Research On Defense Strategy Based On False Data Injection Attacks In Power System

The secure and reliable operation of the power system is crucial to the stability of a national economy.As the result of the high integration of the information and communication technologies,today’s power system becomes more and more vulnerable to cyber attacks.An attacker can disrupt the process of state estimation by intentionally injecting false data into certain measurements without being detected if the network information of the system is known to the attacker.Accordingly,a wrong or insecure dispatch or control signal might be made according to the corrupted data,which may result in severe consequences to a power system.This thesis combines the mechanism of false data injection attacks with machine learning algorithms,and then proposes two defense strategies against false data attacks through the two protection ideas of pre-prevention defense and post-event active defense for DC linear systems,which achieve the effective defense of FDIAs(False data Injection Attacks)and DDAs(Dummy data Attacks)and thus have certain innovation and practical value in the field of power cyber-physical security.To ensure the safe operation of power systems,this paper first investigates the attacking mechanism of false data and then studies the impacts of false data on economic and secure operation of the system.Then try to establish bilevel attack models to evaluate the impact of economy and safety.In order to solve this two-layer problem,this paper employs the KKT(Karush-Kuhn-Tucker)reformulation based method to translate the bilevel optimization into a single one by replacing the lower level problem with its optimality conditions.The simulation experiments on the IEEE14-bus system and IEEE 24-bus system conducted to validate the vulnerability of today’s power systems to cyber attacks.and thus highlight the necessity of defense strategy research.On this basis,this thesis conducts in-depth research on the defense strategies of traditional FDIAs and new DDAs.For traditional FDIAs,after the system is attacked,the corrective protection idea of active defense is used,and the GCNN(Gated Recurrent Unit with Convolutional Neural Network)joint network extracts the measurement features to identify the damaged measurement value and remove it.At the same time,According to the electrical,meteorological and date characteristics mined in historical data,the TCN(Time Convolutional Network)is used to generate data as similar as possible to the real loss data to complete the data,thereby restoring the true power grid operating state and eliminating the influence of traditional FDIAs,the effectiveness of this defense strategy has been verified in the IEEE 14-bus system.Aiming at the new type of high-concealment false data attacks DDAs that are difficult to detect,based on the idea of measuring physical protection in advance,we analyze the interaction between the defender and the attacker and establish a three-layer game defense model to obtain the optimal node protection scheme under the condition of limited enhanced protection budget.With the help of the concept of load node importance based on electrical betweenness and active power flow distribution,the multi-layer game model is transformed to three single layer,and finally the optimal node reinforcement that can against DDAs is obtained.The solution fundamentally inhibits the generation of DDAs,and the effectiveness of this defense strategy has been verified in the IEEE 39-bus system.