Research On Detecting False Data Injection Attacks In Power System

With the wide application of information technology and communication technology in power grid,the vulnerability of power system is becoming more obvious,and it is also more vulnerable to the threat from cyber-attacks.The false data injection attacks(FDIAs)are typical cyber-attacks since they can bypass traditional bad data detection mechanisms and cause erroneous estimation on system states which seriously threatens the stable operation of the power system.In order to accurately detect such attacks and increase the reliability of power systems,three detection strategies are proposed in this paper,and the specific research works are as follows:(1)Aiming at the problem of low discrimination between normal and attack measurements,a new dimensionality-reduction method is proposed.Based on the performance index of the minimum classification error,the orthonormal basis with a reduced dimension is constructed,which makes the two classes measurements distributed in different regions with a greater degree of discrimination in the new space.By verifying that the loss function is independent of the vector length,a set of optimization problems are constructed,which use penalty terms to constrains the orthogonality relation between bases.Then,the Gaussian mixture model detector is fitted by the expectation maximization algorithm,and the detection threshold is searched by the semi-supervised algorithm.Finally,the real world load data are used to generate measurements to verify the effectiveness of the proposed method.(2)Aiming at the problem that the blindness of orthogonal constraints will produce redundant dimensionality,a dimensionality-reduction algorithm based on weighted strategy is proposed.This strategy will put more emphasis on the measurements with higher degree of confusion,which reduces the generation of redundant dimensionality,and makes dimensionality adaptive selection.Then,aiming at the problem that most data-driven methods are only applicable to specific systems,the Euclidean norm detection is used to replace the traditional machine learning-based classifier,and the high-precision detection of FDIAs is realized by tracking the dynamic characteristics of the measurements in the new space.Finally,the accuracy and robustness of the proposed method are verified in the normal case,the case of considering topology changes and the case of integration of renewable energy.(3)Focusing on the problem that the data-driven methods occupy a large storage space and have a high computational complexity,the network communication coding mechanism is utilized to build the FDIAs detection mechanism based on the coding strategy,which aims at the DC linear model and AC nonlinear model of the power grid,respectively.For the DC linear system,a data-driven encoding and decoding scheme is designed,and the optimal data assignments of the encoding and decoding matrix are solved according to whether there are communication constraints or not.For the AC nonlinear model,a stretching matrix is constructed to make the attacked redundant measurements produce large detection residues.Then the corresponding encoding and decoding schemes are put forward,and the conditions for the selection of the decoding matrix are proposed.Finally,the real world load data in IEEE-14 bus system are used to verify the effectiveness of the proposed methods for the DC linear and AC nonlinear systems respectively.