《计算机应用》唯一官方网站 ›› 2022, Vol. 42 ›› Issue (4): 1235-1243.DOI: 10.11772/j.issn.1001-9081.2021071182
• CCF第36届中国计算机应用大会 (CCF NCCA 2021) • 上一篇
刘晶1,2,3, 董志红1, 张喆语1, 孙志刚4, 季海鹏2,3,5()
收稿日期:
2021-07-08
修回日期:
2021-09-03
接受日期:
2021-09-06
发布日期:
2021-09-18
出版日期:
2022-04-10
通讯作者:
季海鹏
作者简介:
刘晶(1979—),女,内蒙古包头人,研究员,博士,CCF高级会员,主要研究方向:工业人工智能基金资助:
Jing LIU1,2,3, Zhihong DONG1, Zheyu ZHANG1, Zhigang SUN4, Haipeng JI2,3,5()
Received:
2021-07-08
Revised:
2021-09-03
Accepted:
2021-09-06
Online:
2021-09-18
Published:
2022-04-10
Contact:
Haipeng JI
About author:
LIU Jing, born in 1979, Ph. D., research fellow. Her research interests include industrial artificial intelligence.Supported by:
摘要:
针对工业物联网(IIOT)新增数据量大、工厂子端数据量不均衡的问题,提出了一种基于联邦增量学习的IIOT数据共享方法(FIL-IIOT)。首先,将行业联合模型下发到工厂子端作为本地初始模型;然后,提出联邦优选子端算法来动态调整参与子集;最后,通过联邦增量学习算法计算出工厂子端的增量加权,从而使新增状态数据与原行业联合模型快速融合。实验结果表明,在美国凯斯西储大学(CWRU)轴承故障数据集上,所提FIL-IIOT使轴承故障诊断精度达到93.15%,比联邦均值(FedAvg)算法和无增量公式的FIL-IIOT(FIL-IIOT-NI)方法分别提高了6.18个百分点和2.59个百分点,满足了基于工业增量数据的行业联合模型持续优化的需求。
中图分类号:
刘晶, 董志红, 张喆语, 孙志刚, 季海鹏. 基于联邦增量学习的工业物联网数据共享方法[J]. 计算机应用, 2022, 42(4): 1235-1243.
Jing LIU, Zhihong DONG, Zheyu ZHANG, Zhigang SUN, Haipeng JI. Data sharing method of industrial internet of things based on federal incremental learning[J]. Journal of Computer Applications, 2022, 42(4): 1235-1243.
编号 | 轴承位置 | 故障部位 | 故障直径/mm |
---|---|---|---|
1 | None | None | 0.000 |
2 | Driver End | IRW | 0.007 |
3 | Driver End | IRW | 0.014 |
4 | Driver End | ORW | 0.007 |
5 | Driver End | ORW | 0.014 |
6 | Driver End | BA | 0.007 |
7 | Driver End | BA | 0.014 |
8 | Fan End | IRW | 0.007 |
9 | Fan End | IRW | 0.014 |
10 | Fan End | ORW | 0.007 |
11 | Fan End | ORW | 0.014 |
12 | Fan End | BA | 0.007 |
13 | Fan End | BA | 0.014 |
表1 轴承故障实验数据描述
Tab. 1 Experimental data description of bearing failure
编号 | 轴承位置 | 故障部位 | 故障直径/mm |
---|---|---|---|
1 | None | None | 0.000 |
2 | Driver End | IRW | 0.007 |
3 | Driver End | IRW | 0.014 |
4 | Driver End | ORW | 0.007 |
5 | Driver End | ORW | 0.014 |
6 | Driver End | BA | 0.007 |
7 | Driver End | BA | 0.014 |
8 | Fan End | IRW | 0.007 |
9 | Fan End | IRW | 0.014 |
10 | Fan End | ORW | 0.007 |
11 | Fan End | ORW | 0.014 |
12 | Fan End | BA | 0.007 |
13 | Fan End | BA | 0.014 |
网络层 | 权重 | 偏置 | |
---|---|---|---|
weight_hh | weight_ih | ||
LSTM层 | 256 × 100 | 256 × 100 | 100 |
dense层1 | 256 × 64 | 64 | |
dense层2 | 64 × 13 | 13 |
表2 LSTM参数
Tab. 2 LSTM parameters
网络层 | 权重 | 偏置 | |
---|---|---|---|
weight_hh | weight_ih | ||
LSTM层 | 256 × 100 | 256 × 100 | 100 |
dense层1 | 256 × 64 | 64 | |
dense层2 | 64 × 13 | 13 |
F值 | 通信轮次 | 每轮时间/s | 训练集准确率 | 测试集准确率 |
---|---|---|---|---|
0.3 | 33 | 12.395 | 0.954 8 | 0.938 8 |
0.5 | 29 | 18.703 | 0.960 2 | 0.943 9 |
0.7 | 26 | 23.787 | 0.962 3 | 0.945 4 |
1.0 | 17 | 28.854 | 0.972 1 | 0.963 1 |
表3 比例系数对模型性能影响
Tab. 3 Influence of scale factor on model performance
F值 | 通信轮次 | 每轮时间/s | 训练集准确率 | 测试集准确率 |
---|---|---|---|---|
0.3 | 33 | 12.395 | 0.954 8 | 0.938 8 |
0.5 | 29 | 18.703 | 0.960 2 | 0.943 9 |
0.7 | 26 | 23.787 | 0.962 3 | 0.945 4 |
1.0 | 17 | 28.854 | 0.972 1 | 0.963 1 |
F 值 | 方法 | 工厂子端 | 不同通信轮次下的极值 | ||||
---|---|---|---|---|---|---|---|
1 | 5 | 10 | 20 | 30 | |||
0.3 | FIL-IIOT | Factory_0 | 3.127 | 0.841 | 0.507 | 1.432 | 0.767 |
Factory_1 | 0.768 | 1.960 | 0.527 | 1.576 | 1.782 | ||
Factory_2 | 2.547 | 1.482 | 1.653 | 2.276 | 2.520 | ||
Factory_3 | 1.926 | 2.568 | 0.911 | 1.374 | 2.167 | ||
Factory_4 | 0.202 | 2.962 | 2.120 | 1.096 | 0.796 | ||
Factory_5 | 2.772 | 3.005 | 0.232 | 2.895 | 1.279 | ||
Factory_6 | 2.352 | 0.628 | 2.937 | 1.652 | 1.581 | ||
Factory_7 | 3.779 | 1.384 | 0.132 | 1.563 | 1.280 | ||
Factory_8 | 3.003 | 2.370 | 0.523 | 1.057 | 1.527 | ||
Factory_9 | 2.830 | 1.196 | 1.268 | 0.904 | 1.666 | ||
FedAvg | Factory_0 | 2.760 | 0.276 | 3.096 | 1.994 | 1.896 | |
Factory_1 | 2.626 | 1.062 | 1.944 | 1.137 | 1.271 | ||
Factory_2 | 2.683 | 1.303 | 3.588 | 1.273 | 2.037 | ||
Factory_3 | 1.415 | 2.944 | 1.911 | 0.994 | 1.199 | ||
Factory_4 | 3.185 | 2.477 | 1.042 | 1.810 | 0.301 | ||
Factory_5 | 1.688 | 2.420 | 1.038 | 2.294 | 3.023 | ||
Factory_6 | 2.151 | 2.709 | 2.737 | 1.252 | 2.332 | ||
Factory_7 | 0.444 | 2.344 | 2.365 | 0.966 | 0.942 | ||
Factory_8 | 0.499 | 1.463 | 2.148 | 0.167 | 2.233 | ||
Factory_9 | 2.947 | 2.710 | 2.616 | 2.350 | 1.277 | ||
1.0 | FedAvg | Factory_0 | 0.357 | 1.789 | 2.212 | 2.352 | 0.738 |
Factory_1 | 2.352 | 0.598 | 2.570 | 1.657 | 2.679 | ||
Factory_2 | 0.766 | 1.832 | 1.408 | 1.253 | 2.489 | ||
Factory_3 | 2.594 | 0.495 | 0.458 | 1.632 | 2.829 | ||
Factory_4 | 0.403 | 2.411 | 2.549 | 2.964 | 1.682 | ||
Factory_5 | 1.386 | 1.394 | 2.096 | 1.823 | 2.205 | ||
Factory_6 | 0.411 | 3.620 | 1.230 | 1.489 | 2.806 | ||
Factory_7 | 2.235 | 1.879 | 1.745 | 0.861 | 0.859 | ||
Factory_8 | 0.641 | 3.447 | 3.177 | 0.630 | 1.399 | ||
Factory_9 | 2.587 | 2.300 | 2.871 | 1.981 | 2.778 |
表4 工厂子端等级值变化情况
Tab. 4 Change of factory sub-end level value
F 值 | 方法 | 工厂子端 | 不同通信轮次下的极值 | ||||
---|---|---|---|---|---|---|---|
1 | 5 | 10 | 20 | 30 | |||
0.3 | FIL-IIOT | Factory_0 | 3.127 | 0.841 | 0.507 | 1.432 | 0.767 |
Factory_1 | 0.768 | 1.960 | 0.527 | 1.576 | 1.782 | ||
Factory_2 | 2.547 | 1.482 | 1.653 | 2.276 | 2.520 | ||
Factory_3 | 1.926 | 2.568 | 0.911 | 1.374 | 2.167 | ||
Factory_4 | 0.202 | 2.962 | 2.120 | 1.096 | 0.796 | ||
Factory_5 | 2.772 | 3.005 | 0.232 | 2.895 | 1.279 | ||
Factory_6 | 2.352 | 0.628 | 2.937 | 1.652 | 1.581 | ||
Factory_7 | 3.779 | 1.384 | 0.132 | 1.563 | 1.280 | ||
Factory_8 | 3.003 | 2.370 | 0.523 | 1.057 | 1.527 | ||
Factory_9 | 2.830 | 1.196 | 1.268 | 0.904 | 1.666 | ||
FedAvg | Factory_0 | 2.760 | 0.276 | 3.096 | 1.994 | 1.896 | |
Factory_1 | 2.626 | 1.062 | 1.944 | 1.137 | 1.271 | ||
Factory_2 | 2.683 | 1.303 | 3.588 | 1.273 | 2.037 | ||
Factory_3 | 1.415 | 2.944 | 1.911 | 0.994 | 1.199 | ||
Factory_4 | 3.185 | 2.477 | 1.042 | 1.810 | 0.301 | ||
Factory_5 | 1.688 | 2.420 | 1.038 | 2.294 | 3.023 | ||
Factory_6 | 2.151 | 2.709 | 2.737 | 1.252 | 2.332 | ||
Factory_7 | 0.444 | 2.344 | 2.365 | 0.966 | 0.942 | ||
Factory_8 | 0.499 | 1.463 | 2.148 | 0.167 | 2.233 | ||
Factory_9 | 2.947 | 2.710 | 2.616 | 2.350 | 1.277 | ||
1.0 | FedAvg | Factory_0 | 0.357 | 1.789 | 2.212 | 2.352 | 0.738 |
Factory_1 | 2.352 | 0.598 | 2.570 | 1.657 | 2.679 | ||
Factory_2 | 0.766 | 1.832 | 1.408 | 1.253 | 2.489 | ||
Factory_3 | 2.594 | 0.495 | 0.458 | 1.632 | 2.829 | ||
Factory_4 | 0.403 | 2.411 | 2.549 | 2.964 | 1.682 | ||
Factory_5 | 1.386 | 1.394 | 2.096 | 1.823 | 2.205 | ||
Factory_6 | 0.411 | 3.620 | 1.230 | 1.489 | 2.806 | ||
Factory_7 | 2.235 | 1.879 | 1.745 | 0.861 | 0.859 | ||
Factory_8 | 0.641 | 3.447 | 3.177 | 0.630 | 1.399 | ||
Factory_9 | 2.587 | 2.300 | 2.871 | 1.981 | 2.778 |
F 值 | 方法 | 训练集准确率 | 测试集准确率 | 时间/s |
---|---|---|---|---|
1.0 | FedAvg | 0.968 7 | 0.944 2 | 754 |
0.3 | FedAvg | 0.933 6 | 0.917 7 | 353 |
FIL-IIOT | 0.955 6 | 0.931 8 | 362 |
表5 联邦优选算法性能对比
Tab. 5 Performance comparison of federal optimization algorithm
F 值 | 方法 | 训练集准确率 | 测试集准确率 | 时间/s |
---|---|---|---|---|
1.0 | FedAvg | 0.968 7 | 0.944 2 | 754 |
0.3 | FedAvg | 0.933 6 | 0.917 7 | 353 |
FIL-IIOT | 0.955 6 | 0.931 8 | 362 |
方法 | 训练精度 | 训练时间/s | 测试精度 | 测试时间/s |
---|---|---|---|---|
FedAvg | 0.882 2 | 722 | 0.869 7 | 83 |
FIL-IIOT-NI | 0.921 4 | 746 | 0.905 6 | 68 |
FIL-IIOT | 0.945 6 | 478 | 0.931 5 | 65 |
表6 增量故障分类结果对比
Tab. 6 Comparison of incremental failure classification results
方法 | 训练精度 | 训练时间/s | 测试精度 | 测试时间/s |
---|---|---|---|---|
FedAvg | 0.882 2 | 722 | 0.869 7 | 83 |
FIL-IIOT-NI | 0.921 4 | 746 | 0.905 6 | 68 |
FIL-IIOT | 0.945 6 | 478 | 0.931 5 | 65 |
1 | 李明时,马跃,尹震宇,等.一种多重冗余的工业物联网智能产线安全通信模型设计[J].小型微型计算机系统,2021,42(3):621-626. 10.3969/j.issn.1000-1220.2021.03.030 |
LI M S, MA Y, YIN Z Y, et al. Design of a multi-redundant secure communication model for intelligent production line of industrial Internet of things[J]. Journal of Chinese Computer Systems, 2021, 42(3):621-626. 10.3969/j.issn.1000-1220.2021.03.030 | |
2 | 张君如,赵晓焱,袁培燕.面向用户隐私保护的联邦安全树算法[J]. 计算机应用,2020,40(10):2980-2985. |
ZHANG J R, ZHAO X Y, YUAN P Y. Federated security tree algorithm for user privacy protection[J]. Journal of Computer Applications, 2020, 40(10):2980-2985. | |
3 | 罗长银,陈学斌,马春地,等.面向区块链的在线联邦增量学习算法[J].计算机应用,2021,41(2):363-371. 10.11772/j.issn.1001-9081.2020050609 |
LUO C Y, CHEN X B, MA C D, et al. Online federated incremental learning algorithm for blockchain[J]. Journal of Computer Applications, 2021, 41(2):363-371. 10.11772/j.issn.1001-9081.2020050609 | |
4 | 芦效峰,廖钰盈,LIO P,等.一种面向边缘计算的高效异步联邦学习机制[J].计算机研究与发展,2020,57(12):2571-2582. 10.7544/issn1000-1239.2020.20190754 |
LU X F, LIAO Y Y, LIO P, et al. An asynchronous federated learning mechanism for edge network computing[J]. Journal of Computer Research and Development, 2020, 57(12):2571-2582. 10.7544/issn1000-1239.2020.20190754 | |
5 | HU B X, GAO Y J, LIU L, et al. Federated region-learning: an edge computing based framework for urban environment sensing[C]// Proceedings of the 2018 IEEE Global Communications Conference. Piscataway: IEEE, 2018:1-7. 10.1109/glocom.2018.8647649 |
6 | YANG W S, ZHANG Y H, YE K J, et al. FFD: a federated learning based method for credit card fraud detection[C]// Proceedings of the 2019 International Conference on Big Data, LNCS 11514. Cham: Springer, 2019:18-32. |
7 | HU Y, SUN X Y, CHEN Y, et al. Model and feature aggregation based federated learning for multi-sensor time series trend following[C]// Proceedings of the 2019 International Work-Conference on Artificial Neural Networks, LNCS 11506. Cham: Springer, 2019:233-246. |
8 | 王蓉,马春光,武朋.基于联邦学习和卷积神经网络的入侵检测方法[J].信息网络安全,2020,20(4):47-54. 10.3969/j.issn.1671-1122.2020.04.006 |
WANG R, MA C G, WU P. An intrusion detection method based on federated learning and convolutional neural network[J]. Netinfo Security, 2020, 20(4):47-54. 10.3969/j.issn.1671-1122.2020.04.006 | |
9 | CHEN Y, SUN X Y, JIN Y C. Communication-efficient federated deep learning with layerwise asynchronous model update and temporally weighted aggregation[J]. IEEE Transactions on Neural Networks and Learning Systems, 2020, 31(10):4229-4238. 10.1109/tnnls.2019.2953131 |
10 | House White. Consumer data privacy in a networked world: a framework for protecting privacy and promoting innovation in the global digital economy[R]. Washington, DC: White House, 2012. |
11 | McMAHAN H B, MOORE E, RAMAGE D, et al. Communication-efficient learning of deep networks from decentralized data[C]// Proceedings of the 20th International Conference on Artificial Intelligence and Statistics. New York: JMLR.org, 2017:1273-1282. |
12 | KONEČNÝ J, McMAHAN H B, YU F X, et al. Federated learning: strategies for improving communication efficiency[EB/OL]. (2017-10-30) [2020-07-12].. |
13 | LU Y L, HUANG X H, ZHANG K, et al. Blockchain empowered asynchronous federated learning for secure data sharing in internet of vehicles[J]. IEEE Transactions on Vehicular Technology, 2020, 69(4):4298-4311. 10.1109/tvt.2020.2973651 |
14 | FedAI. Utilization of FATE in risk management of credit in small and micro enterprises[EB/OL]. [2020-07-12].. 10.14738/abr.49.872 |
15 | OWKIN, Inc. Federated learning to accelerate medical research[EB/OL]. [2020-07-12].. 10.53347/rid-81590 |
16 | QIU J, CHAI Y H, TIAN Z H, et al. Automatic concept extraction based on semantic graphs from big data in smart city[J]. IEEE Transactions on Computational Social Systems, 2020, 7(1):225-233. 10.1109/tcss.2019.2946181 |
17 | ZHENG Y, LIU F R, HSIEH H P. U-Air: when urban air quality inference meets big data[C]// Proceedings of the 19th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. New York: ACM, 2013:1436-1444. 10.1145/2487575.2488188 |
18 | KONEČNÝ J, McMAHAN B, RAMAGE D, et al. Federated optimization: distributed machine learning for on-device intelligence[EB/OL]. (2016-10-08) [2020-08-15].. |
19 | PREUVENEERS D, RIMMER V, TSINGENOPOULOS I, et al. Chained anomaly detection models for federated learning: an intrusion detection case study[J]. Applied Sciences, 2018, 8(12):No.2663. 10.3390/app8122663 |
20 | 林毅,吉鸿江,韩佳佳,等.一种基于马氏距离的系统故障诊断方法[J]. 计算机科学,2020,47(S2):57-63. |
LIN Y, JI H J, HAN J J, et al. System fault diagnosis method based on mahalanobis distance metric[J]. Computer Science, 2020, 47(S2):57-63. | |
21 | 毛文涛,杨超,刘亚敏,等.面向轴承早期故障检测的多尺度残差注意力深度领域适配模型[J].计算机应用,2020,40(10):2890-2898. |
MAO W T, YANG C, LIU Y M, et al. Deep domain adaptation model with multi-scale residual attention for incipient fault detection of bearings[J]. Journal of Computer Applications, 2020, 40(10):2890-2898. | |
22 | LIU J, AN Y C, DOU R L, et al. Helical fault diagnosis model based on data-driven incremental mergence[J]. Computers and Industrial Engineering, 2018, 125:517-532. 10.1016/j.cie.2018.02.002 |
[1] | 王乐, 韩萌, 李小娟, 张妮, 程浩东. 基于动态加权函数的集成分类算法[J]. 《计算机应用》唯一官方网站, 2022, 42(4): 1137-1147. |
[2] | 罗长银, 王君宇, 陈学斌, 马春地, 张淑芬. 改进的联邦加权平均算法[J]. 《计算机应用》唯一官方网站, 2022, 42(4): 1131-1136. |
[3] | 邱鑫源, 叶泽聪, 崔翛龙, 高志强. 联邦学习通信开销研究综述[J]. 《计算机应用》唯一官方网站, 2022, 42(2): 333-342. |
[4] | 刘超, 王磊, 杨文, 钟强强, 黎敏. 属性集变化条件下集值决策信息系统的增量属性约简方法[J]. 《计算机应用》唯一官方网站, 2022, 42(2): 463-468. |
[5] | 郭棉, 张锦友. 移动边缘计算环境中面向机器学习的计算迁移策略[J]. 计算机应用, 2021, 41(9): 2639-2645. |
[6] | 董文涛, 李卓, 陈昕. 基于联邦学习的在线短视频内容分发策略[J]. 计算机应用, 2021, 41(6): 1551-1556. |
[7] | 王家瑞, 谭国平, 周思源. 高速车联网场景下分簇式无线联邦学习算法[J]. 计算机应用, 2021, 41(6): 1546-1550. |
[8] | 罗长银, 陈学斌, 马春地, 王君宇. 面向区块链的在线联邦增量学习算法[J]. 计算机应用, 2021, 41(2): 363-371. |
[9] | 邓文瀚, 张铭, 王李进, 钟一文. 混合群体增量学习算法求解闭环布局问题[J]. 计算机应用, 2021, 41(1): 95-102. |
[10] | 刘丹, 姚立霜, 王云锋, 裴作飞. 面向类不平衡流量数据的分类模型[J]. 计算机应用, 2020, 40(8): 2327-2333. |
[11] | 骆克云, 叶保留, 唐斌, 梅峰, 卢文达. 基于增量学习的RocksDB键值系统主动缓存机制[J]. 《计算机应用》唯一官方网站, 2020, 40(2): 321-327. |
[12] | 张君如, 赵晓焱, 袁培燕. 面向用户隐私保护的联邦安全树算法[J]. 计算机应用, 2020, 40(10): 2980-2985. |
[13] | 忽丽莎, 王素贞, 陈益强, 胡春雨, 蒋鑫龙, 陈振宇, 高兴宇. 基于目标均衡度量的核增量学习跌倒检测方法[J]. 计算机应用, 2018, 38(4): 928-934. |
[14] | 汪金涛, 曹玉东, 孙福明. 稀疏约束图正则非负矩阵分解的增量学习算法[J]. 计算机应用, 2017, 37(4): 1071-1074. |
[15] | 霍纬纲, 屈峰, 程震. 基于演进向量量化聚类的增量模糊关联分类方法[J]. 计算机应用, 2017, 37(11): 3075-3079. |
阅读次数 | ||||||
全文 |
|
|||||
摘要 |
|
|||||