attention mechanism and using L₃ loss function. The optimal utilization of continuous learning classification model is realized, which lays a solid foundation for power grid security situational awareness.
- Transmission line situation awareness /
- Defect classification /
- Continual learning /
- Attention mechanism /
- Ranking agnosticism

HTML全文

图 1 基于对抗性持续学习分类模型整体结构

下载: 全尺寸图片幻灯片

图 2 正常图像与缺陷图像对照

下载: 全尺寸图片幻灯片

图 3 注意力机制模块

下载: 全尺寸图片幻灯片

图 4 改进后ResNet-18网络整体结构

下载: 全尺寸图片幻灯片

图 5 自建数据集样例展示

下载: 全尺寸图片幻灯片

图 6 已学习任务准确率变化图

下载: 全尺寸图片幻灯片

表 1 数据种类及数量(张)

种类	数量	种类	数量
鸟巢	532	异物	102
正常均压环	142	损坏均压环	159
正常绝缘子	1417	掉串绝缘子	541
正常屏蔽环	391	锈蚀屏蔽环	66
正常防震锤	1734	变形防震锤	80

下载: 导出CSV

表 2 不同主干网络对照实验

主干网络	ACC(%)	时间(min)	参数量(MB)	BWT(%)
AlexNet	90.86	553.8	14.3	–0.36
ResNet-18	92.58	711.3	16.4	–0.43
ResNet-34	91.76	1358.0	28.6	–0.52
ResNet-50	90.56	2168.4	32.1	–0.61

下载: 导出CSV

表 3 模型各部分消融实验

S/D	P	ACC(%)	BWT(%)
√		54.07	–17.65
	√	47.70	0
√	√	91.75	–0.92

下载: 导出CSV

表 4 改进网络前后模型性能对比(自建数据集)

主干网络	ACC(%)	时间(min)	参数量(MB)	BWT(%)
ResNet-18	92.58	711.3	16.4	–0.43
ResNet-18-att	94.01	929.4	17.6	–0.20

下载: 导出CSV

表 5 改进网络前后模型性能对比(CIFAR-100)

主干网络	ACC(%)	时间(min)	参数量(MB)	BWT(%)
ResNet-18	71.51	1466.3	31.3	–0.78
ResNet-18-att	72.69	1854.4	35.8	–0.51

下载: 导出CSV

表 6 L₃损失函数消融实验

主干网络	L₃	ACC(%)	时间(min)	BWT(%)
ResNet-18		91.75	514.3	–0.92
ResNet-18	√	92.58	711.3	–0.43
ResNet-18-att		91.76	799.2	–0.16
ResNet-18-att	√	94.01	929.4	–0.20
AlexNet		90.62	471.6	–0.42
AlexNet	√	90.86	553.8	–0.36

下载: 导出CSV

表 7 8组对照实验准确率结果

实验编号	各任务准确率(%)
实验编号	鸟巢异物分类(1)	均压环分类(2)	绝缘子分类(3)	屏蔽环分类(4)	防震锤分类(5)
a	88.75	93.36	96.13	92.97	98.85
b	86.48	89.93	95.81	94.69	98.85
c	89.38	90.63	95.28	89.55	98.87
d	88.84	97.81	95.05	95.83	98.70
e	88.75	90.63	95.72	92.19	98.92
f	88.84	90.31	95.54	93.08	98.80
g	86.59	94.53	95.42	90.94	98.96
h	88.75	93.95	95.92	91.18	98.85
平均	88.30	92.64	95.61	92.55	98.85

下载: 导出CSV

表 8 8组实验结果

编号	ACC(%)	时间(min)	BWT(%)
a	94.01	929.4	–0.20
b	93.15	828.1	0.29
c	92.74	1001.0	–0.07
d	95.25	923.9	0.80
e	93.24	890.2	–0.01
f	93.31	764.7	0.28
g	93.29	831.3	0.06
h	93.73	837.1	–0.03

下载: 导出CSV

参考文献(20)

[1]	于群, 李浩, 屈玉清. 基于深度学习的电网安全态势感知[J]. 科学技术与工程, 2019, 19(35): 273–278. doi: 10.3969/j.issn.1671-1815.2019.35.041 YU Qun, LI Hao, and QU Yuqing. Security situational awareness of power grid based on deep learning[J]. Science Technology and Engineering, 2019, 19(35): 273–278. doi: 10.3969/j.issn.1671-1815.2019.35.041
[2]	刘权莹, 李俊娥, 倪明, 等. 电网信息物理系统态势感知: 现状与研究构想[J]. 电力系统自动化, 2019, 43(19): 9–21,51. doi: 10.7500/AEPS20181027002 LIU Quanying, LI Jun’e, NI Ming, et al. Situation awareness of grid cyber-physical system: Current status and research ideas[J]. Automation of Electric Power Systems, 2019, 43(19): 9–21,51. doi: 10.7500/AEPS20181027002
[3]	张姝, 王昊天, 董骁翀, 等. 基于深度学习的输电线路螺栓检测技术[J]. 电网技术, 2021, 45(7): 2821–2828. doi: 10.13335/j.1000-3673.pst.2020.1336 ZHANG Shu, WANG Haotian, DONG Xiaochong, et al. Bolt detection technology of transmission lines based on deep learning[J]. Power System Technology, 2021, 45(7): 2821–2828. doi: 10.13335/j.1000-3673.pst.2020.1336
[4]	刘思言, 王博, 高昆仑, 等. 基于R-FCN的航拍巡检图像目标检测方法[J]. 电力系统自动化, 2019, 43(13): 162–167. doi: 10.7500/AEPS20180921005 LIU Siyan, WANG Bo, GAO Kunlun, et al. Object detection method for aerial inspection image based on region-based fully convolutional network[J]. Automation of Electric Power Systems, 2019, 43(13): 162–167. doi: 10.7500/AEPS20180921005
[5]	赵振兵, 李延旭, 甄珍, 等. 结合KL散度和形状约束的Faster R-CNN典型金具检测方法[J]. 高电压技术, 2020, 46(9): 3018–3026. doi: 10.13336/j.1003-6520.hve.20200507023 ZHAO Zhenbing, LI Yanxu, ZHEN Zhen, et al. Typical fittings detection method with faster R-CNN combining KL divergence and shape constraints[J]. High Voltage Engineering, 2020, 46(9): 3018–3026. doi: 10.13336/j.1003-6520.hve.20200507023
[6]	赵振兵, 张薇, 戚银城, 等. 融合深度特征的输电线路金具缺陷因果分类方法[J]. 北京航空航天大学学报, 2021, 47(3): 461–468. doi: 10.13700/j.bh.1001-5965.2020.0456 ZHAO Zhenbing, ZHANG Wei, QI Yincheng, et al. Causal classification method of transmission lines fitting defect combined with deep features[J]. Journal of Beijing University of Aeronautics and Astronautics, 2021, 47(3): 461–468. doi: 10.13700/j.bh.1001-5965.2020.0456
[7]	李雪峰, 刘海莹, 刘高华, 等. 基于深度学习的输电线路销钉缺陷检测[J]. 电网技术, 2021, 45(8): 2988–2995. doi: 10.13335/j.1000-3673.pst.2020.1028 LI Xuefeng, LIU Haiying, LIU Gaohua, et al. Transmission line pin defect detection based on deep learning[J]. Power System Technology, 2021, 45(8): 2988–2995. doi: 10.13335/j.1000-3673.pst.2020.1028
[8]	缪希仁, 林志成, 江灏, 等. 基于深度卷积神经网络的输电线路防鸟刺部件识别与故障检测[J]. 电网技术, 2021, 45(1): 126–133. doi: 10.13335/j.1000-3673.pst.2019.1775 MIAO Xiren, LIN Zhicheng, JIANG Hao, et al. Fault detection of power tower anti-bird spurs based on deep convolutional neural network[J]. Power System Technology, 2021, 45(1): 126–133. doi: 10.13335/j.1000-3673.pst.2019.1775
[9]	李彩林, 张青华, 陈文贺, 等. 基于深度学习的绝缘子定向识别算法[J]. 电子与信息学报, 2020, 42(4): 1033–1040. doi: 10.11999/JEIT190350 LI Cailin, ZHANG Qinghua, CHEN Wenhe, et al. Insulator orientation detection based on deep learning[J]. Journal of Electronics &Information Technology, 2020, 42(4): 1033–1040. doi: 10.11999/JEIT190350
[10]	DE LANGE M, ALJUNDI R, MASANA M, et al. A continual learning survey: Defying forgetting in classification tasks[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2022, 44(7): 3366–3385. doi: 10.1109/TPAMI.2021.3057446
[11]	刘志颖, 缪希仁, 陈静, 等. 电力架空线路巡检可见光图像智能处理研究综述[J]. 电网技术, 2020, 44(3): 1057–1069. doi: 10.13335/j.1000-3673.pst.2019.0349 LIU Zhiying, MIAO Xiren, CHEN Jing, et al. Review of visible image intelligent processing for transmission line inspection[J]. Power System Technology, 2020, 44(3): 1057–1069. doi: 10.13335/j.1000-3673.pst.2019.0349
[12]	KIRKPATRICK J, PASCANU R, RABINOWITZ N, et al. Overcoming catastrophic forgetting in neural networks[J]. Proceedings of the National Academy of Sciences of the United States of America, 2017, 114(13): 3521–3526. doi: 10.1073/pnas.1611835114
[13]	ZENKE F, POOLE B, and GANGULI S. Continual learning through synaptic intelligence[C]. The 34th International Conference on Machine Learning, Sydney, Australia, 2017: 3987–3995.
[14]	LI Zhizhong and HOIEM D. Learning without forgetting[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2018, 40(12): 2935–2947. doi: 10.1109/TPAMI.2017.2773081
[15]	CHAUDHRY A, RANZATO M, ROHRBACH M, et al. Efficient lifelong learning with A-GEM[C]. The 7th International Conference on Learning Representations, New Orleans, USA, 2019.
[16]	LOPEZ-PAZ D and RANZATO M. Gradient episodic memory for continual learning[C]. The 31st International Conference on Neural Information Processing Systems, Long Beach, USA, 2017: 6470–6479.
[17]	EBRAHIMI S, MEIER F, CALANDRA R, et al. Adversarial continual learning[C]. The 16th European Conference on Computer Vision, Glasgow, UK, 2020: 386–402.
[18]	SERRÀ S, SURIS D, MIRON M, et al. Overcoming catastrophic forgetting with hard attention to the task[C]. The 35th International Conference on Machine Learning, Stockholm, Sweden, 2018: 4555–4564.
[19]	RUSU A A, RABINOWITZ N C, DESJARDINS G, et al. Progressive neural networks[EB/OL]. https://arxiv.org/abs/1606.04671, 2016.
[20]	WOO S, PARK J, LEE J Y, et al. CBAM: Convolutional block attention module[C]. The 15th European Conference on Computer Vision, Munich, Germany, 2018: 3–19.