Research On Data-Driven Detection And Recovery Of Measurement Data Attack In Power Grid

The development and construction of modern power grids is accompanied by a deep integration with various advanced information and communication technologies.The threat of various cyber-attacks has also become more prominent,such as false data injection attack(FDIA)against data integrity is a representative covert type of attack.By affecting the integrity and reliability of state estimation,FDIA has the potential to cause serious consequences such as grid paralysis.Therefore,it is significance to study the FDIA detections and develop countermeasures to mitigate or even eliminate the attack impacts to ensure the security and stability of grid operation.However,traditional detection methods are difficult to deal with such attacks.Current methods for eliminating attacks are mainly including two research types,pre-attack defense and post-attack repair,while the defense methods for FDIA tend to have high requirements on the infrastructure of grids.Therefore,this paper focuses on the data-driven detection and restoration of measurement data attacks on power grids.Firstly,considering that power measurement data is essentially a non-smooth signal with slight noise characteristics,a noise removal method based on empirical modal decomposition(EMD)of measurement data is studied.EMD is used to process the measurement data and screen them to retain the data strongly correlated components to provide a basis for subsequent attack detection.Then the effectiveness of method is verified based on IEEE 14-bus standard test system.Secondly,a FDIA detection method based on Hellinger distance is proposed by combining the dynamic correlation of adjacent measurement data.The log transformation and antilog transformation are used to enhance the probability distribution characteristics of measurement data variations.The difference between the probability distributions is further calculated based on Hellinger distance.By comparing the magnitude of Hellinger distance at the attacked time with the threshold value set based on the confidence interval to detect FDIA.Then the superiority of the proposed detection method is verified by simulation research.Finally,considering the correlation between normal measurement data and between their historical variation patterns,a strategy based on unsupervised learning to cope with FDIA is investigated.The resistance capability of power grid after the attack is enhanced by introducing the grid topology as the label and proposing an improved generative adversarial net(GAN)based on repair of the attack-affected measurement data.Then the effectiveness of the restoration method is verified by simulation results.