基于XGBoost和无迹卡尔曼滤波自适应混合预测的电网虚假数据注入攻击检测

Grid False Data Injection Attacks Detection Based on XGBoost and Unscented Kalman Filter Adaptive Hybrid Prediction

随着信息技术在电力系统中的广泛应用,电网正发展为一类信息系统与物理系统高度融合的电力信息物理系统(cyber-physical system,CPS)。而虚假数据注入攻击(false data injection attacks,FDIA)是影响电力CPS安全运行的隐患之一。为了能够检测与修正虚假数据注入攻击,提出一种基于极端梯度提升(extreme gradient boosting,XGBoost)结合无迹卡尔曼滤波(unscented Kalman filter,UKF)的电网虚假数据注入攻击检测方法。首先通过改进的加权灰色关联分析法进行相似日的选取,然后使用XGBoost进行电力系统日前负荷预测;将负荷预测结果经潮流计算得到的状态量与UKF动态状态估计得到的状态量进行自适应混合预测,以降低FDIA对状态预测的影响;最后基于预测值和静态状态估计值构造随机变量,通过中心极限定理比较随机变量的分布以进行FDIA检测与修正。IEEE-14和IEEE-118节点测试系统仿真结果验证了文中提出方法的有效性和准确性。

With the widespread application of information technology in power system,power grid has become a type of power cyber-physical system(CPS)in which cyber system and physical system are highly integrated.False data injection attacks(FDIA)is one of the hidden dangers that affect the safe operation of power CPS.In order to detect and correct FDIA,a detection method of FDIA based on the combination of extreme gradient boosting(XGBoost)and unscented Kalman filter(UKF)was proposed in this paper.Firstly,the similar day was selected by the improved weighted grey relational analysis method,and the XGBoost was used to forecast day-ahead load of power system.Then the state quantity obtained by power flow computation with load prediction results and UKF dynamic state estimation were used to carry out adaptive hybrid prediction to reduce the impact of FDIA on state prediction.Finally,the random variable was constructed based on the predicted value and the value of the static state estimation.And the distribution of the random variable was compared by the central limit theorem to detect and correct FDIA.The simulation results of IEEE-14 and IEEE-118 bus test systems verify the effectiveness of the proposed method.