分布式新能源场景下配电网虚假数据注入攻击检测OA
Detection of False Data Injection Attacks on Power Distribution Networks in Distributed Renewable Energy Scenarios
[目的]随着新型电力系统中分布式节点广泛接入配电网,频繁的数据交互增加了配电网遭受虚假数据注入攻击(false data injection attacks,FDIA)的风险.常规的数据驱动检测方法在挖掘数据特征时往往将所有数据作为一个整体,忽略了不同节点数据中的个性特征.针对这一问题,文章提出了一种基于最大信息系数的个性化联邦训练方法,用于分布式新能源场景下的虚假数据注入攻击检测.[方法]所提方法将检测模型部署在分布式边缘节点,提高了边缘节点的网络安全防护能力及本地数据隐私保护能力;通过应用多层神经网络进行个性化联邦训练,将其分为不同特征层来进行共性和个性特征分离,在分布式检测的基础上加强对异构节点数据的特征处理;考虑量测数据中的时间特征,通过引入最大信息系数深入挖掘数据中潜在的规律性特征,将分析结果融合个性化联邦训练,以提高对节点本身数据个性特征的提取能力.[结果]以含分布式新能源节点的园区数据为例进行仿真分析,所提方法相比传统联邦框架和不考虑相关性分析的检测方法,检测准确率、精确率、召回率和F1分数均有所提升;最大信息系数在处理周期性数据时具有较好的个性特征提取能力.[结论]所提方法增加了对数据共性和个性特征的分离和提取,在客户端数量较多时检测模型具有较快的收敛速率,更适合分布式新能源场景下的FDIA检测.
[Objective]With the extensive integration of distributed nodes in new power systems into distribution networks,frequent data interactions increase the risk of false data injection attacks(FDIA)on the distribution networks.Conventional data-driven detection methods tend to treat all data holistically when mining data features,usually ignoring individual characteristics in data from different nodes.To address this problem,this paper proposes a personalized federated training method based on maximum information coefficient for false data injection attack detection in distributed renewable energy scenarios.[Methods]The proposed method deploys the detection model in distributed edge nodes,which improves the network security protection and local data privacy protection of the edge nodes.Multi-layer neural networks subjected to personalized federated training are separated into distinct feature layers to decouple common and individual features,thereby enhancing the feature processing of heterogeneous node data on the basis of distributed detection.Considering the temporal features in the measurement data,the potential regular features in the data are deeply mined by introducing the maximum information coefficient,and the analysis results are fused into the personalized federated training in order to improve the ability of extracting the personality features of the nodes'own data.[Results]The park data containing distributed renewable energy nodes is taken as an example for simulation analysis,and the proposed method improves the detection accuracy,precision,recall,and F1 score compared to the traditional federated framework and the detection method that does not consider correlation analysis.Maximum information coefficient demonstrates better personality feature extraction when dealing with periodic data.[Conclusions]The proposed method enhances the separation and extraction of common and individual features of the data,and the detection model exhibits a faster convergence rate when there are a large number of clients,rendering it more suitable for FDIA detection in distributed renewable energy scenarios.
龚钢军;张晓炜;王路遥;李璐含;黄雨菲;王浩淼;扬爽
北京市能源电力信息安全工程技术研究中心(华北电力大学),北京市 102206北京市能源电力信息安全工程技术研究中心(华北电力大学),北京市 102206北京市能源电力信息安全工程技术研究中心(华北电力大学),北京市 102206北京市能源电力信息安全工程技术研究中心(华北电力大学),北京市 102206北京市能源电力信息安全工程技术研究中心(华北电力大学),北京市 102206国网辽宁省电力有限公司,沈阳市 110002国网辽宁省电力有限公司,沈阳市 110002
信息技术与安全科学
虚假数据注入攻击(FDIA)分布式节点个性化联邦学习最大信息系数数据安全
false data injection attack(FDIA)distributed nodespersonalized federated learningmaximum information coefficientdata security
《电力建设》 2026 (4)
16-27,12
国家重点研发计划资助项目(2022YFB3105100) This work is supported by National Key R&D Program of China(No.2022YFB3105100).
评论