Intrusion detection framework based on homomorphic encryption in AMI network
In order to alleviate the privacy issue of traditional smart grids, some researchers have proposed a power metering system based on a federated learning framework, which jointly trains the model by exchanging gradients between multiple data owners instead of raw data. However, recent research shows...
| Main Authors: | , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Frontiers Media S.A.
2022-12-01
|
| Series: | Frontiers in Physics |
| Subjects: | |
| Online Access: | https://www.frontiersin.org/articles/10.3389/fphy.2022.1102892/full |
| _version_ | 1797821805266731008 |
|---|---|
| author | Jing Wang Zhuoqun Xia Yaling Chen Yaling Chen Chang Hu Fei Yu |
| author_facet | Jing Wang Zhuoqun Xia Yaling Chen Yaling Chen Chang Hu Fei Yu |
| author_sort | Jing Wang |
| collection | DOAJ |
| description | In order to alleviate the privacy issue of traditional smart grids, some researchers have proposed a power metering system based on a federated learning framework, which jointly trains the model by exchanging gradients between multiple data owners instead of raw data. However, recent research shows that the federated learning framework still has privacy and security issues. Secondly, since the server does not have direct access to all parties data sets and training process, malicious attackers may conduct poisoning attacks. This is a new security threat in federated learning - poisoning attack. However, solving the two problems at the same time seems to be contradictory because privacy protection requires the inseparability of the training gradients of all parties, and security requires the server to be able to identify the poisoned client. To solve the above issues, this paper proposes an intrusion detection method based on federated learning client-side security in AMI networks, which uses CKKS to protect model parameters. In addition, to resist the poisoning attack in federated learning, the model trained by the data processing center and the model trained by each client are firstly calculated for the direction similarity, and the similarity value is scaled as the adaptive weight of the aggregation model. Then, the size of each client model update is normalized to be the same size as the data processing center model update. Finally, the normalized updates and adaptive weights are weighted averaged to form a global model update. The research results show that the method in this paper can effectively resist inference attacks and poisoning attacks. In the AMI network, the intrusion detection method based on federated learning can maintain a good detection performance. |
| first_indexed | 2024-03-13T09:58:14Z |
| format | Article |
| id | doaj.art-01e8b1fc4c594244a2045b71ca2f0326 |
| institution | Directory Open Access Journal |
| issn | 2296-424X |
| language | English |
| last_indexed | 2024-03-13T09:58:14Z |
| publishDate | 2022-12-01 |
| publisher | Frontiers Media S.A. |
| record_format | Article |
| series | Frontiers in Physics |
| spelling | doaj.art-01e8b1fc4c594244a2045b71ca2f03262023-05-23T08:56:48ZengFrontiers Media S.A.Frontiers in Physics2296-424X2022-12-011010.3389/fphy.2022.11028921102892Intrusion detection framework based on homomorphic encryption in AMI networkJing Wang0Zhuoqun Xia1Yaling Chen2Yaling Chen3Chang Hu4Fei Yu5School of Computer and Communication Engineering, Changsha University of Science and Technology, Changsha, ChinaSchool of Computer and Communication Engineering, Changsha University of Science and Technology, Changsha, ChinaSchool of Electrical and Information Engineering, Hunan Institute of Traffic Engineering, Hengyang, ChinaSchool of Computer and Communication Engineering, Changsha University of Science and Technology, Changsha, ChinaSchool of Computer and Communication Engineering, Changsha University of Science and Technology, Changsha, ChinaSchool of Computer and Communication Engineering, Changsha University of Science and Technology, Changsha, ChinaIn order to alleviate the privacy issue of traditional smart grids, some researchers have proposed a power metering system based on a federated learning framework, which jointly trains the model by exchanging gradients between multiple data owners instead of raw data. However, recent research shows that the federated learning framework still has privacy and security issues. Secondly, since the server does not have direct access to all parties data sets and training process, malicious attackers may conduct poisoning attacks. This is a new security threat in federated learning - poisoning attack. However, solving the two problems at the same time seems to be contradictory because privacy protection requires the inseparability of the training gradients of all parties, and security requires the server to be able to identify the poisoned client. To solve the above issues, this paper proposes an intrusion detection method based on federated learning client-side security in AMI networks, which uses CKKS to protect model parameters. In addition, to resist the poisoning attack in federated learning, the model trained by the data processing center and the model trained by each client are firstly calculated for the direction similarity, and the similarity value is scaled as the adaptive weight of the aggregation model. Then, the size of each client model update is normalized to be the same size as the data processing center model update. Finally, the normalized updates and adaptive weights are weighted averaged to form a global model update. The research results show that the method in this paper can effectively resist inference attacks and poisoning attacks. In the AMI network, the intrusion detection method based on federated learning can maintain a good detection performance.https://www.frontiersin.org/articles/10.3389/fphy.2022.1102892/fulladvanced metering infrastructureintrusion detectionfederated learninghomomorphic encryptionprivacy protectionpoisoning attack |
| spellingShingle | Jing Wang Zhuoqun Xia Yaling Chen Yaling Chen Chang Hu Fei Yu Intrusion detection framework based on homomorphic encryption in AMI network Frontiers in Physics advanced metering infrastructure intrusion detection federated learning homomorphic encryption privacy protection poisoning attack |
| title | Intrusion detection framework based on homomorphic encryption in AMI network |
| title_full | Intrusion detection framework based on homomorphic encryption in AMI network |
| title_fullStr | Intrusion detection framework based on homomorphic encryption in AMI network |
| title_full_unstemmed | Intrusion detection framework based on homomorphic encryption in AMI network |
| title_short | Intrusion detection framework based on homomorphic encryption in AMI network |
| title_sort | intrusion detection framework based on homomorphic encryption in ami network |
| topic | advanced metering infrastructure intrusion detection federated learning homomorphic encryption privacy protection poisoning attack |
| url | https://www.frontiersin.org/articles/10.3389/fphy.2022.1102892/full |
| work_keys_str_mv | AT jingwang intrusiondetectionframeworkbasedonhomomorphicencryptioninaminetwork AT zhuoqunxia intrusiondetectionframeworkbasedonhomomorphicencryptioninaminetwork AT yalingchen intrusiondetectionframeworkbasedonhomomorphicencryptioninaminetwork AT yalingchen intrusiondetectionframeworkbasedonhomomorphicencryptioninaminetwork AT changhu intrusiondetectionframeworkbasedonhomomorphicencryptioninaminetwork AT feiyu intrusiondetectionframeworkbasedonhomomorphicencryptioninaminetwork |