Power system attack target identification method based on customized generative model

Abstract

The invention discloses a power system attack target identification method based on a customized generative model, and belongs to the technical field of power information physical system security, and the method comprises the following steps: S1, collecting power system operation data; identifying a to-be-attacked eye; s2, establishing a physical constraint model, and solving the physical constraint model to obtain an attack vector; and S3, superposing the attack vector in the S2 to real measurement data by using a power system communication network through a software defined radio or protocol stack attack means, performing injection attack on the to-be-attacked target in the S1, and verifying the attack effect. According to the power system attack target recognition method based on the customized generative model, attack vector design is converted into an optimization problem containing physical constraints, the physical feasibility and concealment of attack vectors are enhanced, and automatic integration of attack processes is achieved.

Description

Technical Field

[0001] The present invention belongs to the field of power cyber-physical system security technology, and specifically relates to a power system attack target identification method based on a customized generative model. Background Art

[0002] As power systems become increasingly reliant on SCADA systems and advanced metering infrastructure, the need for cybersecurity protection has significantly increased. False Data Injection Attacks (FDIA) have become a key threat due to their high stealth and destructive nature. FDIAs inject specially crafted interference vectors into power system measurement data, disrupting the state estimation process and potentially causing generation plan deviations, branch circuit overloads, and even systemic blackouts.

[0003] Existing FDIA research primarily focuses on attack paths and defense chains. Attack path-based research often relies on graph analysis or static load metrics to identify vulnerable nodes / branches before an attack, but relies on fixed topological parameters and lacks dynamic integration of real-time operational data. During an attack, generative adversarial networks or heuristic algorithms are used to construct attack vectors, but physical constraints are generally ignored. Post-attack evaluation and feedback are attempted, but standardized systems and strategies have yet to be established. Defense chain-based research uses topological analysis or data modeling to provide risk warnings before an attack, but primarily relies on static defenses and lacks dynamic monitoring. During an attack, residual statistics or deep learning are used to detect data, but these methods lack robustness against targeted attacks and interpretability. Post-attack research focuses on data reconstruction, but research is fragmented and lacks a closed-loop defense system. Existing technologies face bottlenecks such as insufficient intelligent target selection, a lack of physical constraints, and a lack of closed-loop processes.

[0004] Therefore, a new solution is urgently needed. Summary of the Invention

[0005] The purpose of the present invention is to provide a method for identifying attack targets in power systems based on a customized generative model. This method transforms attack vector design into an optimization problem with physical constraints, thereby enhancing the physical feasibility and concealment of the attack vector. By forming a closed-loop framework through a fixed instruction logic system, the stability and consistency of the model execution are ensured, and the automated integration of the attack process is realized.

[0006] To achieve the above objectives, the present invention provides a method for identifying power system attack targets based on a customized generative model, comprising the following steps:

[0007] S1. Collect power system operation data and store it as an mpc structure in Matlab.mat format; identify the target to be attacked;

[0008] S2. Establish a physical constraint model and solve the physical constraint model to obtain the attack vector;

[0009] S3, using the power system communication network, superimposes the attack vector in S2 onto the real measurement data, implements an injection attack on the target to be attacked in S1, and verifies the attack effect.

[0010] Preferably, the mpc structure in S1 includes a bus matrix and a branch matrix. The bus matrix records the node number, node type, voltage amplitude, voltage phase angle, active injection power and reactive injection power node electrical characteristics; the branch matrix records the branch head and end node numbers, branch transmission power upper limit, real-time inflow active power, real-time inflow reactive power, real-time outflow active power and real-time outflow reactive power branch operating parameters.

[0011] Preferably, in S1, the target to be attacked is identified by calculating the branch load ratio. The branch load ratio LR is calculated as follows:

[0012]

[0013] Among them, P in P is the real-time active power flowing into the branch; out P is the real-time active power flowing out of the branch; max The upper limit of branch transmission power.

[0014] Preferably, in S1, the branch load rate is first calculated, then all branches are traversed, and then sorted in descending order according to the branch load rate, the branch with the highest branch load rate is selected as the target, and the unloaded branches with a branch load rate of 0 are filtered.

[0015] Preferably, in S2, the attack vector model construction is specifically expressed as follows:

[0016]

[0017] Where a is the one-dimensional norm of the attack vector; ΔP is the false disturbance value of the active power of the node; ΔQ is the false disturbance value of the reactive power of the node; ΔP ij is the false disturbance of active power flow in the branch; ΔQ ij False disturbances in reactive power flow in branches;

[0018] a=∑ i∈N |ΔP i |+∑ i∈N |ΔQ i |+∑ i∈N |ΔV i |+∑ i∈N |Δθ i |+∑ (i,j)∈ε |ΔP ij |+∑ (i,j)∈ε |ΔQ ij |;

[0019] Where ΔP i is the active power disturbance value of node i; ΔQ i is the reactive power disturbance value of node i; ΔV i is the voltage amplitude disturbance value of node i; Δθ i The voltage phase angle disturbance value of node i; ΔP ij Active power flow disturbance value of branch (i, j); ΔQ ij The reactive power flow disturbance value of branch (i, j);

[0020]

[0021] in, is the injected active power of node i; is the active power consumed by the load at node i; N r is the set of generator nodes; N is the set of all nodes;

[0022]

[0023] in, is the injected reactive power of node i; is the reactive power consumed by the load at node i;

[0024]

[0025] in, is the reactive power measurement value of node i; is the voltage amplitude measurement value of node i; is the voltage amplitude of the adjacent node j connected to node i; G ij is the conductance between node i and node j; B ij is the susceptance between node i and node j; is the voltage phase angle difference between node i and node j;

[0026]

[0027] in, is the measured active power value of node i;

[0028]

[0029] in, is the measured active power value of branch (i, j); X ij is the reactance of branch (i, j); ε is the set of all branches in the power system;

[0030]

[0031] in, is the reactive power measurement value of branch (i, j);

[0032]

[0033] in, is the upper limit of active power of branch (i, j); ε T is the target branch set;

[0034]

[0035] in, is the tampered measurement value; A is the true measurement value; ΔA is the false disturbance vector; P is the active power measurement value of the node; Q is the reactive power measurement value of the node; V is the voltage amplitude measurement value of the node; θ is the voltage phase angle measurement value of the node; active power.

[0036] The present invention also provides a power system attack target identification system based on a customized generative model, comprising an information acquisition module, an attack vector design module, and a false data injection module;

[0037] The information acquisition module builds a logic chain based on a customized generative model to achieve power system operation data collection and intelligent identification of attack targets;

[0038] The attack vector design module is connected to the information acquisition module and generates an attack vector that complies with power grid operation rules and is highly concealed through an optimization algorithm that incorporates physical constraints;

[0039] The false data injection module is connected to the attack vector design module and is used to apply the attack vector to the power system to trigger a preset attack effect.

[0040] Therefore, the present invention adopts the above-mentioned power system attack target identification method based on a customized generative model. Compared with the existing technology, the present invention has the following significant beneficial effects:

[0041] (1) The present invention intelligently identifies the attack targets most likely to cause overload based on the branch load ratio (the ratio of the transmission power flow to the power limit), replacing the traditional target selection method based on manual experience or heuristic rules;

[0042] (2) The present invention transforms the attack vector design into an optimization problem with physical constraints of the power system, incorporating constraints such as power conservation, power flow equations, and Kirchhoff's law. The goal is to minimize the L1 norm of the attack vector (i.e., minimize the disturbance amplitude), ensuring that the attack vector is physically feasible (in compliance with power grid operation rules). At the same time, the disturbance amplitude is reduced, enhancing concealment, and avoiding being identified by the detection mechanism due to excessive amplitude, thus solving the defect of "lack of physical constraints" in the existing technology.

[0043] (3) The present invention uses a closed-loop framework, which is different from the traditional GPT model, which is sensitive to input prompts and has large output fluctuations, and solves the defect of "lack of process closed loop" in the existing technology.

[0044] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 This is an architectural diagram of a method for identifying power system attack targets based on a customized generative model according to the present invention. DETAILED DESCRIPTION

[0046] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention. Unless otherwise defined, the technical terms or scientific terms used in the present invention should be the common meanings understood by people with ordinary skills in the field to which the present invention belongs.

[0047] Example 1

[0048] like Figure 1 As shown, a method for identifying attack targets of a power system based on a customized generative model of the present invention includes an information acquisition module, an attack vector design module and a false data injection module;

[0049] The information acquisition module builds a logic chain based on a customized generative model to achieve power system operation data collection and intelligent identification of attack targets;

[0050] The attack vector design module uses an optimization algorithm that incorporates physical constraints to generate attack vectors that comply with grid operation rules and are highly concealed.

[0051] The false data injection module applies the designed attack vector to the power system, triggering the preset attack effect.

[0052] The three modules form a complete attack chain of "power system information collection, intelligent identification of attack targets, physical constraint attack vector design, and false data injection", accurately targeting abnormal conditions such as power overload in the power system. The specific steps are as follows:

[0053] S1. Accurately collect and analyze power system data. The power system's operating state is complex, requiring the collection of multi-dimensional data to support attack analysis. This invention specifies that the data input format is Matlab.mat format (mpc structure), which is a standardized data carrier in the field of power system simulation and analysis. It can fully encapsulate the grid topology and real-time operating parameters, specifically including the bus matrix and branch matrix:

[0054] The bus matrix is used to store basic information about power system nodes, including fields such as node number (uniquely identifying grid nodes), node type (such as generator nodes, load nodes, and balance nodes; different types of nodes play significantly different roles in power flow calculations and attack responses), voltage amplitude (reflecting the stable state of node voltage, either per-unit or actual value, and can be used as a disturbance-related parameter in attacks), voltage phase angle (a key factor affecting power flow distribution; small phase angle changes can trigger power flow redistribution), active power injection (active power supply and demand at the node, and associated branch power transmission pressure), and reactive power injection (affecting node voltage regulation and system reactive power balance). For example, in the IEEE 14-node test system, the bus matrix clearly presents the electrical characteristics of each node, providing node-level basic data for attack target screening.

[0055] The branch matrix focuses on branch operation data. Key fields include the branch's head-end connection node number (which identifies the branch's physical connection relationship in the grid topology and clarifies the power transmission path), the branch's transmission power limit (determined by physical equipment constraints such as the line conductor's current carrying capacity and the transformer's rated capacity, and serves as the core threshold for determining branch overload risk, corresponding to column 6 of the matrix), real-time inflow active power (active power input at the branch's head end, reflecting the current power transmission intensity, corresponding to column 14 of the matrix), real-time inflow reactive power (reactive power input at the branch's head end, affecting node voltage and system reactive power flow, corresponding to column 15 of the matrix), real-time outflow active power (active power output at the branch's end end, corresponding to column 16 of the matrix), and real-time outflow reactive power (reactive power output at the branch's end end, corresponding to column 17 of the matrix). These data dynamically reflect the branch's power transmission status and are key to calculating branch load factors and identifying high-value targets for attack.

[0056] By standardizing the input of power system data, we ensure that the basic data required for attack analysis is complete and accurate, laying a solid data foundation for subsequent intelligent identification of attack targets.

[0057] Intelligent screening of attack targets based on load rate. The customized generative model follows a fixed logical process, deeply mines the value of branch matrix data, and accurately screens high-value FDIA attack targets. The specific steps are as follows:

[0058] The branch load rate (LR) is defined as the ratio of the absolute value of the branch transmission power flow to the power upper limit. The calculation formula is:

[0059]

[0060] Among them, P in is the real-time active power flowing into the branch (taken from the 14th column of the branch matrix); P out is the real-time outflow active power of the branch (taken from the 16th column of the branch matrix); P max is the upper limit of branch transmission power (taken from the 6th column of the branch matrix), reflecting the physical capacity boundary of the branch for safe operation.

[0061] The load rate is essentially a quantitative indicator of the "tension" of the branch's actual power transmission and physical capacity: the closer the load rate is to 1, the closer the branch is to its safe operating limit. At this time, if a tiny false data disturbance is injected, it will easily exceed the power limit, triggering the branch overload protection action or causing a large-scale redistribution of the power system flow, achieving the FDIA's attack goal of causing great damage with small disturbances. Therefore, this type of branch is a high-value target of FDIA.

[0062] The customized generative model traverses all branches of the power system, calculating the load factor for each branch and constructing a "branch number-load factor" association dataset. It then sorts the data in descending order by load factor, selecting the branches with the highest load factors as attack targets. For example, in a power system with 100 branches, the model automatically calculates the LR value for each branch. If the load factor of branch A is 0.92, branch B is 0.75, and the remaining branches are lower, branch A, being closest to the critical overload state, is selected as the priority attack target. Attacking such branches significantly reduces the amount of false data perturbation required, greatly improving the stealth of the attack.

[0063] At the same time, the model has built-in abnormal branch filtering logic, automatically ignoring branches with a load rate of 0 (such as no-load lines. Even if false data is injected into such branches, it is difficult to trigger power overload and has no attack value), focusing on high-value target calculations and improving recognition efficiency.

[0064] To ensure smooth integration of all links in the attack process, the customized generative model strictly adheres to a standardized output format. An example of the output result is: "The most vulnerable branch of this power system is XXX," where "XXX" corresponds to the unique identifier of the branch in the power system (usually composed of a combination of the branch's head and end node numbers, such as "Bus1-Bus2," which precisely matches the branch connection relationship field in the branch matrix).

[0065] This standardized design enables subsequent attack vector design modules (such as the YALMIP solver in Matlab) to directly parse the output results, quickly locate the target branch's index position in the branch matrix, and automatically associate the branch's physical constraint parameters (such as power limit and current power flow), providing precise input for attack vector mathematical modeling. Furthermore, during attack testing and defense review, technicians can use the standardized output to quickly locate the target branch in the power grid topology, analyze the attack path and system response, and improve the efficiency of attack and defense research.

[0066] S2. Power system operation strictly adheres to the laws of physics. If an attack vector violates these laws, it is easily detected by power system monitoring devices (such as state estimators and PMU measurement units). Therefore, this invention deeply embeds the physical laws of the power system into the attack vector design process to construct an attack "feasible space". The main constraints include:

[0067] Power conservation constraint: For any node in the power system, the injected power and the outflow power must satisfy the conservation relationship. The calculation formula is:

[0068] ∑P in +ΔP inj =∑P out ;

[0069] Among them, ∑P in is the total active power flowing into the node; ∑P out is the total active power flowing out of the node; ΔP inj is the false active power injected into the node (the active component of the attack vector); similarly, the reactive power must also satisfy:

[0070] ∑Q in +ΔQ inj =∑Q out ;

[0071] Among them, ∑Q in is the total reactive power flowing into the node; ∑Q out The total reactive power flowing out of the node; ΔQ inj is the false reactive power injected into the node.

[0072] The power flow equation constraint, the power system power flow distribution follows Kirchhoff's law. For AC systems, the branch power flow and node voltage and phase angle satisfy a nonlinear relationship. Taking the active power flow as an example, the calculation formula is:

[0073]

[0074] Among them, P ij is the active power flow from node i to node j; V i 、V j is the voltage amplitude of nodes i and j; δij is the voltage phase angle difference between nodes i and j; G ij 、B ij are the branch conductance and susceptance parameters; G ii is the self-conductance of node i; after the attack vector is injected, it is necessary to ensure that the power flow equation still holds, otherwise the attack will be detected due to abnormal power flow distribution.

[0075] Branch power limit constraint: Except for the target branch, the power transmission of other branches must be strictly lower than its power limit (that is, the load factor must be less than 1, unless the overload outside the target branch is the chain reaction that the attack intends to trigger, but conventional attacks are covert and will limit such situations). The calculation formula is:

[0076] max(|P in,k |,|P out,k |)≤P max,k ,k≠target;

[0077] Here, k is the non-target branch number, ensuring that the attack only acts on the target branch to cause overload, avoiding premature exposure of the attack due to abnormal overload of multiple branches.

[0078] By embedding physical constraints, attack vector design is transformed into a constrained optimization problem. That is, under the premise of satisfying the physical operation rules of the power grid, the minimum amplitude of false data that can trigger the overload of the target branch is found, achieving the dual goals of "feasible and concealed attack".

[0079] The concealment of the attack vector is key to the successful implementation of FDIA. If the false data disturbance amplitude is too large, it is easily detected by the power system bad data detection mechanism (such as state estimation detection based on residual analysis). Therefore, the optimization objective of this invention is to minimize the L1 norm of the attack vector, calculated as follows:

[0080]

[0081] Where Δx is the attack vector (including false disturbances to the measured quantities such as node voltage and power); Δx i is the i-th component of the attack vector; ||Δx||1 is the L1 norm, reflecting the overall disturbance amplitude of the attack vector.

[0082] The advantage of choosing the L1 norm is that, compared to the L2 norm, the L1 norm places greater emphasis on suppressing "large disturbance components," guiding the optimization algorithm to generate a combination of multi-dimensional, small disturbances rather than focusing on large disturbances in a few dimensions. This better meets the power system's need for covert attacks, which demand "small-amplitude, multi-node coordinated disturbances." For example, injecting false power into the three nodes associated with a target branch might reduce the active power disturbance to just 0.05 pu (per unit) per node after optimization. However, the coordinated effect can cause the target branch's load factor to exceed 1, triggering an overload. Furthermore, the amplitude of the disturbance at a single node is small, making it difficult to detect with conventional detection.

[0083] Based on the physical constraints of the power system, the attack vector model (power conservation, power flow equation, branch power limit) and the optimization objective (minimum L1 norm) are constructed, which can be expressed as follows:

[0084]

[0085] Where a is the one-dimensional norm of the attack vector; ΔP is the false disturbance value of the active power of the node; ΔQ is the false disturbance value of the reactive power of the node; ΔP ij is the false disturbance of active power flow in the branch; ΔQ ij False disturbances in reactive power flow in branches;

[0086] a=∑ i∈N |ΔP i |+∑ i∈N |ΔQ i |+∑ i∈N |ΔV i |+∑ i∈N |Δθ i |+∑ (i,j)∈ε |ΔP ij |+∑ (i,j)∈ε |ΔQ ij |;

[0087] Where ΔP i is the active power disturbance value of node i; ΔQ i is the reactive power disturbance value of node i; ΔV i is the voltage amplitude disturbance value of node i; Δθ i The voltage phase angle disturbance value of node i; ΔP ij Active power flow disturbance value of branch (i, j); ΔQ ij The reactive power flow disturbance value of branch (i, j);

[0088]

[0089] in, is the injected active power of node i; is the active power consumed by the load at node i; N ris the set of generator nodes; N is the set of all nodes;

[0090]

[0091] in, is the injected reactive power of node i; is the reactive power consumed by the load at node i;

[0092]

[0093] in, is the reactive power measurement value of node i; is the voltage amplitude measurement value of node i; is the voltage amplitude of the adjacent node j connected to node i; G ij is the conductance between node i and node j; B ij is the susceptance between node i and node j; is the voltage phase angle difference between node i and node j;

[0094]

[0095] in, is the measured active power value of node i;

[0096]

[0097] in, is the measured active power value of branch (i, j); X ij is the reactance of branch (i, j); ε is the set of all branches in the power system;

[0098]

[0099] in, is the reactive power measurement value of branch (i, j);

[0100]

[0101] in, is the upper limit of active power of branch (i, j); ε T is the target branch set;

[0102]

[0103] in, is the tampered measurement value; A is the true measurement value; ΔA is the false disturbance vector; P is the active power measurement value of the node; Q is the reactive power measurement value of the node; V is the voltage amplitude measurement value of the node; θ is the voltage phase angle measurement value of the node; active power.

[0104] S3. False data injection: The generated attack vector must be accurately injected into the power system measurement link (such as the SCADA system and the data acquisition end of the PMU device). In specific operations, the power system communication network (such as the IEC61850 protocol transmission channel) is used to superimpose false data on the real measurement data, so that the power system state estimation, scheduling decision-making and other links "receive" the tampered measurement value. For example, for the power measurement device of the target branch-related node, the false power value in the attack vector is superimposed on the real collection value through software-defined radio (SDR) technology or protocol stack attack methods, causing the power system control center to misjudge the node power state, and then use the wrong data in the power flow calculation and scheduling decision-making, triggering the target branch power overload.

[0105] After the false data is injected, the power system performs power flow calculations and dispatch decisions based on the tampered measurements. Because the attack vector is designed to target the high load characteristics of the target branch and meet physical constraints, the actual power flow transmitted by the target branch (including the false disturbance) will exceed the power limit, triggering branch overload protection (such as tripping the line circuit breaker) or causing power flow redistribution in the power system (causing cascading overload risks in other branches), achieving the FDIA attack's goal of disrupting the stable operation of the power system.

[0106] Verify the stealthiness of the attack, including that the attack vector satisfies rules such as power conservation and power flow equations, making it difficult for conventional "bad data detection" mechanisms of the power system (such as residual-based state estimation detection) to identify the residuals caused by false data within the normal fluctuation range (limited by the minimum L1 norm optimization); the attack vector L1 norm is minimized, so that the single-dimensional disturbance amplitude is far below the power system measurement error threshold, further evading detection.

[0107] By building a power system simulation platform (such as a collaborative simulation environment based on MATPOWER and Python), after injecting the attack vector, the system state estimation residuals, branch load rate changes, and detection system alarm conditions before and after the injection can be compared to verify the concealment and effectiveness of the attack.

[0108] Therefore, the present invention adopts the above-mentioned method for identifying power system attack targets based on a customized generative model. This method transforms the attack vector design into an optimization problem with physical constraints, enhancing the physical feasibility and concealment of the attack vector; by forming a closed-loop framework through a fixed instruction logic system, the stability and consistency of the model execution are ensured, and the automated integration of the attack process is realized.

[0109] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit the same. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solutions of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for identifying attack targets in a power system based on a customized generative model, characterized in that: The following steps are involved: S1. Collect power system operation data and store it as an mpc structure in Matlab.mat format; identify the target to be attacked; S2. Establish a physical constraint model and solve the physical constraint model to obtain the attack vector; S3, using the power system communication network, superimposes the attack vector in S2 onto the real measurement data, implements an injection attack on the target to be attacked in S1, and verifies the attack effect.

2. A method for identifying power system attack targets based on a customized generative model according to claim 1, characterized in that: The MPC structure in S1 includes a bus matrix and a branch matrix. The bus matrix records the node number, node type, voltage amplitude, voltage phase angle, active injection power, and reactive injection power node electrical characteristics; the branch matrix records the branch head and end node numbers, branch transmission power upper limit, real-time inflow active power, real-time inflow reactive power, real-time outflow active power, and real-time outflow reactive power branch operating parameters.

3. The method for identifying attack targets of a power system based on a customized generative model according to claim 2, characterized in that: In S1, the target to be attacked is identified by calculating the branch load ratio. The branch load ratio LR is calculated as follows: Among them, P in P is the real-time active power flowing into the branch; out P is the real-time active power flowing out of the branch; max The upper limit of branch transmission power.

4. The method for identifying attack targets of a power system based on a customized generative model according to claim 3, characterized in that: In S1, the branch load rate is first calculated, and then all branches are traversed and sorted in descending order according to the branch load rate. The branch with the highest branch load rate is selected as the target, and the unloaded branches with a branch load rate of 0 are filtered out.

5. The method for identifying attack targets of a power system based on a customized generative model according to claim 4, characterized in that: In S2, the attack vector model is constructed as follows: Where a is the one-dimensional norm of the attack vector; ΔP is the false disturbance value of the active power of the node; ΔQ is the false disturbance value of the reactive power of the node; ΔP ij is the false disturbance of active power flow in the branch; ΔQ ij False disturbances in reactive power flow in branches; a=∑ i∈N |ΔP i |+∑ i∈N |DQ i |+∑ i∈N |ΔV i |+∑ i∈N |Dth i |+ ∑ (i,j)∈ε |ΔP ij |+∑ (i,j)∈ε |ΔQ ij |; Where ΔP i is the active power disturbance value of node i; ΔQ i is the reactive power disturbance value of node i; ΔV i is the voltage amplitude disturbance value of node i; Δθ i The voltage phase angle disturbance value of node i; ΔP ij Active power flow disturbance value of branch (i, j); ΔQ ij The reactive power flow disturbance value of branch (i, j); in, is the injected active power of node i; is the active power consumed by the load at node i; N r is the set of generator nodes; N is the set of all nodes; in, is the injected reactive power of node i; is the reactive power consumed by the load at node i; in, is the reactive power measurement value of node i; is the voltage amplitude measurement value of node i; is the voltage amplitude of the adjacent node j connected to node i; G ij is the conductance between node i and node j; B ij is the susceptance between node i and node j; is the voltage phase angle difference between node i and node j; in, is the measured active power value of node i; in, is the measured active power value of branch (i, j); X ij is the reactance of branch (i, j); ε is the set of all branches in the power system; in, is the reactive power measurement value of branch (i, j); in, is the upper limit of active power of branch (i, j); ε T is the target branch set; in, is the tampered measurement value; A is the true measurement value; ΔA is the false disturbance vector; P is the active power measurement value of the node; Q is the reactive power measurement value of the node; V is the voltage amplitude measurement value of the node; θ is the voltage phase angle measurement value of the node; active power.

6. A power system attack target identification system based on a customized generative model, including an information acquisition module, an attack vector design module, and a false data injection module; The information acquisition module builds a logic chain based on a customized generative model to achieve power system operation data collection and intelligent identification of attack targets; The attack vector design module is connected to the information acquisition module and generates an attack vector that complies with power grid operation rules and is highly concealed through an optimization algorithm that incorporates physical constraints; The false data injection module is connected to the attack vector design module and is used to apply the attack vector to the power system to trigger a preset attack effect.

Images