Research On Detection And Defense Methods Of False Data Injection Attack In Smart Grid

The power infrastructure is highly coupled with information and communication systems,making the power system more intelligent,but also more vulnerable to malicious cyber attacks.A malicious attacker can remotely connect to the internal core system of the smart grid through external network equipment and tamper with the grid state information.The false data injection(FDI)attacks,as one of the most important forms of attack on power systems in recent years,can bypass system detection and cause deviations in state estimation results,which can affect power generation,power flow optimization,power markets,and transmission and distribution.Electricity and other causes interference,which seriously harms the stable operation of the power grid.In order to solve the problems of measurement data integrity and protection state estimation in smart grids,the construction methods of FDI attacks in different application scenarios of smart grids are studied,and effective attack detection and active defense mechanisms are further proposed.First,under the traditional bad data detection(BDD)of smart grid,the construction principle of FDI attack is studied.Based on the interaction characteristics of the power grid and the information system under the DC and AC models,the method of constructing the FDI attack vector under non-complete topology information is studied,and the effectiveness is verified through simulation experiments.The probability distribution characteristics of the power system measurement changes in order to study the method of real-time detection and precise location of concealed FDI attacks.Finally,by analyzing the impact of FDI attacks on the continuous operation of state estimation in the power grid,an online active defense method for smart grids is proposed to ensure the stable operation of the system.Specific research contents include:(1)Proposed the attack vector construction algorithm with the incomplete topology informationA new method for constructing effective FDI attack vectors is proposed in the case of incomplete topology information.This method can construct the attack vector based on the incomplete topological information of the power grid without disturbing a large number of measured values to cause interference with the state estimation.First,the kernel independent component analysis method is used to transform the limited system topology information into the measurement data space to obtain the Jacobian matrix corresponding to the incomplete topology information,which is used to represent the current state of the power system.Tthe incomplete topology information corresp-onding to the the Jacobian matrix is used to construct a generating function model of the FDI attack vector,which will not be detected by the traditional BDD device.Finally,the lagrange function method is used to solve the vector generating function to obtain the attack vector.The proposed method can construct FDI attack vectors based on the Jacobian matrices with different sparsity corresponding to the completeness of different topological information.The experimental results prove that when the attack can only tamper with the limited measurement value,the attack vector constructed by this method still has a high success rate.(2)Proposed the real-time attack detection method combing the routing mechanism of the capsule network with the improved graph networkBy constructing a bi-directional graph model to represent the power system,the input is based on an improved graph networks(GN)detection algorithm,automatically learning and extracting measurements features,training parameters in the computation block structure,and classifying the measured values of the tested attack.Determine if or not there is any tampered measurement value in the measurement data.The experimental results show that under the condition that the topology of the power system is unchanged,the improved graph network detection mechanism based on a simple hierarchical structure is more efficient than standard neural networks,convolutional neural networks,and recursive neural networks with the same number of layers in the structure of the grid.The cost of updating caculation is smaller,and the tampered data in the measurement data can be identified more accurately.However,in many cases,the power system frequently changes in load configuration,power generation equipment,and system topology over time.Therefore,when the power system detects an attack,it is necessary to obtain the node location information of the attacked data source at the same time.Aiming at this problem,the routing mechanism in capsule networks is researched.Capsule network is combined with graph network,and Caps-GN algorithm is proposed.Capsule vector is used to save relevant attribute information in the power system,such as location,direction,and connection.Calculate the probability of the differential loss classification function,and output the classification result of the attack,so as to detect the FDI attack position.Experimental results show that compared with scalar-based graphic neural networks,vector-based capsule neural networks can provide more accurate localization functions.(3)Proposed the adaptive window on-line generation algorithm for physical network models to combat network defense attacksThis dissertation proposes an active defense method against FDI attacks.After detecting the attack and deleting the tampered measurement value,the non-tampered measurement data is reconstructed into the network physical operation model and input to the improved online generation game function in the adversarial network.The objective function takes turns to find the Nash equilibrium value,and the generation model generates measurement data that approximates the original value.Using the online active defense algorithm,the original measurement data is accurately recovered from the tampered measurement values,and the recovered data is transmitted to the state estimator to ensure the correct operation of the state estimation.In order to improve the speed of data generation and ensure the uninterrupted operation of state estimation,smooth gradient operation and adaptive window structure are used to improve the generation adversarial network algorithm and speed up the training process of the operation parameters in the algorithm.The improved online generation confrontation network can recover the initial measurement data from the incomplete measurement data after the FDI attack,and transmit the generated data to the state estimator,thereby effectively reducing the impact of the FDI attack on the power system.