Research On Power System Security Detection Technology Under False Data Attack

Accurate state estimation is critical to wide-area situational awareness of smart grid.Through the real-time estimation and analysis of power system state,electric power operators issue the optimal decisions on each core component of the smart grid,to ensure the safe,stable and economic operation of power grids.However,power system state estimators are vulnerable to a newly type of cyber-attack,called false data injection attack(FDIA).Aiming at destroying the integrity of the power system data,this attack modifies the results of state estimation by tampering with state estimation measurement data,and can avoid the detection of the traditional bad data identification mechanism,causing power system control center to misjudge the situation of power system,directly threatening the system running safely and stably.Therefore,in order to ensure the security of power system data,it is of great significance to study the attack mechanism of such cyber-attack,develop effective detection methods and defensive measures.This article focuses on the study of the detection and defensive method of false data injection attacks:Firstly,a hybrid FDIA detection mechanism considers time correlation is proposed.At present,for most of the attackers,the design principle of FDIA data is to make the attack vectors satisfying the constraints of physical characteristics of the power grid,however,the time correlation existing in power system data is neglected.According to this idea,variational modal decomposition technology is used in our proposed detection method to extract the features of the FDIA attack behavior among the time series of system states.Then,several statistically-based indicators are used to describe the abrupt features.Further,according to the obtained feature indicators,OS-extreme learning machine which has sequential learning ability is trained as detector to fill the demand of real-time detection.Finally,F-test is used to conduct a global assessment of the security status of the power system to distinguish the false data attacks from physical grid failures.Secondly,most of the existing FDIA detection methods are designed for a specific network structure,require prior knowledge and a huge amounts of historical data to train the detector.However,the lack of real-world attack samples is a common problem faced by power security researchers.To tackle the problem,a double-layer FDIA defense mechanism is proposed.The upper layer is the detection layer.After performing the dimension reduction on the distribution features of the system state data,the anomaly detection method based on unsupervised learning is used for attack detection.The lower layer is the recovery layer.In order to solve the unobservable problem of the power system after delete the abnormal data,a predict compensation mechanism based on improved extreme learning machine is proposed to recover the state estimation results.IEEE standard test systems are employed in our simulation experiments.The experimental results prove the effectiveness of the proposed two methods.