Detection, Isolation and Elimination Methods for False Data Injection Attacks in Microgrid Systems

Abstract

The present invention provides a method for detecting, isolating and eliminating false data injection attacks in a microgrid system. The method includes: analyzing the load frequency control of a microgrid system involving electric vehicles affected by wind power generation fluctuations and load disturbances, and establishing a state equation of a dynamic model; analyzing the attack mechanism of false data injection attacks and establishing a mathematical model of the attacks; designing an HBF neural network observer to detect and isolate network attacks, and analyzing the boundedness of the estimation error of the observer; designing an H-infinity security controller to eliminate the negative impact of network attacks. The present invention overcomes the deficiencies of traditional methods that cannot quickly detect and self-repair, reduces the risk of malicious network attacks on complex microgrid systems, and improves the security of the operation of complex microgrid systems.

Description

Technical Field

[0001] The present invention relates to the technical field of information security of microgrid systems, and in particular, to a method for detecting, isolating, and eliminating false data injection attacks in a microgrid system. Background Art

[0002] Nowadays, complex microgrid systems are widely used in military sites, urban households, and large factories due to their advantages such as economy, low carbon, and easy expandability. In different types of microgrid systems, load frequency control (LFC) is effectively utilized to automatically maintain frequency stability during load disturbances and power generation fluctuations. In addition, in order to assist LFC in regulating voltage and frequency, vehicle-to-grid (V2G) has been successfully incorporated into the grid.

[0003] Generally, a complex microgrid system is a typical networked control system, and different components exchange information through a shared communication network. Compared with traditional power systems connected through dedicated communication networks, networked microgrid systems have the advantages of low cost, easy maintenance, and easy expansion. However, the widespread application of shared communication networks brings convenience to microgrids while also making them extremely vulnerable to the adverse effects of malicious attacks in the network. In recent years, some accidents have shown that the emergence of malicious attacks can easily cause various catastrophic damages such as data leakage, direct economic losses, and large-scale power outages. Therefore, in order to ensure the stable operation of microgrids and reduce economic losses, it is urgent to formulate corresponding strategies to timely detect, accurately isolate, and effectively mitigate malicious network attacks.

[0004] Existing technologies mainly detect false data injection attacks, and rarely involve the isolation and elimination of false data injection attacks. The main idea of detection is to compare the system output residual with a pre-set threshold based on the analysis of data consistency to complete the detection of network attacks. However, for new types of intelligent network attacks, such as stealth attacks, zero-dynamics attacks, and pole-dynamics attacks, they can not only damage the performance of power systems but also eliminate the impact of network attacks on the output of power systems, making the system output residual not have obvious abnormal characteristics, resulting in the difficulty of traditional residual-based detection methods in detecting such false data injection attacks.

[0005] Therefore, a method for detecting, isolating, and eliminating false data injection attacks in a complex microgrid system with the participation of electric vehicles is needed. Summary of the Invention

[0006] Aiming at the problem that traditional residual-based detection methods are difficult to detect new types of false data injection attacks, the purpose of the present invention is to provide a method for detecting, isolating, and eliminating false data injection attacks in a microgrid system with the participation of electric vehicles, which can reduce the risk of malicious network attacks on the microgrid system and improve the security of the microgrid system operation.

[0007] To solve the above problems, the technical solution of the present invention is as follows:

[0008] A method for detecting, isolating, and eliminating false data injection attacks in a microgrid system, comprising the following steps:

[0009] Analyze the load frequency control of a microgrid system with electric vehicles affected by wind power fluctuations and load disturbances, and establish the state equation of the dynamic model;

[0010] Analyze the attack mechanism of false data injection attacks and establish a mathematical model of the attacks;

[0011] Design an HBF neural network observer to detect and isolate network attacks, and analyze the boundedness of the observer estimation error;

[0012] Design an H-infinity security controller to eliminate the negative impact of network attacks.

[0013] Optionally, in the step of analyzing the load frequency control of a microgrid system with electric vehicles affected by wind power fluctuations and load disturbances and establishing the state equation of the dynamic model, the state equation of the dynamic model is:

[0014]

[0015] Where x(t) is the state variable,

[0016] x T (t) = [Δf(t), ΔP t (t), ΔP g (t), ΔP Ev1 (t), ΔP Ev2 (t)]; y(t) is the measured output, y T (t) = [Δf(t), ΔP Ev1 (t), ΔP Ev2 (t)]; Δf(t) represents the frequency deviation; ΔP t (t) represents the output power of the diesel generator; ΔP g (t) represents the governor valve position; ΔP Ev1 (t), ΔP Ev2 (t) represent the output powers of the first and second electric vehicle stations. u(t) is the control input; ΔP d (t) is the wind power fluctuation and load disturbance; A, B, H, C are matrices of appropriate dimensions.

[0017] Optionally, the matrices of A, B, H, C are:

[0018]

[0019] In the formula, T t represents the time constant of the diesel generator; H t represents the equivalent inertia constant; T g represents the governor time constant; R g represents the governor constant; T Ev1 and T Ev2 represent the time constants of the first and second electric vehicle stations.

[0020] Optionally, in the step of analyzing the attack mechanism of the false data injection attack and establishing a mathematical model of the attack, the false data injection attack destroys the data integrity by tampering with the data packets transmitted between different components of the microgrid, thereby disrupting the normal operation of the power system.

[0021] Optionally, in the step of analyzing the attack mechanism of the false data injection attack and establishing a mathematical model of the attack, the mathematical model of the false data injection attack is:

[0022]

[0023] In the formula, u i (t), respectively represent the i-th component of the normal control signal, the false data injected by the intelligent attacker, and the control signal received by the actuator. When a false data injection attack occurs, the model state equation is established as:

[0024]

[0025] In the formula, is the attacked system state vector, is the attack signal of the injection forward channel, is the measured output under attack.

[0026] Optionally, the step of designing an HBF neural network observer to detect and isolate network attacks and analyzing the boundedness of the observer estimation error specifically includes: designing an HBF neural network observer:

[0027]

[0028] In the formula, and respectively represent the observer state, the observer output, the output of the HBF neural network, and the observer gain to be designed.

[0029] Optionally, the step of designing an HBF neural network observer to detect and isolate network attacks and analyzing the boundedness of the observer estimation error specifically includes: Since the HBF neural network has a strong approximation ability, given an approximation error ε > 0, for any attack signal f at (t, xat ) The formula is:

[0030] f at (t, x at ) = W T Φ(x at , c, Ψ) + ε

[0031] Where W = [W1, W2,..., W n T is the ideal weight matrix, c = [c1, c2,..., c n T is the center of the hyper basis function, Ψ is the hyper basis function on the hidden layer neurons and The formula is:

[0032]

[0033] Where represents the similarity between x at and c i The attack signal f at (t, x at ) can be approximated as the formula:

[0034]

[0035] Define the state estimation error as The weight update error is The output estimation error is The estimation error dynamics equation is:

[0036]

[0037] Where is a bounded disturbance term, A L = A - LC.

[0038] Optionally, both the state estimation error and the weight update error are bounded, and the weight update method of the neural network is:

[0039] Optionally, the steps of designing the H-infinity security controller to eliminate the negative impact of network attacks specifically include: The H-infinity security controller based on the HBF neural network is: Where K is the controller gain to be designed, is the attack signal calculated based on the neural network, and the influence of the disturbance term η(t) on the state x at (t) is limited to:

[0040]

[0041] ​​where γ1 > 0, γ2 > 0, and γ3 > 0 are performance indicators.

[0042] Optionally, the steps for the designed H-infinity security controller to eliminate the negative impact of cyberattacks specifically include: The sufficient condition for the microgrid system to remain asymptotically stable when suffering from false data injection attacks is: For given parameters K, L, γ1 > 0, γ2 > 0, and γ3 > 0, when there exist proper-dimension matrices P1 > 0 and P2 > 0 such that:

[0043]

[0044] where

[0045]

[0046] Ξ2 = A T P2 + P2A - C T L T P2 - P2LC.

[0047] Then the microgrid system is asymptotically stable and has an H-infinity norm bound γ = γ1γ3 + γ2.

[0048] Compared with the prior art, the method for detecting, isolating, and eliminating false data injection attacks in the microgrid system of the present invention fully considers the wind power generation fluctuations and load disturbances that may occur in the microgrid system, designs an HBF neural network observer to accurately estimate the internal state of the complex power grid, fully utilizes the approximation ability of the neural network to online reconstruct the possible attack signals, and at the same time designs an H-infinity security controller to timely eliminate the negative impact of cyberattacks, realizing the integrated design of the algorithm for detecting, isolating, and eliminating false data injection attacks in the complex microgrid system participated by electric vehicles, effectively reducing the risk of the complex microgrid system suffering from malicious cyberattacks, and improving the security and self-repair ability of the complex microgrid system operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] By reading the detailed description of the non-restrictive embodiments with reference to the following drawings, other features, objectives, and advantages of the present invention will become more apparent:

[0050] Figure 1 is the flowchart of the method for detecting, isolating, and eliminating false data injection attacks in the microgrid system provided by the embodiment of the present invention;

[0051] Figure 2 is the load frequency control structure diagram of the microgrid system provided by the embodiment of the present invention;

[0052] Figure 3The estimation effect diagram of the attack signal in the first four communication channels by the neural network provided in the embodiment of the present invention;

[0053] Figure 4a The estimation effect diagram of the attack signal in the fifth communication channel by the neural network provided in the embodiment of the present invention;

[0054] Figure 4b The comparison diagram of the elimination effects of the false data injection attack by the method of the present invention and the traditional method provided in the embodiment of the present invention. Detailed implementation manners

[0055] The present invention will be described in detail below with reference to specific embodiments. The following embodiments will help those skilled in the art to further understand the present invention and make the technical solutions and their beneficial effects of the present invention obvious.

[0056] Specifically, Figure 1 The flowchart of the method for detecting, isolating and eliminating false data injection attacks in the microgrid system provided in the embodiment of the present invention is as Figure 1 shown, and the method includes the following steps:

[0057] S1: Analyze the load frequency control of the microgrid system participated by electric vehicles affected by wind power generation fluctuations and load disturbances, and establish the state equation of the dynamic model;

[0058] Specifically, the state equation of the dynamic model of the load frequency control (LFC) system participated by electric vehicles can be constructed by the following method:

[0059] According to Figure 2 the power flow and transfer function model shown, it is deduced that:

[0060]

[0061] In the formula: Δf(t) represents the frequency deviation; ΔP t (t) represents the output power of the diesel generator; ΔP g (t) represents the governor valve position; ΔP Ev1 (t), ΔP Ev2 (t) represent the output powers of the first and second electric vehicle stations; ΔP d (t) is the wind power generation fluctuation and load disturbance; Δu1, Δu2 and Δu3 are generated by the load frequency control and are used as the inputs of the electric vehicles and the diesel generator. T t represents the diesel generator time constant; H t represents the equivalent inertia constant; T g represents the governor time constant; R g represents the governor constant; T Ev1 , T Ev2Denote the time constants of the first and second electric vehicle charging stations. By the inverse Laplace transform, Equation (1) can be written as:

[0062]

[0063] Define the state vector x T (t) = [Δf(t), ΔP t (t), ΔP g (t), ΔP Ev1 (t), ΔP Ev2 (t)]; Measure the output vector y T (t) = [Δf(t), ΔP Ev1 (t), ΔP Ev2 (t)]; Wind power fluctuations and load disturbances

[0064] ΔP d (t) = ΔP l (t) - ΔP wind (t), The control input u T (t) = [Δu1(t), Δu2(t), Δu3(t)]. The state equation of the dynamic model is obtained as:

[0065]

[0066] Where the matrices of A, B, H, and C are:

[0067]

[0068] S2: Analyze the attack mechanism of false data injection attacks and establish a mathematical model of the attack;

[0069] Specifically, false data injection attacks are a typical malicious attack that disrupts data integrity by tampering with the data packets transmitted between different components of the microgrid, thereby disrupting the normal operation of the power system. To successfully implement the attack, an intelligent attacker needs to possess the following two capabilities: one is to interfere with the normal operation of the microgrid, and the other is to successfully evade the detection mechanism of the microgrid. This destructiveness and concealment make false data injection attacks even more lethal.

[0070] Generally speaking, false data injection attacks occurring in the forward channel can be either state-independent or state-dependent, and they can usually be modeled in the following form:

[0071]

[0072] Where u i (t), They respectively represent the $i$-th component of the normal control signal, the false data injected by the intelligent attacker, and the control signal received by the actuator. Based on this, the model state equation when suffering from false data injection attack can be established:

[0073]

[0074] In the formula, is the attacked system state vector, is the attack signal on the injection forward channel, is the attacked measurement output.

[0075] S3: Design an HBF neural network observer to detect and isolate network attacks, and analyze the boundedness of the observer estimation error;

[0076] The designed HBF neural network observer is:

[0077]

[0078] In the formula, and respectively represent the observer state, the observer output, the output of the HBF neural network, and the observer gain to be designed.

[0079] Since the HBF neural network has a strong approximation ability, given the approximation error $\varepsilon>0$, for any attack signal $f$ at (t, $x$ at ) can be written in the following form:

[0080] $f$ at (t, $x$ at ) = $W$ T $\varPhi(x$ at , $c$, $\varPsi$) + $\varepsilon$

[0081] In the formula, $W$ = [W1, W2,...., W n T is the ideal weight matrix and $c$ = [c1, c2,..., c n T is the center of the super basis function, $\varPsi$ is the super basis function on the hidden layer neurons and it can usually be selected in the following form:

[0082]

[0083] In the formula, represents the similarity between $x$ at and $c$ i . Therefore, the attack signal $f$ at (t, $x$ at ) can be approximated in the following form:​​

[0084]

[0085] Define the state estimation error as The weight update error is The output estimation error is The estimation error dynamics equation is:

[0086]

[0087] Where, is a bounded disturbance term, A L = A - LC

[0088] Furthermore, both the state estimation error and the weight update error are bounded. The weight update method of the neural network is as follows:

[0089]

[0090] Where ρ1 and ρ2 are the learning rate and the damping coefficient respectively. Next, analyze the boundedness of both the state estimation error and the weight update error.

[0091] Select a positive definite Lyapunov function Where P is a positive definite matrix and satisfies (Q is a positive definite matrix). By taking the derivative of V, we can obtain Define And combining equations (10) and (11), we can get:

[0092]

[0093] Substitute the following inequality into (12)

[0094]

[0095] We can obtain the following expression:

[0096]

[0097] Then define l1 = 0.5|δ| and We can get:

[0098]

[0099] Therefore, when And When, is negative definite. That is, the state estimation error e at (t) and the weight update error are bounded, which means that the HBF neural network accurately detects and isolates the attack signal.

[0100] S4: Design an H-infinity security controller to eliminate the negative impact of cyberattacks.

[0101] The H-infinity security controller based on the HBF neural network is:

[0102]

[0103] where K is the controller gain to be designed, is the attack signal calculated based on the neural network. Then the physical dynamics model (5) and the error dynamics (10) can be written as:

[0104]

[0105] where is the actual approximation error of the HBF neural network. It is easy to know that where is the space of square-integrable functions with the Euclidean norm. Next, the influence of the perturbation term η(t) on the state x at (t) is restricted as:

[0106]

[0107] where γ1 > 0, γ2 > 0, and γ3 > 0 are performance indicators.

[0108] Furthermore, the sufficient condition for the microgrid system to remain asymptotically stable under false data injection attacks is:

[0109] For the given parameters K, L, γ1 > 0, γ2 > 0, and γ3 > 0, when there exist proper-dimensional matrices P1 > 0 and P2 > 0 such that:

[0110]

[0111] where

[0112]

[0113] Ξ2 = A T P2 + P2A - C T L T P2 - P2LC.

[0114] Then the complex microgrid system is asymptotically stable and has an H-infinity norm bound γ = γ1γ3 + γ2, that is, the security controller (13) effectively eliminates the negative impact of false data injection attacks.

[0115] Prove the above sufficient condition by constructing a Lyapunov function and (P1 > 0, P2 > 0), from equations (14) and (15), we can obtain:

[0116]

[0117] In the formula, Ξ 11 = Ξ1 + I, Ξ 21 = Ξ2 + I, and Using the Schur complement lemma, it can be known that if conditions (18) and (19) hold and J1 < 0 and J2 < 0 are also satisfied, then equations (16) and (17) will be satisfied simultaneously.

[0118] Transform the above sufficient conditions. For given parameters γ1 > 0, γ2 > 0, and γ3 > 0, when there exist matrices P1 > 0, P2 > 0, Y, and S with appropriate dimensions such that:

[0119]

[0120] Then the system (14) is asymptotically stable and has an H-infinity norm bound γ = γ1γ3 + γ2. The control gain can be calculated as The observer gain where

[0121] Ξ3 = AP1 + P1A T - BY - Y T B T ,

[0122] Ξ4 = A T P2 + P2A - C T S T - SC.

[0123] Define Y = KP1, S = P2L, and it is easy to obtain conditions (20) and (21) from equations (18) and (19).

[0124] The following provides a specific simulation experiment to verify the method of the present invention:

[0125] A complex microgrid system involving an electric vehicle, and the parameters of the system are shown in the following table:

[0126] <![CDATA[T t > <![CDATA[H t > <![CDATA[T g > <![CDATA[R g > <![CDATA[T Ev1 > <![CDATA[T Ev2 > 8 7.11 0.1 2.5 1 1

[0127] Assume that the initial state of the system is x(0) = [0.1; 0.1; 0.2; 0.2; 0.3], take ρ1 = 3×10 9 , ρ2 = 0.5. Obtain the observer gain L and the control gain K by solving the LMIs (20) and (21).

[0128] Assuming that the attacker launches an attack on the fifth channel at the 10th second, the actual attack signal is as follows:

[0129]

[0130] in The output of the neural network and the state response of the microgrid are as follows: Figure 3 and Figure 4a , Figure 4b As shown, from Figure 3 and Figure 4a It can be seen that the fifth output of the HBF neural network It becomes non-zero from the 10th second and approaches the real attack signal with higher accuracy But the first four outputs and That is to say, the HBF neural network designed in this embodiment successfully detects and reconstructs the attack signal and locates it in the fifth channel. By comparing the detection effect of the BP neural network, it is obvious that the HBF neural network in this embodiment is reconstructed faster and more accurately, which will help to mitigate the negative impact of the false data injection attack. Figure 4b It can be seen that compared with the traditional mitigation scheme, the mitigation scheme proposed in this paper can more effectively alleviate the negative impact of false data injection attacks, that is, the complex microgrid can still maintain stable operation under the false data injection attack.

[0131] The above simulation experiments show that, considering the fluctuation of wind power generation and load disturbance, the HBF neural network in the present invention can reconstruct possible attack signals more accurately and quickly. At the same time, the estimated complex microgrid state and attack signal will be used in the design of the H infinity safety controller, that is, the solution in this embodiment can detect and locate network attacks more quickly and accurately, and more effectively eliminate the negative impact of network attacks.

[0132] Compared with the prior art, the method for detecting, isolating and eliminating false data injection attacks in a microgrid system of the present invention fully considers the wind power generation fluctuations and load disturbances that may occur in the microgrid system, designs a HBF neural network observer to accurately estimate the internal state of a complex power grid, fully utilizes the approximation capability of the neural network to reconstruct possible attack signals online, and designs an H-infinity security controller to timely eliminate the negative impact of network attacks, thereby realizing the integrated design of the detection, isolation and elimination algorithms for false data injection attacks in a complex microgrid system involving electric vehicles, effectively reducing the risk of the complex microgrid system being subjected to malicious network attacks, and improving the safety and self-repairing capability of the operation of the complex microgrid system.

[0133] The specific embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the above specific embodiments, and those skilled in the art can make various changes or modifications within the scope of the claims, which does not affect the essence of the present invention. Without conflict, the embodiments of the present application and the features in the embodiments can be combined with each other arbitrarily.

Claims

1. A method for detecting, isolating, and eliminating false data injection attacks in a microgrid system, characterized in that, The method includes the following steps: Analyze the load frequency control of a microgrid system with electric vehicles affected by wind power generation fluctuations and load disturbances, and establish the state equation of a dynamic model; Analyze the attack mechanism of false data injection attacks and establish a mathematical model of the attacks; Design an HBF neural network observer to detect and isolate network attacks, and analyze the boundedness of the observer estimation error. Specifically, it includes: Design an HBF neural network observer: wherein, and respectively represent the observer state, the observer output, the output of the HBF neural network, and the observer gain to be designed. u(t) is the control input, and A, B, H, and C are matrices of appropriate dimensions. is the attacked measurement output; Given an approximation error ε > 0 and any attack signal f at (t, x at ) is given by: f at (t, x at ) = W T Φ(x at , c, Ψ) + ε where W = [W1, W2,..., W n T is the ideal weight matrix, c = [c1, c2,..., c n T is the center of the hyper basis function, Ψ is the hyper basis function on the hidden layer neurons and the formula is:​​ In the formula, represents x at and c i The similarity between them, the attack signal f at (t, x at ) can be approximated by the formula: Define the state estimation error as Define the weight update error as Define the output estimation error as The estimation error dynamics equation is as follows: wherein, is a bounded disturbance term, A L = A - LC; ΔP d (t) is the wind power generation fluctuation and load disturbance; Design an H-infinity security controller to eliminate the negative impact of network attacks.

2. The method for detecting, isolating and eliminating false data injection attacks on the microgrid system according to claim 1, characterized in that, In the step of analyzing the load frequency control of a microgrid system with electric vehicles affected by wind power generation fluctuations and load disturbances and establishing the state equation of a dynamic model, the state equation of the dynamic model is: Where x(t) is the state variable, x T (t) = [Δf(t), ΔP t (t), ΔP g (t), ΔP Ev1 (t), ΔP Ev2 (t)]; y(t) is the measured output, y T (t) = [Δf(t), ΔP Ev1 (t), ΔP Ev2 (t)]; Δf(t) represents the frequency deviation; ΔP t (t) represents the output power of the diesel generator; ΔP g (t) represents the governor valve position; ΔP Ev1 (t), ΔP Ev2 (t) represent the output powers of the first and second electric vehicle stations, u(t) is the control input; A, B, H, C are matrices of appropriate dimensions.

3. The method for detecting, isolating and eliminating false data injection attacks in the microgrid system according to claim 2, characterized in that, The matrices of A, B, H, and C are: where T t represents the diesel generator time constant; H t represents the equivalent inertia constant; T g represents the governor time constant; R g represents the governor constant; T Ev1 , T Ev2 represent the time constants of the first and second electric vehicle stations.

4. The method for detecting, isolating and eliminating false data injection attacks on the microgrid system according to claim 1, characterized in that, In the step of analyzing the attack mechanism of false data injection attacks and establishing a mathematical model of the attacks, the false data injection attack is to damage data integrity by tampering with data packets transmitted between different components of the microgrid.

5. The method for detecting, isolating and eliminating false data injection attacks on a microgrid system according to claim 4, characterized in that, In the step of analyzing the attack mechanism of false data injection attacks and establishing a mathematical model of the attacks, the mathematical model of the false data injection attack is: where \(u\) i (t), represent the \(i\)-th component of the normal control signal, the false data injected by the intelligent attacker, and the control signal received by the actuator respectively. When suffering from false data injection attacks, the model state equation is established as follows: In the formula, is the system state vector under attack, is the attack signal injected into the forward channel.

6. The method for detecting, isolating and eliminating false data injection attacks on the microgrid system according to claim 1, characterized in that Both the state estimation error and the weight update error are bounded. The weight update method of the neural network is as follows: where ρ1 and ρ2 are the learning rate and the damping coefficient respectively.

7. The method for detecting, isolating and eliminating false data injection attacks on the microgrid system according to claim 1, characterized in that The steps for the described designed H-infinity security controller to eliminate the negative impact of cyberattacks specifically include: The H-infinity security controller based on the HBF neural network is as follows: In the formula, K is the controller gain to be designed, is the attack signal calculated based on the neural network. The influence of the disturbance term η(t) on the state x at (t) is limited to: where γ1 > 0, γ2 > 0 and γ3 > 0 are performance indicators.

8. The method for detecting, isolating, and eliminating false data injection attacks in the microgrid system according to claim 7, characterized in that, The step of designing an H-infinity security controller to eliminate the negative impact of network attacks specifically includes: The sufficient condition for the microgrid system to remain asymptotically stable when suffering from false data injection attacks is: For given parameters K, L, γ1>0, γ2>0, and γ3>0, when there exist appropriate dimension matrices P1>0 and P2>0 such that: Where, Ξ2 = A T P2 + P2A - C T L T P2 - P2LC Then the microgrid system is asymptotically stable and has an H-infinity norm bound γ = γ1γ3 + γ2.

Images