基于自适应生成对抗网络的智能电网状态重构的虚假数据攻击检测  

False Data Injection Attack Detection for Smart Grid State Reconstruction Based on Adaptive Generative Adversarial Networks

作　　者：王新宇[1] 王相杰 罗小元[1] WANG Xinyu;WANG Xiangjie;LUO Xiaoyuan(School of Electrical Engineering,Yanshan University,Qinhuangdao 066004,Hebei Province,China)

机构地区：[1]燕山大学电气工程系,河北省秦皇岛市066004

出　　处：《电力信息与通信技术》2024年第9期1-7,共7页Electric Power Information and Communication Technology

基　　金：国家自然科学基金(62103357);河北省自然科学基金青年项目(F2021203043)

摘　　要：考虑到电力系统与能源互联网的深度耦合,虚假数据注入攻击对电力系统的威胁不断提升。文章针对虚假数据注入攻击(false data injection attack,FDIA)设计自适应生成对抗网络(adaptive generative adversarial networks,AGAN)状态重构的虚假数据注入攻击检测方法。该方法在生成对抗网络(generative adversarial networks,GAN)基础上融入卷积神经网络(convolutional neural network,CNN)以及自适应约束下的自注意力机制(self-attention,SA),实现节点间全局参考性,从而实现状态的有效重构和异常状态的准确预测;根据AGAN的异常数据预测结果设计结合网络判别值的检测逻辑。最后,在IEEE14节点的电力系统上验证所提方法的有效性,且对比GAN、CNN,AGAN重构的平均绝对百分比误差为0.0001%,检测准确率可达到98%。Considering the deep coupling between power system and energy Internet,the threat of false data injection attack to power system is increasing.This paper designs an adaptive generative adversarial networks(AGAN)state reconstruction method for false data injection attack(FDIA)detection.AGAN state reconstruction method for false data injection attack detection.The method is based on generative adversarial networks(GAN),incorporates convolutional neural network(CNN)and self-attention(SA)mechanism under adaptive constraints to achieve global referentiality among nodes,so as to realize effective reconstruction of states and accurate prediction of abnormal states.The detection logic combining network discriminant values is proposed based on the abnormal data prediction results of AGAN.Finally,the effectiveness of the proposed method is verified on the power system of IEEE14 nodes,and the average absolute percentage error of AGAN reconstruction is 0.0001%compared with GAN and CNN,and the detection accuracy can reach 98%.

关 键 词：生成对抗网络 自注意力机制 自适应约束 电力系统 虚假数据注入攻击 

分 类 号：TN915.08[电子电信—通信与信息系统]