基于数据驱动的电力系统虚假数据注入攻击防御框架的研究

Research on defense framework for false data injection attacks in power system based on data-driven algorithm

电力系统状态估计(power system state estimation,PSSE)在现代智能电网的稳定运行中起着至关重要的作用,但它也容易遭受网络攻击。虚假数据注入攻击(false data injection attacks,FDIA)是最常见的网络攻击方式之一,它可以篡改量测数据并绕过不良数据检测(bad data detection,BDD)机制,从而导致不正确的状态估计结果。文中提出一种基于数据驱动的针对PSSE的FDIA防御框架,该框架包含异常检测子框架和数据恢复子框架。异常检测部分采用改进的图卷积网络(improved graph convolutional network,IGCN)模型,该模型采用动态的边缘条件滤波器作用于图结构中,有效利用电力系统的拓扑信息、节点特征和边特征,从而检测出异常值。数据恢复部分采用变分自编码器(variational auto-encoder,VAE)模型,该模型将深度学习思想与贝叶斯推理相结合,可以有效地将异常数据恢复到在正常运行情况下的数值。针对不同攻击强度和攻击程度下的IEEE 14系统进行案例研究,以评估防御框架的检测与恢复性能。仿真结果表明,基于IGCN的异常检测子框架性能优于常规的数据驱动模型框架,其总体精确率为99.348%,召回率为99.331%,F1值为99.324%,基于VAE的数据恢复子框架的总体平均绝对误差为0.00534 p.u.,证明了防御框架优异的检测与恢复性能。

Power system state estimation(PSSE)plays a crucial role in the stable operation of modern smart grid,while it is vulnerable to cyber attacks.False data injection attacks(FDIA)are one of the most common cyber attacks,which can tamper with measurement data and bypass bad data detection(BDD)mechanism and lead to incorrect PSSE results.This paper proposes a data-driven-based FDIA defense framework against PSSE,which includes an anomaly detection sub-framework and a data recovery sub-framework.The anomaly detection part adopts improved graph convolutional network(IGCN)model.The model uses dynamic edge-conditioned filters to act on the graph structure,which effectively utilizes the topology information,node features and edge features of power systems to detect abnormal values.The data recovery part adopts the variational auto-encoder(VAE)model.The model combines deep learning ideas with Bayesian inference,which can effectively restore the abnormal data to values under normal operating conditions.Case studies are conducted on IEEE 14-bus system under different attack intensities and attack degrees to evaluate the detection and recovery performance of the defense framework.Simulation results show that the IGCN-based anomaly detection model outperforms the conventional data-driven model framework,with the overall precision rate of 99.348%,the recall rate of 99.331%,and the F 1-score of 99.324%and the overall mean absolute error of the VAE-based data recovery model is 0.00534 p.u.,demonstrating the excellent detection and recovery performance of the defense framework.