Power system data driving damping controller under false data injection attack

Abstract

The invention discloses an electric power system data driving damping controller under false data injection attack, and relates to the technical field of novel electric power system control and network security, and the electric power system data driving damping controller comprises the following steps: S1, aiming at a wide area measurement signal of a damping controller, determining a condition that an attack signal reaches system stability destroying, and constructing multiple types of false data injection models; s2, designing an attack detection mechanism and an attack compensation module based on a threshold value according to the attacked controller measurement signal; s3, establishing a data driving model corresponding to the controlled system under the action of attack compensation; according to the method, the output signal of the controller is updated and the output of the controlled power system at the next moment is predicted by using the online I / O data of the controlled power system, so that the method can adapt to various operation conditions and is high in self-adaptability; the method has the capability of resisting false data injection attacks, the influence of the communication risk of network attacks on the system stability can be avoided, and the robustness of the controller is improved.

Description

Technical Field

[0001] The invention relates to the technical field of novel power system control and network security, and in particular to a data-driven damping controller for a power system under false data injection attacks. Background Art

[0002] In the context of promoting the construction of a new power system and supporting green power cooperation, new energy industries represented by wind power have developed rapidly. However, with the increase in large-scale interconnected power systems and the increase in installed capacity of wind farms, low-frequency oscillations between regions have occurred frequently, threatening the safe and stable operation of the power system. Double-fed induction generator (DFIG) wind turbines have become an effective device for additional damping control of interval oscillations due to their energy-saving and low-carbon, independent power control, and low operation and maintenance costs.

[0003] Existing DFIG-based additional damping controllers are divided into two categories: model-based and data-driven. Model-driven ones mostly rely on accurate modeling of the power system dynamics, and the stability and robustness of the controller are often affected by this. In the actual operation of the system, the fan control process is complex, and the DFIG and its control model provided by the manufacturer are mostly black box models, which makes the model-driven DFIG damping controller face high uncertainty challenges. In order to solve the problem of model-driven, many scholars have proposed data-driven methods that only require the input and output data of the controlled system, such as heuristic dynamic programming algorithms, model-free adaptive control technology, etc. to design additional damping controllers. The above method only uses the real-time I / O data of the controlled system, does not require additional internal information, has a simple structure, and has low computational complexity. Among them, damping controllers based on model-free adaptive control algorithms have been widely used. However, this type of controller relies on a wide area measurement system (WAMS) to provide remote signals as its input signals. Since WAMS contains a large number of sensors and communication networks, it is vulnerable to various network threats, thereby blocking the signal transmission of the communication network or injecting false data into it. The attacked signal will cause the additional damping controller to make wrong decisions, and its impact is similar to the low-frequency oscillation caused by a short-circuit fault in the system.

[0004] Among them, some scholars have considered the uncertainty of communication networks in wide-area damping control, such as time delay, packet loss, and communication failure, and only a few studies have considered the impact of network attacks. Compared with time delay, packet loss, or communication failure, network attacks are malicious and secretive. Therefore, the methods developed to deal with time delay, packet loss, or communication failure are not suitable for dealing with these attacks. For example, all these methods lack a detection mechanism for attacks, and even if an attack occurs, it will not be noticed. For the zone oscillation of the wind turbine-containing power system under network attack, how to use the real-time input and output data of the controlled system to design network attack detection and compensation methods, and realize the effective damping of the oscillation by the system after compensation, so as to ensure the safe and stable operation of the power system under the attack of false data injection. Summary of the invention

[0005] The purpose of the present invention is to solve the problems mentioned in the above background and to propose a data-driven damping controller for a power system under false data injection attacks.

[0006] The object of the present invention can be achieved by the following technical scheme: A data-driven damping controller for a power system under a false data injection attack, comprising a damping controller installed on the side of a wind turbine converter, and a control method of the damping controller comprising:

[0007] S1: For the wide-area measurement signal of the damping controller, determine the conditions under which the attack signal can destroy the system stability and build various types of false data injection models;

[0008] S2: Design a threshold-based attack detection mechanism and attack compensation module for the measured signal of the damping controller under attack;

[0009] S3: establishing a data-driven model corresponding to the controlled system under attack compensation, wherein the data-driven model represents the relationship between the output signal of the power system and the control signal of the connected wind turbine through pseudo partial derivative matrix parameters and control laws;

[0010] S4: The generated control signal is transmitted to the control circuit of the wind turbine, so that the corresponding wind turbine control circuit controls the wind turbine according to the control signal, so as to achieve low-frequency oscillation damping of the wind turbine power system including the network attack.

[0011] As a preferred embodiment of the present invention, the conditions for the attack signal to destroy stability include:

[0012] S11: The power system with wind turbine is represented as a discrete nonlinear model with single input and single output, including the control input u(k) of the damping controller, the wide-area measurement signal y(k) of the damping controller and the bounded disturbance d(k); where:

[0013] y(k+1)=f(y(k),y(k-1),...,y(kn y ),

[0014] u(k),u(k-1),...,u(kn u ),

[0015] d(k),d(k-1),...,d(kn d ))

[0016] S12: For the wide-area measurement signal y(k) of the damping controller, the attacker satisfies the following conditions: obtain the state information in the transmission network to construct the false data injection attack model y a (k), attack strategy Υy(k)∈{0,1}, when the channel is attacked at time k, we get Υ y (k) = 1, otherwise Υ y (k) = 0, the output of the attacked system is expressed as Since the attacker's ability is limited, there must be an upper limit for the attack signal, which can be expressed as

[0017] S13: Select the speed difference of the synchronous machine as the input signal of the damping controller. During the transmission of the input signal, the attacker invades the communication network to intercept the signal and inject false data, including parameter attack, random attack, data replay attack and data missing attack types. The attack signal y a (k) Tampering with the actual measured value y(k) causes the damping controller to perceive incorrect state information, thereby generating incorrect control commands; the established false data injection attack model is:

[0018] (1) Parameter attack:

[0019]

[0020] Where A represents the amplitude of the attack signal, e ζ ·sin(ω s t) represents the damped sine wave associated with the system oscillation mode, ζ is the damping ratio of the dominant oscillation mode, ω s is the natural oscillation frequency of the system, and the variable y a (t) represents the injected attack signal, t represents the current time, τ a Indicates the duration of the attack;

[0021] (2) Random attack:

[0022]

[0023] Among them, α and β are the minimum and maximum values ​​in the input signal range of the damping controller, respectively, and θ is a variable uniformly distributed on [0, 1], that is, θ~Uniform(0,1). The upper and lower limits of the real system output are used as constraints to model the random attack signal.

[0024] (3) Data Missing Attack:

[0025]

[0026] (4) Data replay attack:

[0027]

[0028] Among them, tt 0 Indicates the time at which data is to be recorded.

[0029] As a preferred embodiment of the present invention, the specific process of designing a threshold-based attack detection mechanism and an attack compensation module is as follows:

[0030] S21: The attacker chooses to inject false data into any measurement channel of the damping controller input signal; detect whether the attack exists, and the threshold detection for FDIA:

[0031]

[0032] Among them, Indicates the transmission signal of the measurement channel attacked by FDI, y p (k) is the predicted control input calculated by the damping controller using the existing I / O data, γ is the attack detection threshold; when Υ y When (k) = 0, it means that the control input has not been attacked or the attack impact is negligible. Otherwise, it means that the control input has been attacked by the network. At this time, the transmitted signal is unreliable. In the threshold detection, the threshold γ is set according to the prediction error, that is, γ min ≤|y(k)-y p (k)|;

[0033] S22: After the attack signal is identified, the attacked signal needs to be compensated to restore the authenticity of its data. The attack compensation module is as follows:

[0034]

[0035] in, represents the last unattacked transmission signal value before time k, or represents Υ y (kl)=0;l∈{1,2,3…,l M}; After the input signal of the damping controller completes the attack detection and compensation in the "attack detection and attack compensation" link, the damping controller uses the attack-compensated data to update the input of the controlled system and predict the output of the controlled system at a certain time later.

[0036] As a preferred embodiment of the present invention, the data-driven model includes a pseudo partial derivative matrix estimator, a control law based on a model-free adaptive control algorithm, and a predicted output value of the controlled system at the next moment; the data-driven model corresponding to the controlled system under the attack compensation is established specifically as follows:

[0037] S31: The controlled power system is transformed into:

[0038]

[0039] u(k),u(k-1),...,u(kn u ),

[0040] d(k),d(k-1),...,d(kn d ))

[0041] When the controlled power system satisfies the assumption and when ||ΔU(k)||≠0, for any length L, there exists a bounded time-varying matrix Φ∈R L , called the pseudo partial derivative matrix, makes the transformed data-driven model:

[0042]

[0043] in:

[0044] ΔU(k)=[Δu(k),Δu(k-1),...,Δu(k-L+1)] T

[0045] Δu(k)=u(k)-u(k-1)

[0046] Φ(k)=[φ 1 (k),φ 2 (k),...,φ L (k)] T ;

[0047] S32: Online estimation of Φ(k) is performed based on the I / O data of the controlled power system, and the pseudo partial derivative estimator is:

[0048]

[0049] in is the pseudo partial derivative vector estimate, is the tracking error of the controlled system output, Γ(k)=η(μ+||ΔU(k-1)|| 2 ) -1 , weighting factor μ>0, step size factor η∈(0,2], if the pseudo partial derivative at the current moment is less than or equal to the set value, or the difference between the control signal at the previous moment and the previous moment is less than or equal to the set value, then the pseudo gradient at the current moment is reset to the set initial value of the pseudo gradient;

[0050] S33: Set the expected value y of the output signal of the controlled power system * (k) = 0. After the system recovers to stability, its control signal u(k) is 0. The output signal of the damping controller at the next moment is obtained according to the data-driven model of the controlled power system:

[0051]

[0052] Among them, a linearization step vector [ρ 0 , ρ 1 , ..., ρ L ] T , ρ i >0, if i<1 or i>L,

[0053] S34: Pseudo-partial derivative based estimator The input signal of the damping controller at the current moment And the ΔU(k) ​​of the damping controller sliding time window, the output of the controlled power system at the next moment is obtained:

[0054]

[0055] Among them, y p (k+1) is the predicted value of the next moment k+1 calculated by the damping controller based on historical data and the data-driven model.

[0056] Compared with the prior art, the present invention has the following beneficial effects:

[0057] 1. The damping controller of the present invention does not require the internal signal of the system, has a simple structure, low calculation amount, and is not affected by the accuracy of system modeling. The damping controller is installed on the inverter side of the fan, and the cost is low.

[0058] 2. The present invention utilizes the online I / O data of the controlled power system to update the output signal of the damping controller and predict the output of the controlled power system at the next moment, and can adapt to various operating conditions with strong adaptive performance.

[0059] 3. The present invention has the ability to resist false data injection attacks, can avoid the impact of network attack communication risks on system stability, and improve the robustness of the damping controller.

[0060] 4. The data-driven damping controller of the present invention has a relatively complete theoretical proof under some practical assumptions, which can ensure the monotonic convergence of the closed-loop system tracking error and the bounded input-bounded output stability under the action of network attacks. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] In order to facilitate understanding by those skilled in the art, the present invention is further described below with reference to the accompanying drawings.

[0062] Figure 1 This is a flow chart of a control method of an interval low-frequency oscillation data driven damping controller under false data attack in this embodiment;

[0063] Figure 2 Block diagram of data-driven damping controller for power system under false data injection attack;

[0064] Figure 3 It is the schematic diagram of wind speed of fan;

[0065] Figure 4a Schematic diagram of CWADC damping level under parameter attack;

[0066] Figure 4b Schematic diagram of MFAC damping level under parameter attack;

[0067] Figure 4c Schematic diagram of CAD-MFAC damping level under parameter attack;

[0068] Figure 5a Schematic diagram of CWADC damping level under random attack;

[0069] Figure 5b Schematic diagram of MFAC damping level under random attack;

[0070] Figure 5c Schematic diagram of CAD-MFAC damping level under random attack;

[0071] Figure 6a Schematic diagram of CWADC damping level under data missing attack;

[0072] Figure 6b Schematic diagram of MFAC damping level under data missing attack;

[0073] Figure 6c Schematic diagram of CAD-MFAC damping level under data missing attack;

[0074] Figure 7a Schematic diagram of CWADC damping level under data replay attack;

[0075] Figure 7b Schematic diagram of MFAC damping level under data replay attack;

[0076] Figure 7c Schematic diagram of CAD-MFAC damping level under data replay attack. DETAILED DESCRIPTION

[0077] The technical solution of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0078] Example 1

[0079] See also Figure 1 - Figure 3 As shown, a data-driven damping controller for a power system under a false data injection attack includes a damping controller installed on a wind turbine converter side, and a control method of the damping controller includes:

[0080] S1. For the wide-area measurement signal of the damping controller, determine the conditions under which the attack signal can destroy the system stability and build various types of false data injection models;

[0081] S2. Design a threshold-based attack detection mechanism and attack compensation module for the controller measurement signal under attack;

[0082] S3, establishing a data-driven model corresponding to the controlled system under attack compensation, wherein the data-driven model represents the relationship between the output signal of the power system and the control signal of the connected wind turbine through pseudo partial derivative matrix parameters and control laws;

[0083] S4. The generated control signal is transmitted to the control circuit of the wind turbine, so that the corresponding wind turbine control circuit controls the wind turbine according to the control signal, so as to achieve low-frequency oscillation damping of the wind turbine power system caused by network attacks.

[0084] In step S1, a false data injection attack model is established when the measurement signal of the controller is injected, so that the attack signal reaches the condition of destroying stability. It includes:

[0085] S11. The power system containing wind turbines can be represented as a discrete nonlinear model with single input and single output, including the control input u(k) of the damping controller, the measurement signal y(k) of the damping controller, and the bounded disturbance d(k):

[0086] y(k+1)=f(y(k),y(k-1),...,y(kn y ),

[0087] u(k),u(k-1),...,u(kn u ),

[0088] d(k),d(k-1),...,d(kn d ))

[0089] In addition to satisfying the attack signal assumptions described in S1, the system also needs to satisfy the following assumptions: (1) The system is observable and controllable; (2) Except for a finite time, the partial derivatives of the function f(·) with respect to u(k), u(k-1), …u(k-L+1) are continuous; (3) Except for a finite time, the system (1) satisfies the generalized Lipschitz condition: |y(k 1 +1)-y(k 2 +1)|≤c||U L (k 1 )-U L (k 2 )||, where c is a positive constant.

[0090] S12. For the controller’s wide-area measurement signal y(k), the attacker needs to meet certain conditions to achieve the purpose of destroying the system stability. The following assumptions need to be made: the attacker understands the network security detection mechanism and can obtain the status information in the transmission network to construct the false data injection attack model y a (k), attack strategy Υ y (k)∈{0,1}. When the channel is attacked at time k, it can be known that Υ y (k) = 1, otherwise Υ y (k) = 0. Therefore, the output of the attacked system can be expressed as In addition, since the attacker's ability is limited, there must be an upper limit on the attack signal, which can be expressed as

[0091] S13, select the speed difference of the synchronous machine as the input signal of the controller. In the signal transmission, in order to affect the stable operation of the system, the attacker may invade the communication network to intercept the signal and inject false data such as parameter attack, random attack, data replay attack, data missing attack, etc., where the attack signal y a (k) can tamper with the actual measured value y(k), causing the damping controller to perceive incorrect state information, thereby generating incorrect control commands and affecting the stability of the system. The following is the established false data injection attack model:

[0092] (1) Parameter attack:

[0093]

[0094] Where A represents the amplitude of the attack signal, e ζ ·sin(ω s t) represents the damped sine wave associated with the system oscillation mode, ζ is the damping ratio of the dominant oscillation mode, ω s is the natural oscillation frequency of the system. a (t) represents the injected attack signal, t represents the current time, and the duration of the attack is represented by τ a express.

[0095] (2) Random attack:

[0096]

[0097] Among them, α and β are the minimum and maximum values ​​within the range of the controller input signal, respectively, and θ is a variable uniformly distributed on [0, 1], i.e., θ~Uniform(0,1). The upper and lower limits of the real system output are used as constraints to model the random attack signal. Since the attack signal is within the detection range of the bad data detection algorithm, it is not easy to be detected and has good concealment.

[0098] (3) Data Missing Attack:

[0099]

[0100] Unlike the above attacks, in a data missing attack, the attacker may prevent the data or control signals of some sensors from reaching the control center, resulting in the inability to obtain complete information. The data missing attack signal established in this paper is reflected as the measurement value transmitted by the wide area communication network is zero, which may be caused by a communication failure or loss.

[0101] (4) Data replay attack:

[0102]

[0103] Among them, tt 0Indicates the time when data needs to be recorded. In a data replay attack, the attacker records the actual output data of the controlled power system in normal operation or interference scenarios, and uses these data instead of the actual data during the attack, thereby misleading the controller's actions. For example, when a three-phase fault occurs in the power system, the attacker replays these operating data in the corresponding signal, resulting in the phenomenon of two consecutive faults. In addition, the attacker can replay the recorded data when the power system is in normal operation, resulting in a decrease in the damping performance of the controller. Since the attack signal injected by the replay attack is derived from the real recorded operating data, it has better confidentiality.

[0104] In step S2, a threshold is set to detect network attacks, and the stored controller history data is used to compensate the attacked controller input signal.

[0105] S21. The attacker can choose to inject false data into any measurement channel of the controller input signal to reduce the controller damping performance. In order to eliminate the impact of the attack signal on the controller input signal, it is first necessary to detect whether the attack exists. The following is the threshold detection design for FDIA:

[0106]

[0107] Among them, Indicates the transmission signal of the measurement channel attacked by FDI, y p (k) is the predicted control input calculated by the damping controller using the existing I / O data, and γ is the attack detection threshold. y When (k) = 0, it means that the control input has not been attacked or the impact of the attack can be ignored. Otherwise, it means that the control input has been attacked by the network, and the transmitted signal is unreliable. In threshold detection, the setting of the threshold γ is relatively important. If the γ value is too high, some attack behaviors will be missed. If the γ value is too low, some attacks may be misreported, thus affecting the performance of the controller. Therefore, the set threshold should be as close to the prediction error as possible, that is, γ min ≤|y(k)-y p (k)|.

[0108] S22. After identifying the attack signal, it is necessary to compensate the attacked signal to restore the authenticity of its data. The attack compensation module is designed as follows:

[0109]

[0110] in, represents the last unattacked transmission signal value before time k, or represents Υ y(kl) = 0. In order to prevent the controller output from not changing and affecting its damping performance when the system is attacked for too long, there is also an upper limit for the l value, that is, l∈{1,2,3…,l M After the input signal of the damping controller completes the attack detection and compensation in the "attack detection and attack compensation" link, the damping controller will use the attack-compensated data to update the input of the controlled system and predict the output of the controlled system at a later time.

[0111] In step S3, a data-driven model corresponding to the controlled system under the attack detection and compensation in step S2 is established, including a pseudo-partial derivative matrix estimator and a control law based on a model-free adaptive control algorithm, and a predicted output value of the controlled system at the next moment. Including:

[0112] The controlled power system described in S31 and S1 is transformed into the following forms under the network attack:

[0113]

[0114] u(k),u(k-1),...,u(kn u ),

[0115] d(k),d(k-1),...,d(kn d ))

[0116] When the controlled power system satisfies the assumption S1 and ||ΔU(k)||≠0, for any length L, there exists a bounded time-varying matrix Φ∈R L , called the pseudo partial derivative matrix, so that the system S3 can be transformed into the following data-driven model:

[0117]

[0118] in:

[0119] ΔU(k)=[Δu(k),Δu(k-1),...,Δu(k-L+1)] T

[0120] Δu(k)=u(k)-u(k-1)

[0121] Φ(k)=[φ 1 (k),φ 2 (k),...,φ L (k)] T

[0122] S32, based on the I / O data of the controlled power system, Φ(k) is estimated online, and the pseudo partial derivative estimator is:

[0123]

[0124] in is the pseudo partial derivative vector estimate, is the tracking error of the controlled system output, Γ(k)=η(μ+||ΔU(k-1)|| 2 ) -1 , weighting factor μ>0, step size factor η∈(0,2]. If the pseudo partial derivative at the current moment is less than or equal to the set value, or the difference between the control signal at the previous moment and the previous moment is less than or equal to the set value, the pseudo gradient at the current moment is reset to the set initial value of the pseudo gradient.

[0125] S33, set the expected value y of the output signal of the controlled power system * (k) = 0. Considering that the data-driven damping controller is an additional control, its control signal u(k) should be 0 after the system recovers stability. The output signal of the controller at the next moment can be obtained according to the data-driven model of the controlled power system:

[0126]

[0127] Among them, a linearization step vector [ρ 0 , ρ 1 ,…,ρ L ] T , ρ i >0, if i<1 or i>L,

[0128] S34, based on the pseudo partial derivative estimator described in S3 The current input signal of the controller And the ΔU(k) ​​of the controller sliding time window, the output of the controlled power system at the next moment can be obtained:

[0129]

[0130] Among them, y p (k+1) is the predicted value of the next moment k+1 calculated by the controller based on historical data and the data-driven model. One-step forward predictive control can only predict future values ​​in a short period of time, and there is a certain deviation. In addition, since the controlled power system has strong nonlinear and complex characteristics, it is very sensitive to the setting of initial conditions. Therefore, only the predicted value is used to determine whether the system is attacked, and the predicted value is not used to participate in the pseudo-partial derivative update.

[0131] In the step S4, the generated control signal is transmitted to the control circuit of the wind turbine, so that the corresponding wind turbine control circuit controls the wind turbine according to the control signal, so as to achieve effective damping of oscillation of the power system containing the wind turbine under the action of network attack.

[0132] To verify the effectiveness of the present invention, the example method of the present invention is now applied to a power system scenario containing a double-fed wind farm. The entire system block diagram after adding the Cyber ​​Attacks-Defensivemodel-free Wide-area Damping Controller (CAD-MFWADC) of the present invention is as follows: Figure 2 Specifically, the CAD-MFWADC is installed on the DFIG, and the signal containing the dominant oscillation modal component of the system is selected as the output signal y(k) of the controlled system, which is introduced into the control loop of the CAD-MFWADC to adjust the reactive power delivered by the DFIG to the grid. During the input signal transmission process of the controller, the y(k) after the false data is injected is converted into After passing through the attack detection and compensation module, the controller output signal u(k) is generated by the pseudo partial derivative estimator and the MFAC controller and transmitted to the control loop of the wind turbine RSC current inner loop control to realize the reactive power regulation of the wind turbine injected into the power system.

[0133] In order to verify the effectiveness and adaptability of the present invention, the capacity of the doubly fed wind farm is set to 600MW and it is operated at a random wind speed of 5-24m / s as the basic operating point. According to the results of eigenvalue analysis, the system has an interval weak damping oscillation mode with a frequency of 0.674Hz and a damping ratio of 1.928%. The relative speed difference ω between the synchronous machines G1 and G4 is selected by the main mode ratio method. 1-4 As feedback signals, the optimized and adjusted CAD-MFWADC parameters are shown in Table 1.

[0134] Table 1 CAD-MFWADC related parameters

[0135] Tab.1 Parameters of the CAD-MFWADC

[0136]

[0137] When wide-area damping control is not used, the system is in undamped oscillation, damped oscillation and unstable state respectively. For comparison, the MFAC-based wide-area damping controller (MFAC-WADC) and the conventional wide-area damping controller (CWADC) are introduced. The conventional wide-area damping controller is designed by the residue method, and its lead-lag module is as follows:

[0138]

[0139] Example 1: The system runs at the basic operating point. When t=2, a three-phase instantaneous short circuit fault occurs between bus 1 and bus 2. The fault clears itself after 100ms. When t=2-7, random uniform data is injected into the input signal ω of the controller. 1-4 middle. Figure 4a , Figure 4b , Figure 4c , Figure 5a , Figure 5b , Figure 5c , Figure 6a , Figure 6b , Figure 6c The figure shows the dynamic response of G2-G10 relative to the rotor angle of G1 under different types of false data injection attacks. It can be seen that under CWADC and MFAC-WADC, the controller is directly attacked without detection and protection. Under CAD-MFWADC, the attacked data will be identified by the attack detection and attack compensation modules, and then compensated. When the attacker stops at t=7s, CAD-MFWADC can immediately restore its ability to damp oscillation, while the system using CWADC and MFAC-WADC will fall into continuous divergent oscillation until the system is unstable. In addition, under three different types of parameter attacks, random attacks, and data loss attacks, CAD-MFWADC shows good adaptability and robustness.

[0140] Example 2: The system is running at a stable operation point, and the system response data after the three-phase instantaneous short circuit fault recorded in Example 1 is recorded and archived. At this time, the system has not been attacked by a network. The data replay attack is initiated at t = 2s and injected into the input signal of the damping controller for 5s. Figure 7a , Figure 7b and Figure 7c It shows the dynamic response of G2-G10 relative to the G1 rotor angle under data replay attack. It can be seen that the system with CWADC and CAD-MFWADC will oscillate during the attack phase, and CWADC cannot suppress this interval oscillation. CAD-MFWADC will damp the oscillation quickly for a period of time after the attack stops, while the system with MFAC will have unsuppressible interval oscillation.

[0141] The preferred embodiments of the present invention disclosed above are only used to help explain the present invention. The preferred embodiments do not describe all the details in detail, nor do they limit the invention to only specific implementation methods. Obviously, many modifications and changes can be made according to the content of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the present invention, so that those skilled in the art can understand and use the present invention well. The present invention is limited only by the claims and their full scope and equivalents.

Claims

1. A data-driven damping controller for a power system under a false data injection attack, comprising a damping controller installed on a wind turbine converter side, characterized in that: The control method of the damping controller comprises the following steps: S1: For the wide-area measurement signal of the damping controller, determine the conditions under which the attack signal can destroy the system stability and build various types of false data injection models; S2: Design a threshold-based attack detection mechanism and attack compensation module for the measured signal of the damping controller under attack; S3: establishing a data-driven model corresponding to the controlled system under attack compensation, wherein the data-driven model represents the relationship between the output signal of the power system and the control signal of the connected wind turbine through pseudo partial derivative matrix parameters and control laws; S4: The generated control signal is transmitted to the control circuit of the wind turbine, so that the corresponding wind turbine control circuit controls the wind turbine according to the control signal, so as to achieve low-frequency oscillation damping of the wind turbine power system including the network attack.

2. A data-driven damping controller for a power system under false data injection attack according to claim 1, characterized in that: The conditions under which the attack signal reaches the point of destabilization include: S11: The power system with wind turbine is represented as a discrete nonlinear model with single input and single output, including the control input u(k) of the damping controller, the wide-area measurement signal y(k) of the damping controller and the bounded disturbance d(k); where: y(k+1)=f(y(k),y(k-1),...,y(k-n y ), u(k),u(k-1),...,u(kn u ), d(k),d(k-1),…,d(k-n d )) S12: For the wide-area measurement signal y(k) of the damping controller, the attacker satisfies the following conditions: obtain the state information in the transmission network to construct the false data injection attack model y a (k), attack strategy Υ y (k)∈{0,1}, when the channel is attacked at time k, we get y (k) = 1, otherwise Υ y (k) = 0, the output of the attacked system is expressed as Since the attacker's ability is limited, there must be an upper limit for the attack signal, which can be expressed as S13: Select the speed difference of the synchronous machine as the input signal of the damping controller. During the transmission of the input signal, the attacker invades the communication network to intercept the signal and inject false data, including parameter attack, random attack, data replay attack and data missing attack types. The attack signal y a (k) Tampering with the actual measured value y(k) causes the damping controller to perceive incorrect state information, thereby generating incorrect control commands; the established false data injection attack model is: (1) Parameter attack: Where A represents the amplitude of the attack signal, e ζ ·sin(ω s t) represents the damped sine wave associated with the system oscillation mode, ζ is the damping ratio of the dominant oscillation mode, ω s is the natural oscillation frequency of the system, and the variable y a (t) represents the injected attack signal, t represents the current time, τ a Indicates the duration of the attack; (2) Random attack: Among them, α and β are the minimum and maximum values ​​in the input signal range of the damping controller, respectively, and θ is a variable uniformly distributed on [0, 1], that is, θ~Uniform(0,1). The upper and lower limits of the real system output are used as constraints to model the random attack signal. (3) Data Missing Attack: (4) Data replay attack: Among them, t-t0 represents the time when data needs to be recorded.

3. A data-driven damping controller for a power system under false data injection attack according to claim 1, characterized in that: The specific process of designing the threshold-based attack detection mechanism and attack compensation module is as follows: S21: The attacker chooses to inject false data into any measurement channel of the damping controller input signal; detect whether the attack exists, and the threshold detection for FDIA: Among them, Indicates the transmission signal of the measurement channel attacked by FDI, y p (k) is the predicted control input calculated by the damping controller using the existing I / O data, γ is the attack detection threshold; when Υ y When (k) = 0, it means that the control input has not been attacked or the attack impact is negligible. Otherwise, it means that the control input has been attacked by the network. At this time, the transmitted signal is unreliable. In the threshold detection, the threshold γ is set according to the prediction error, that is, γ min ≤|y(k)-y p (k)|; S22: After the attack signal is identified, the attacked signal needs to be compensated to restore the authenticity of its data. The attack compensation module is as follows: in, represents the last unattacked transmission signal value before time k, or represents Υ y (kl)=0;l∈{1,2,3…,l M }; After the input signal of the damping controller completes the attack detection and compensation in the "attack detection and attack compensation" link, the damping controller uses the attack-compensated data to update the input of the controlled system and predict the output of the controlled system at a certain time in the future.

4. A data-driven damping controller for a power system under false data injection attack according to claim 1, characterized in that: The data-driven model includes a pseudo partial derivative matrix estimator, a control law based on a model-free adaptive control algorithm, and the predicted output value of the controlled system at the next moment; the data-driven model corresponding to the controlled system under the attack compensation is established as follows: S31: The controlled power system is transformed into: u(k),u(k-1),...,u(kn u ), d(k),d(k-1),…,d(k-n d )) When the controlled power system satisfies the assumption and when ||ΔU(k)||≠0, for any length L, there exists a bounded time-varying matrix Φ∈R L , called the pseudo partial derivative matrix, makes the transformed data-driven model: in: ΔU(k)=[Δu(k),Δu(k-1),...,Δu(k-L+1)] T Δu(k)=u(k)-u(k-1) Φ(k)=[φ1(k),φ2(k),...,φ L (k)] T ; S32: Online estimation of Φ(k) is performed based on the I / O data of the controlled power system, and the pseudo partial derivative estimator is: in is the pseudo partial derivative vector estimate, is the tracking error of the controlled system output, Γ(k)=η(μ+||ΔU(k-1) 2 ) -1 , weighting factor μ>0, step size factor η∈(0,2], if the pseudo partial derivative at the current moment is less than or equal to the set value, or the difference between the control signal at the previous moment and the previous moment is less than or equal to the set value, then the pseudo gradient at the current moment is reset to the set initial value of the pseudo gradient; S33: Set the expected value y of the output signal of the controlled power system * (k) = 0. After the system recovers to stability, its control signal u(k) is 0. The output signal of the damping controller at the next moment is obtained according to the data-driven model of the controlled power system: Among them, a linearization step vector [ρ0, ρ1, ..., ρ L ] T , ρ i >0, if i<1 or i>L, S34: Pseudo-partial derivative based estimator The input signal of the damping controller at the current moment And the ΔU(k) ​​of the damping controller sliding time window, the output of the controlled power system at the next moment is obtained: Among them, y p (k+1) is the predicted value of the next moment k+1 calculated by the damping controller based on historical data and the data-driven model.

Images