Two-dimensional security risk assessment method for data-driven state estimation

Abstract

The invention discloses a two-dimensional security risk assessment method for data-driven state estimation, and relates to the technical field of security risk assessment, and the method comprises the steps: firstly, building a nonlinear relation among a data-driven state estimation model, analog quantity measurement and state quantity based on an artificial neural network architecture; then, based on an attacker view angle, utilizing the CGAN to initiate an adversarial attack to the data-driven state estimation model; then, the security risk faced by state estimation under the adversarial attack is assessed from two dimensions, the first dimension refers to attacker strategy analysis and assessment of attack cost and resources needed to be input by an attacker when the adversarial attack is implemented, and the second dimension refers to service vulnerability assessment and assessment of attack intensity and concealment of the adversarial attack; and finally, establishing a two-dimensional security risk assessment system according to a logic sequence of attack occurrence. According to the method, security threats existing in existing data-driven state estimation are revealed, and a basis is provided for formulating defense measures.

Description

Technical Field

[0001] The present invention relates to the technical field of security risk assessment, and in particular to a dual-dimensional security risk assessment method based on data-driven state estimation. Background Art

[0002] The diverse operating modes, randomized power flow distribution, and complex stability mechanisms of new power systems place higher demands on the real-time and accuracy of wide-area system state perception. Data-driven systems within new architectures offer new possibilities for expanding the observability of state estimation. However, the ubiquitous access, massive data transmission, and frequent real-time interactions of new power systems pose threats to the data security of state estimation. The sensitivity to small perturbations and limited generalization capabilities make data-driven state estimation a vulnerable target for adversarial attacks.

[0003] In this context, there is a need to continuously optimize data-driven state estimation methods while promptly detecting and assessing potential safety risks to ensure stable and reliable operation of power systems. It is particularly urgent to systematically evaluate and study the safety issues that may arise from data-driven algorithms in state estimation. However, research on these safety risks is currently relatively lacking. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to overcome the shortcomings of the existing technology and provide a two-dimensional security risk assessment method for data-driven state estimation. The present invention reveals the security risks existing in the existing data-driven state estimation and provides a scientific basis for formulating corresponding defense measures.

[0005] The present invention adopts the following technical solutions to solve the above technical problems:

[0006] A dual-dimensional safety risk assessment method for data-driven state estimation proposed in the present invention includes:

[0007] Establish a data-driven state estimation model based on artificial neural network architecture;

[0008] From the attacker's perspective, we use the conditional generative network (CGAN) to launch an adversarial attack on the data-driven state estimation model.

[0009] The security risks faced by state estimation under adversarial attacks are evaluated from the first and second dimensions, where:

[0010] The first dimension refers to the attacker's strategy analysis. The first dimension evaluates the security risks faced by state estimation under adversarial attacks, which means: evaluating the attack cost and resources required by the attacker to implement the adversarial attack;

[0011] The second dimension refers to business vulnerability assessment. The second dimension assesses the security risks faced by state estimation under adversarial attacks, which means: assessing the attack intensity and concealment of adversarial attacks;

[0012] By weighting and summing the evaluation indicators of attack cost and resources in the first dimension and attack intensity and concealment in the second dimension, a two-dimensional security risk assessment system is established to implement security risk assessment based on data-driven state estimation.

[0013] As a further optimization scheme of the dual-dimensional security risk assessment method of data-driven state estimation described in the present invention, a data-driven state estimation model is established; the details are as follows:

[0014] First, the attacker captures real-time and historical data from multiple data sources through the power system monitoring mechanism. The data includes measurement data and system operation status, and integrates the historical data into the original data set I = {X 1:S ,Z 1:S}, where X 1:S is the original state quantity of group S, Z 1:S For the S group of original quantity measurements; then, based on the artificial neural network architecture, according to the obtained original data set I = {X 1:S ,Z 1:S}Build a data-driven state estimation model f θ (*), the objective function is defined as minimizing the prediction error; then, the optimization algorithm of gradient descent is used to refine f θ (*) predictive ability, nonlinear relationship between analog measurement and state quantity.

[0015] As a further optimization scheme for the dual-dimensional security risk assessment method for data-driven state estimation described in the present invention, using CGAN to launch an adversarial attack on the data-driven state estimation model includes:

[0016] The process of launching an adversarial attack is as follows: first, the attacker obtains the original data set of the attack node under different operating states of the power system; then, the two deep network architectures of the generator and discriminator in the CGAN are constructed, and the CGAN is updated through alternating adversarial training; then, the label information that meets the attacker's intention is determined, and the original measurement of the power system is collected. The attacker uses the generator to generate a disturbance u, and superimposes the disturbance with the original measurement z to form an analog measurement z′ that the discriminator cannot distinguish. This analog measurement is consistent with the label information; finally, the predicted state x′ output by the data-driven state estimation model is tested, completing the adversarial attack on the data-driven state estimation model under the given label information.

[0017] As a further optimization scheme for the dual-dimensional security risk assessment method of data-driven state estimation described in the present invention, the generator and discriminator in CGAN include:

[0018] The generator G in CGAN takes as input a set of raw measurements z and label information y. During the iteration process, the generator generates perturbations u with the same distribution as the raw measurements, and superimposes the perturbations u with the raw measurements z to form simulated measurements z′.

[0019] The discriminator D in CGAN combines a set of raw measurements z and simulated measurements z′ into a data batch, associates this data batch with label information y, and then inputs them into the discriminator. The discriminator distinguishes between raw and simulated measurements and classifies a given data batch as real or fake. When the simulated measurement approaches the raw measurement and meets the pre-set label information, it is judged as real.

[0020] As a further optimization scheme of the dual-dimensional security risk assessment method of data-driven state estimation described in the present invention, a dual-dimensional security risk assessment system is established, specifically as follows:

[0021] In a power system with a total of K nodes, assume that an attacker launches A counterattacks against N attack nodes, 1≤N≤K. Determine the evaluation indicators corresponding to the two dimensions of attacker strategy analysis and business vulnerability assessment.

[0022] The security risk assessment score is obtained by weighting and summing the four evaluation indicators of attack cost and resources in the first dimension and attack intensity and concealment in the second dimension.

[0023] As a further optimization scheme of the dual-dimensional security risk assessment method of data-driven state estimation described in the present invention, the evaluation indicators include:

[0024] First dimension:

[0025] Define the average attack cost c and the resources r required for the attack a ,Analyze the attack entry point and sustainability based on the obtained attack cost and resource assessment results;

[0026] The ratio of the perturbation generated by the generator in the adversarial attack to the original amount, i.e., the relative perturbation error, is used as the metric of the attack cost; for the i-th attack node, 1≤i≤N, c i is the attack cost of the i-th attack node. Calculate the attack cost of the j-th adversarial attack for all attack nodes, 1≤j≤A. Assuming that each attack is independent, sum the attack costs of all attacks to obtain the average attack cost c of the total A adversarial attacks.

[0027] The resources that the attack needs to obtaina It is represented by the ratio of the number of attack nodes selected by the attacker to the total number of nodes in the power system;

[0028]

[0029] Among them, a i It is used to characterize whether the attack cost of the i-th attack node meets the pre-set threshold requirement λ, u i is the perturbation amount generated by the generator on the i-th attack node, z i The raw quantity measurement of the i-th attack node;

[0030] The average attack cost represents the cost of an attacker launching an adversarial attack against a data-driven state estimation model. The sustainability of an adversarial attack depends on the cost of the attack and the resources required for the attack. The selected attack nodes represent the resources required for the attack and indicate the entry point of the attack.

[0031] Second Dimension:

[0032] Define the average attack result deviation Δx and the probability B of the attack bypassing bad data detection. Evaluate the attack strength and stealth of the adversarial attack based on the obtained attack result deviation and the probability of the attack bypassing the bad data detection mechanism.

[0033] The distance between the predicted state quantity output by the data-driven state estimation model under adversarial attack and the original state quantity is used as the metric for the attack result deviation. For the i-th attack node, 1≤i≤N, the attack result deviation of the j-th adversarial attack is calculated for all attack nodes, 1≤j≤A. Assuming that each attack is independent, the attack result deviations of all attacks are summed to obtain the average attack result deviation Δx of the total A adversarial attacks.

[0034]

[0035]

[0036] Among them, B j It is used to characterize whether the jth attack in the total A adversarial attacks can bypass the bad data detection, x i ′ is the predicted state quantity output after the analog measurement is input into the data-driven state estimation model, x i is the original state of the i-th attack node;

[0037] The average attack result deviation represents the attack strength of the adversarial attack, and the stealthiness of the adversarial attack is measured by the probability that the attack can bypass the detection of bad data.

[0038] Compared with the prior art, the present invention adopts the above technical solution and has the following technical effects:

[0039] Taking into account the sensitivity of current data-driven algorithms to small perturbations, this paper provides a two-dimensional security risk assessment method for data-driven state estimation under adversarial attacks based on CGAN. It explores existing security vulnerabilities from the two dimensions of attacker strategy analysis and business vulnerability assessment, and evaluates the risk impact of attack cost and attack intensity to reveal the security threats existing in existing data-driven state estimation and provide a scientific basis for formulating corresponding defense measures. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 is a flow chart of the method of the present invention;

[0041] Figure 2 This is a schematic diagram of a data-driven state estimation model based on an artificial neural architecture in the present invention;

[0042] Figure 3 Schematic diagram of the CGAN structure of the present invention;

[0043] Figure 4 Schematic diagram of the dual-dimensional security risk assessment method of the present invention;

[0044] Figure 5 The IEEE-39 node power grid model of the present invention;

[0045] Figure 6 : This is the test curve of the data-driven state estimation model of the present invention; wherein (a) is the voltage amplitude V, and (b) is the phase angle θ;

[0046] Figure 7 The results of the anti-attack diagram of the present invention are shown in FIG. 1 , where (a) is normal to upper boundary, (b) is upper boundary to lower boundary, (c) is upper boundary to normal, and (d) is lower boundary to upper boundary.

[0047] Figure 8 The figure compares the attack costs of the present invention and the FGSM-based counterattack method; wherein, (a) is the FGSM-based counterattack, and (b) is the attack method proposed by the present invention. DETAILED DESCRIPTION

[0048] In order to make the objectives, technical solutions and advantages of the present invention more clear, the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0049] like Figure 1 As shown, a two-dimensional security risk assessment method for data-driven state estimation (SE) under adversarial attack based on conditional generative adversarial network (CGAN) is proposed. The method includes the following steps:

[0050] Establish a data-driven state estimation model based on artificial neural network architecture;

[0051] From the attacker's perspective, we use the conditional generative network (CGAN) to launch an adversarial attack on the data-driven state estimation model.

[0052] The security risks faced by state estimation under adversarial attacks are evaluated from the first and second dimensions, where:

[0053] The first dimension refers to the attacker's strategy analysis. The first dimension evaluates the security risks faced by state estimation under adversarial attacks, which means: evaluating the attack cost and resources required by the attacker to implement the adversarial attack;

[0054] The second dimension refers to business vulnerability assessment. The second dimension assesses the security risks faced by state estimation under adversarial attacks, which means: assessing the attack intensity and concealment of adversarial attacks;

[0055] According to the logical order of attack occurrence, the attack cost and resources in the first dimension and the attack intensity and concealment in the second dimension are weighted and summed to establish a two-dimensional security risk assessment system to implement data-driven state estimation security risk assessment.

[0056] The data-driven state estimation model is established as follows:

[0057] like Figure 2 As shown, the present invention selects the deep neural network (DNN) in the artificial neural network architecture to establish a data-driven state estimation model. The specific process is as follows:

[0058] First, the attacker captures real-time and historical data from multiple data sources through the power system monitoring mechanism. The data includes measurement data and system operation status, and integrates the historical data into the original data set I = {X 1:S ,Z 1:S}, where X 1:S is the original state quantity of group S, Z 1:S The original quantity measurement of group S is obtained; then, based on DNN, according to the original data set I={X 1:S ,Z 1:S}Build a data-driven state estimation model f θ (*), the objective function is defined as minimizing the prediction error; then, the optimization algorithm of gradient descent is used to refine f θ (*) predictive ability, nonlinear relationship between analog measurement and state quantity;

[0059]

[0060] in, The S group of predicted state quantities estimated by the data-driven state estimation model. The loss function L(θ) is used to measure the predicted state quantity of the data-driven state estimation model. With the original state quantity X 1:S The difference between. is a pre-set loss function, θ t+1 is the updated model parameter after the t+1th round of training, θ t is the model parameter obtained from the current t-th round of training. η is the learning rate, is the loss function L(θ after the current t-th round of training (t) ) with respect to the gradient of the model parameters θ;

[0061] The model parameters θ are updated according to the gradient descent algorithm to minimize the loss function.

[0062] Using CGAN to launch adversarial attacks on data-driven state estimation models includes:

[0063] The adversarial attack process is as follows: First, the attacker obtains the original data set of the attack node under different operating states of the power system. Then, the two deep network architectures of the generator and discriminator in the CGAN are constructed and updated through alternating adversarial training. Then, the attacker determines the label information that meets the attacker's intention and collects power system measurements. The attacker uses the generator to generate a disturbance u, which is superimposed on the original measurement z to form an analog measurement z′ that the discriminator cannot accurately distinguish. This analog measurement is consistent with the label information. Finally, the predicted state x′ output by the data-driven state estimation model is tested, completing the adversarial attack on the data-driven state estimation model under the given label information.

[0064] In a power system, assume that there are N attack nodes, and the attack node is an M-dimensional vector. The original quantity measurement of each attack node is recorded as z = [z1, z2, ... z N ] T , z n’ The n'th original quantity measurement, 1≤n'≤N, the superscript T represents the transposition, the original state quantity is recorded as x=[x1,x2,…x N ] T , x n’ is the n'th original state quantity, and the perturbation quantity generated by the generator is u=[u1,u2,…u N ] T ,u n’ is the n'th disturbance, u is superimposed on the original measurement to form the simulated measurement z'. The predicted state output by z' after passing through the data-driven state estimation model is recorded as x'. G(*) represents the mapping relationship of the generator, and D(*) represents the mapping relationship of the discriminator;

[0065] Combined with the formula, the process of launching an adversarial attack against the i-th attack node is expressed as follows, 1≤i≤N:

[0066] u i =G(z i ,y)

[0067]

[0068] {d i ,d i ′}=D(z i ,z i ′,y)

[0069] x i ′=f θ (z i ′)

[0070] Among them, u i is the perturbation amount generated by the generator on the i-th attack node, z i is the raw quantity measurement of the i-th attack node, z i (M,:) means taking z i All M-dimensional attack vectors, z i ′ is the analog measurement formed by the attacker at the i-th attack node, and F(*) is z i 、u i and z i ′, F(*) represents the superposition relationship, d i The original quantity measurement and label information y are input into the discriminator’s discrimination result, d i ′ is the discrimination result of the discriminator inputted with the analog measurement and the label information y, d i and d i ′ are all probability vectors output by the sigmoid function of the last layer of the discriminator, x i ′ is the analog measurement z i ′ is input into the data-driven state estimation model and outputs the predicted state quantity.

[0071] The generator and discriminator in the CGAN include:

[0072] like Figure 3 As shown in Figure 2, the generator G in CGAN takes as input a set of original measurements z and label information y. During the iteration process, the generator generates perturbations u with the same distribution as the original measurements, and superimposes the perturbations u with the original measurements z to form simulated measurements z′.

[0073] The discriminator D in the CGAN takes a set of raw measurements z and simulated measurements z′, groups them into a data batch, associates this data batch with label information y, and then feeds both into the discriminator. The discriminator distinguishes between raw and simulated measurements and classifies a given data batch as real or fake. When the simulated measurement approaches the raw measurement and meets the pre-set label information, it is judged as real.

[0074] Train CGAN according to the loss function J(*);

[0075]

[0076] in, Indicates that the original quantity measurement z obeys P z The distribution of (z).

[0077] The establishment of a two-dimensional security risk assessment system includes:

[0078] In a power system with a total of K nodes, assume that an attacker launches A counterattacks against N attack nodes, 1≤N≤K. Determine the evaluation indicators corresponding to the two dimensions of attacker strategy analysis and business vulnerability assessment.

[0079] The security risk assessment score R is obtained by weighting and summing the four evaluation indicators of attack cost and resources in the first dimension and attack intensity and concealment in the second dimension:

[0080] R=(R1c+R2r a )+(R3Δx+R4B)

[0081] Among them, R1~R4 are the weighted coefficients of the four evaluation indicators of attack cost and resources in the first dimension, and attack intensity and concealment in the second dimension. The weighted coefficients are pre-set according to the system scale, attack target and evaluation criteria.

[0082] The evaluation indicators include:

[0083] First dimension:

[0084] Define the average attack cost c and the resources r required for the attack a ,Analyze the attack entry point and sustainability based on the obtained attack cost and resource assessment results;

[0085] The ratio of the perturbation generated by the generator in the adversarial attack to the original amount, i.e., the relative perturbation error, is used as the metric of the attack cost; for the i-th attack node, 1≤i≤N, c iis the attack cost of the i-th attack node. Calculate the attack cost of the j-th adversarial attack for all attack nodes, 1≤j≤A. Assuming that each attack is independent, sum the attack costs of all attacks to obtain the average attack cost c of the total A adversarial attacks.

[0086] The resources that the attack needs to obtain a It is represented by the ratio of the number of attack nodes selected by the attacker to the total number of nodes in the power system.

[0087]

[0088] Among them, a i It is used to characterize whether the attack cost of the i-th attack node meets the pre-set threshold requirement λ, u i is the perturbation amount generated by the generator on the i-th attack node, z i The raw quantity measurement of the i-th attack node;

[0089] The average attack cost represents the cost of an attacker launching an adversarial attack against a data-driven state estimation model. The sustainability of an adversarial attack depends on the cost of the attack and the resources required for the attack. The selected attack nodes represent the resources required for the attack and indicate the entry point of the attack.

[0090] Second Dimension:

[0091] Define the average attack result deviation Δx and the probability B of the attack bypassing bad data detection. Evaluate the attack strength and stealth of the adversarial attack based on the obtained attack result deviation and the probability of the attack bypassing the bad data detection mechanism.

[0092] The distance between the predicted state quantity output by the data-driven state estimation model under adversarial attack and the original state quantity is used as a measure of the attack result deviation. For the i-th attack node, 1≤i≤N, the attack result deviation of the j-th adversarial attack is calculated for all attack nodes, 1≤j≤A. Assuming that each attack is independent, the attack result deviations of all attacks are summed to obtain the average attack result deviation Δx for the total A adversarial attacks.

[0093]

[0094] Among them, B j It is used to characterize whether the jth attack in the total A adversarial attacks can bypass the bad data detection, x i ′ is the predicted state quantity output after the analog measurement is input into the data-driven state estimation model, x i is the original state of the i-th attack node;

[0095] The average attack result deviation represents the attack strength of the adversarial attack, and the stealthiness of the adversarial attack is measured by the probability of the attack bypassing the detection of bad data.

[0096] Figure 4 Schematic diagram of the two-dimensional security risk assessment method.

[0097] An optional implementation manner of the present invention is described in detail below.

[0098] In one embodiment of the present invention, the above-mentioned security risk assessment method is applied to Figure 5 The IEEE-39-node grid model shown in the figure consists of 39 buses, 19 loads, and 10 generators. In this scenario, uniform sampling is used to obtain 10,000 steady-state samples, or raw quantity measurements, including node active power P, reactive power Q, and voltage amplitude V. The dimensions of P, Q, and V are all 39 x 10,000. The validation scenario database is generated using the Monte Carlo method, with the following random parameters: load levels follow a uniform distribution in the range [0.8, 1], and load variations follow a normal distribution in the range N(0, 0.03).

[0099] For the state estimation problem in the power system, the main focus is on the stable state of the power grid at a specific time point, including the voltage amplitude and angle, the injected power of the bus, etc., which plays a key role in preventing faults and optimizing power grid operation. The present invention uses the voltage amplitude state as an attack guide to verify the effectiveness of the method in assessing voltage safety risks.

[0100] We further supplemented the data with 2000 sets of abnormal voltage amplitude status samples, including 1000 sets of over-bounded and under-bounded samples. In this example, the normal voltage per unit value range is set to 0.95 to 1.05. Over-bounded refers to samples where the voltage per unit value of some nodes exceeds 1.05, and under-bounded refers to samples where the voltage per unit value is below 0.95.

[0101] After learning and optimizing the training data set, the DNN-based data-driven state estimation model can effectively identify and measure the complex relationship between data and system state, such as Figure 6 The test results shown, Figure 6 (a) is the voltage amplitude V, Figure 6 (b) in the figure is the phase angle θ. When the model predicts the voltage amplitude and phase angle, the mean absolute error (MAE) is 0.0013 and 0.0019 respectively, indicating that the model has high prediction accuracy, thus providing a business-side state estimation model foundation for the CGAN-based adversarial attack.

[0102] The method performs a preliminary attack cost and resource assessment on all nodes, determining which nodes, from the attacker's perspective, have smaller true values ​​and are more susceptible to attack. Based on the assessment results, the attacker selects nodes with the best attack effectiveness and lower resource requirements for attack. In the example described in the present invention, the attacker only conducts adversarial attacks on 28 nodes of the IEEE-39 node power grid model after screening, using these nodes as the entry point for the attack, thereby more efficiently utilizing limited resources.

[0103] Figure 7 This is the result diagram of the anti-attack of the present invention. Figure 7 (a) is normal to over-bounds, Figure 7 (b) in the example is from over-bounds to under-bounds. Figure 7 (c) in the figure is over the limit to normal. Figure 7 (d) in the figure represents the transition from lower to upper bound; the dashed line represents the upper bound of 1.05, and the dotted line represents the lower bound of 0.95. In this security risk assessment method, from the perspective of state estimation, the CGAN-based adversarial attack can achieve targeted modification of the output of the data-driven SE model. Samples subjected to the adversarial attack can cause all state quantities V to be incorrectly estimated as having a preset label category. For example, attacking an originally normal sample can cause it to be misclassified as an out-of-bounds sample, thereby disrupting the reliable operation of subsequent services.

[0104] In terms of attacker strategy analysis, the CGAN uses node information as a starting point. It replaces direct modeling of the physical mechanisms of the power system with a learning process based on the distribution and characteristics of data acquired by the attacker. Leveraging the resulting simulated measurement impact model, as long as attackers have access to system measurements for a period of time, they can launch a sustained adversarial attack, resulting in severe consequences for system decision-making. Furthermore, the reduction and efficient utilization of attack nodes also reduces attack costs and the difficulty of acquiring attack resources.

[0105] The possibility of attacks bypassing the bad data detection mechanism in the security risk assessment method was examined. Table 1 shows the average bypass rates of the simulated and original measurements at different detection thresholds. It can be observed that for thresholds greater than 1.2, the bad data detection bypass rates of the two are very close, indicating that the simulated and original measurements generated by the proposed method have relatively similar distributions. Furthermore, after analysis, the network has 39 degrees of freedom. Assuming that the weighted residual sum of squares follows a chi-square distribution and taking the significance level α as 0.01, the test value found in the distribution table is 62.42. The corresponding weighted sum of squares estimate for the generated simulated measurement is 19.95, thus meeting the estimation requirements and passing the residual test.

[0106] Table 1. Bypass rate of bad data detection in analog measurement under different thresholds

[0107] Threshold Raw quantity measurement Analog measurement 1.02 0.98 0.78 1.05 0.98 0.84 1.08 0.99 0.92 1.2 0.99 0.98 2 0.99 0.99

[0108] In terms of service vulnerability assessment, the proposed attack method's strength was quantified. Given label information that matches the attack intent, the proposed method can redirect the system's original state quantity to the preset state quantity corresponding to the label information, causing a significant deviation in the voltage amplitude, thereby achieving the goal of manipulating the data-driven SE model. Furthermore, by analyzing the bypass rate of bad data detection, the proposed attack also demonstrates a certain degree of concealment, injecting potential security risks into the state estimation service and confirming its vulnerability to countermeasures.

[0109] In order to further verify the effectiveness of the proposed security risk assessment method for data-driven state estimation under adversarial attacks, the following two attack scenarios are used for comparison:

[0110] Solution 1: FGSM-based counter-attack method;

[0111] Solution 2: The CGAN-based adversarial attack method proposed in this invention.

[0112] The perturbation amount ε of the FGSM algorithm is set to 0.1, and the relative error of superimposing perturbations on the measured data in scheme 1 is tested compared with the proposed method, i.e., the attack cost. The comparison results are as follows: Figure 8 As shown, Figure 8 (a) is an adversarial attack based on FGSM. Figure 8 (b) in the figure represents the attack method proposed in this paper. As can be seen from the figure, in Scheme 1, the relative error of the superimposed perturbation on each node is up to 30 times, and the average perturbation relative error of over 70% of the measurements is higher than 10%. In Scheme 2, the errors of the vast majority of attack nodes are within 10%, and only less than 3% of the measurements have errors exceeding 20%. The relative errors of the perturbations superimposed on reactive power and voltage amplitude are all less than 20%. The attack cost is generally much lower than the counterattack results of Scheme 1.

[0113] Table 2 compares the two methods across five buses. Compared to Scheme 1, Scheme 2 produces less cumulative perturbations and lowers the attack cost, given a fixed number of attack buses. For node 38, Scheme 1 achieves relative errors of 4.3% for active power and 0.48% for reactive power, respectively. Scheme 2, on the other hand, requires only cumulative perturbations of 0.01% and 0.183%, significantly less than the former.

[0114] Table 2 Comparison of attack costs of two methods

[0115]

[0116] Also in the dimension of attacker strategy analysis, the method can achieve the attack effect that meets the attack intention with less attack cost than Solution 1, and has more significant attack sustainability.

[0117] In summary, this paper proposes a two-dimensional security risk assessment method for data-driven state estimation under adversarial attacks. First, based on the artificial neural network architecture, a data-driven state estimation model is established to simulate the nonlinear relationship between measured quantities and state quantities. Then, from the attacker's perspective, CGAN is used to launch an adversarial attack on the data-driven state estimation model. Then, the security risks faced by state estimation under adversarial attacks are evaluated from two dimensions. Dimension one refers to attacker strategy analysis, which evaluates the attack cost and resources required to implement the adversarial attack. Dimension two refers to business vulnerability assessment, which evaluates the attack intensity and concealment of the adversarial attack. Finally, a two-dimensional security risk assessment system is established according to the logical sequence of attack occurrence. Examples verify the actual operation effect of the proposed scheme. From the perspective of security risk assessment, the security risks of data-driven state estimation are revealed, providing a theoretical basis and practical guidance for the design and implementation of effective defense measures.

[0118] In today's new power system landscape, with its diverse operating modes and complex stability mechanisms, ubiquitous access and massive data transmission pose threats to the data security of state estimation. Data-driven state estimation's sensitivity to small perturbations and limited generalization capabilities make it a vulnerable target, allowing attackers to launch covert adversarial attacks that threaten the stable operation of state estimation services. However, research on these security risks and attack threats is currently lacking.

[0119] The purpose of the present invention is to provide a two-dimensional security risk assessment method for data-driven state estimation, so as to comprehensively reveal the security risks existing in existing data-driven state estimation, thereby providing a scientific basis for formulating corresponding defense measures.

[0120] The innovation of the present invention is:

[0121] 1. An innovative CGAN-based adversarial attack on the data-driven state estimation model is proposed. The basic model is improved and optimized to reduce the attack cost while making the attack results produce preset deviations according to the attacker's intentions, thereby maximizing the security risk of the state estimation business under adversarial attacks and providing an attack basis for the evaluation system.

[0122] 2. An innovative two-dimensional security risk assessment method is proposed from the perspectives of both attackers and system services. Following the logical sequence of attacks, from the acquisition of attack sources to the implementation of attack behaviors, and then to the malicious consequences caused by the attacks and the corresponding attack detection methods, the impact of adversarial attacks on state estimation services is considered in a relatively comprehensive manner.

[0123] The application points of the present invention are:

[0124] 1. The proposed attack method is not limited to the artificial neural network architecture DNN used in the data-driven state estimation model example of the present invention, but is also applicable to many other data-driven state estimation models, and has strong adaptability.

[0125] 2. The method described in this invention is primarily applicable to the field of power system security risk assessment technology. By using the attack method described in this invention to uncover potential vulnerabilities in state estimation services, the security risk assessment system analyzes attacker strategies and service vulnerabilities, providing a scientific basis for developing appropriate defense measures and offering strong support for future power system security.

[0126] 3. Actual applications can enhance system security by building an active defense system, conducting in-depth research on network intrusion and attack models, and combining artificial intelligence technology to quickly assess and warn of security risks.

[0127] Throughout this specification, references to terms such as "one embodiment," "example," or "specific example" indicate that a specific feature, structure, material, or characteristic described in conjunction with that embodiment or example is included in at least one embodiment or example of the present disclosure. In this specification, schematic representations of these terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.

[0128] The above shows and describes the basic principles, main features and advantages of the present disclosure. Those skilled in the art should understand that the present disclosure is not limited to the above embodiments. The above embodiments and descriptions are merely illustrative of the principles of the present disclosure. Various changes and improvements may be made to the present disclosure without departing from the spirit and scope of the present disclosure, and such changes and improvements shall fall within the scope of the present disclosure.

Claims

1. A dual-dimensional safety risk assessment method for data-driven state estimation, characterized in that: include: Based on the artificial neural network architecture, a data-driven state estimation model is established; From the attacker's perspective, we use the conditional generative network (CGAN) to launch adversarial attacks on the data-driven state estimation model. The security risks faced by state estimation under adversarial attacks are evaluated from the first and second dimensions, where: The first dimension refers to the attacker's strategy analysis. The first dimension evaluates the security risks faced by state estimation under adversarial attacks, which means: evaluating the attack cost and resources required by the attacker to implement adversarial attacks; The second dimension refers to the business vulnerability assessment. The second dimension assesses the security risks faced by state estimation under adversarial attacks, which means: assessing the attack intensity and concealment of adversarial attacks; By weighting and summing the attack cost and resources in the first dimension and the attack intensity and concealment in the second dimension, a two-dimensional security risk assessment system is established to implement security risk assessment based on data-driven state estimation.

2. According to claim 1, a dual-dimensional safety risk assessment method for data-driven state estimation is characterized in that: Establish a data-driven state estimation model; the details are as follows: First, the attacker captures real-time and historical data from multiple data sources through the power system monitoring mechanism. The data includes measurement data and system operation status, and integrates the historical data into the original data set I = {X 1:S ,Z 1:S }, where X 1:S is the original state quantity of group S, Z 1:S is the original quantity measurement of group S; then, based on the artificial neural network architecture, according to the original data set I = {X 1:S ,Z 1:S }Build a data-driven state estimation model f θ (*), the objective function is defined as minimizing the prediction error; then, the gradient descent optimization algorithm is used to refine f θ (*) predictive ability, nonlinear relationship between analog measurement and state quantity.

3. According to claim 1, a data-driven state estimation dual-dimensional safety risk assessment method is characterized in that: Using CGAN to launch adversarial attacks on data-driven state estimation models includes: The process of launching an adversarial attack is as follows: first, the attacker obtains the original data set of the attack node under different operating states of the power system; then, two deep network architectures of the generator and discriminator in CGAN are constructed, and CGAN is updated through alternating adversarial training; then, the label information that meets the attacker's intention is determined, and the original quantity measurement of the power system is collected. The attacker uses the generator to generate a disturbance u, and superimposes the disturbance with the original quantity measurement z to form an analog quantity measurement z′ that the discriminator cannot distinguish. This analog quantity measurement is consistent with the label information; finally, the predicted state quantity x′ output by the data-driven state estimation model is verified to complete the adversarial attack on the data-driven state estimation model under the given label information.

4. According to claim 3, a data-driven state estimation dual-dimensional safety risk assessment method is characterized in that: The generator and discriminator in CGAN include: The generator G in CGAN takes as input a set of original quantity measurements z and label information y. During the iteration process, the generator generates a perturbation u with the same distribution as the original quantity measurements, and superimposes the perturbation u with the original quantity measurements z to form a simulated quantity measurement z′. The discriminator D in CGAN combines a set of original quantity measurements z and simulated quantity measurements z′ into a data batch, corresponds this data batch to the label information y respectively, and then inputs them into the discriminator together; the discriminator distinguishes between original quantity measurements and simulated quantity measurements, and the discriminator classifies a given data batch as true or false; when the simulated quantity measurement is close to the original quantity measurement and meets the pre-set label information, it is judged as true.

5. The dual-dimensional safety risk assessment method for data-driven state estimation according to claim 1 is characterized in that: Establish a two-dimensional security risk assessment system, as follows: In a power system with a total of K nodes, assume that the attacker launches A counterattacks on N attack nodes, 1≤N≤K, and determine the evaluation indicators corresponding to the two dimensions of attacker strategy analysis and business vulnerability assessment; The four evaluation indicators of attack cost and resources in the first dimension and attack intensity and concealment in the second dimension are weighted and summed to obtain the security risk assessment score.

6. A dual-dimensional safety risk assessment method for data-driven state estimation according to claim 5, characterized in that the assessment index include: First Dimension: Define the average attack cost c and the resources r required for the attack a ,Analyze the attack entry point and sustainability based on the attack cost and resource evaluation results; The ratio of the perturbation generated by the generator in the adversarial attack to the original measurement, i.e., the relative perturbation error, is used as the metric of the attack cost; for the i-th attack node, 1≤i≤N, c i is the attack cost of the i-th attack node. Calculate the attack cost of the j-th adversarial attack for all attack nodes, 1≤j≤A. Assuming that each attack is independent, sum the attack costs of all attacks to get the average attack cost c of the total A adversarial attacks. The resources that the attack needs to obtain a It is represented by the ratio of the number of attack nodes selected by the attacker to the total number of nodes in the power system; Among them, a i It is used to characterize whether the attack cost of the i-th attack node meets the preset threshold requirement λ, u i is the perturbation amount generated by the generator on the i-th attack node, z i The raw quantity measurement of the i-th attack node; The average attack cost represents the attack cost that an attacker needs to invest when launching an adversarial attack on the data-driven state estimation model. Whether the adversarial attack is sustainable depends on the cost of the attack and the resources required for the attack. The selected attack nodes represent the resources required for the attack and are used to indicate the entry point of the attack. Second Dimension: Define the average attack result deviation Δx and the probability B of the attack bypassing the bad data detection mechanism; evaluate the attack strength and concealment of the adversarial attack based on the obtained attack result deviation and the probability of the attack bypassing the bad data detection mechanism; The distance between the predicted state quantity output by the data-driven state estimation model under adversarial attack and the original state quantity is used as the metric of attack result deviation; for the i-th attack node, 1≤i≤N, the attack result deviation of the j-th adversarial attack is calculated for all attack nodes, 1≤j≤A, assuming that each attack is independent, the attack result deviations of all attacks are summed to obtain the average attack result deviation Δx of the total A adversarial attacks; Among them, B j It is used to characterize whether the jth attack in the total a adversarial attacks can bypass the bad data detection, x i ′ is the predicted state quantity output after the analog measurement is input into the data-driven state estimation model, x i is the original state of the i-th attack node; The average attack result deviation represents the attack strength of the adversarial attack, and the stealth of the adversarial attack is measured by the probability of the attack bypassing the detection of bad data.

Images