Methods, devices, equipment, and media for extrapolating cross-domain attacks on power cyber-physical systems

By acquiring multi-source knowledge from the power information physical system for credibility verification and converting it into executable rule templates, and constructing cross-domain coupled hypergraphs and causal graphs, the problems of difficulty in executing multi-source heterogeneous knowledge and difficulty in expressing high-order coupling relationships are solved. This enables accurate deduction of attack paths and verification of physical feasibility, and improves the accuracy and credibility of the deduction results.

CN123074801APending Publication Date: 2026-11-10CHINA ELECTRIC POWER RESEARCH INSTITUTE CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611318915.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-28
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

In existing cross-domain attack simulation methods for power cyber-physical systems, multi-source heterogeneous knowledge lacks a unified executable mechanism, making it difficult to directly drive attack path generation. High-order coupling relationships are difficult to express, and there is a lack of dual constraints from cross-domain coupling topology and physical mechanisms, resulting in simulation results that do not fit real-world scenarios.

Method used

By acquiring multi-source knowledge for credibility verification, converting it into executable rule templates, constructing cross-domain coupled hypergraphs and causal graphs, generating candidate hyperedges and causal edges, generating causal edges only between nodes that have been verified to support hyperedges and whose corresponding constraints are met, performing attack path deduction, and conducting physical feasibility verification and closed-loop calibration.

Benefits of technology

This improves the accuracy and reliability of attack path projection results, ensuring that path generation conforms to the actual connection relationships and business logic of the power information physical system, thus enhancing the accuracy and reliability of the projection results.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123074801A_ABST
    Figure CN123074801A_ABST
Patent Text Reader

Abstract

This invention relates to the field of power system network security technology, and discloses a method, apparatus, device, and medium for cross-domain attack simulation in power cyber-physical systems. The method includes: acquiring knowledge entries corresponding to multi-source knowledge, performing credibility verification, and converting the knowledge entries into executable rule templates; generating candidate hyperedges based on the rule templates and scenario nodes, performing preset verification, and writing the candidate hyperedges as verified hyperedges into a cross-domain coupled hypergraph; generating candidate causal edges only between nodes supported by verified hyperedges and satisfying corresponding constraints, and constructing a cross-domain causal graph; deriving attack paths based on the cross-domain causal graph; verifying the physical feasibility of the attack paths, and performing closed-loop calibration based on the attack path simulation results and the physical feasibility verification results. This invention solves the technical problems in cross-domain attack simulation, such as the difficulty in executing knowledge, the difficulty in expressing high-order coupling relationships, the lack of domain constraints on causal paths, and the difficulty in closed-loop calibration of simulation results.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of power system network security technology, specifically to methods, apparatus, equipment, and media for cross-domain attack simulation of power cyber-physical systems. Background Technology

[0002] As the cyber-physical integration of power systems deepens, cross-domain cyberattacks have become a key risk factor threatening the safe operation of power systems. Attackers typically exploit vulnerabilities on the information side to penetrate the system, and then transmit the impact to the physical domain through methods such as communication protocol tampering and control command forgery, ultimately causing serious consequences such as equipment failure and power grid instability.

[0003] Existing attack simulation methods mainly include attack graph-based, knowledge graph-based, Bayesian network-based, and power simulation-based techniques. Attack graph-based schemes use binary edges to represent vulnerability exploitation or state transition relationships. When a cross-domain transition requires multiple conditions to be met simultaneously, such as communication reachability, protocol state, control permissions, device binding, and physical topology, the binary edge structure needs to decompose the joint premise into multiple local relations, making it difficult to directly express high-order dependencies under the common constraints of multiple nodes. Knowledge graph-based schemes mainly focus on the storage and retrieval of entity relationships, lacking a mechanism to convert knowledge entries into executable rules and inference constraints, making it difficult to drive the dynamic generation of cross-domain paths. While power simulation-based schemes can verify physical impacts, they still need to establish verifiable mapping relationships between information-side events, control actions, and physical objects.

[0004] Existing cross-domain attack simulation methods for power information physical systems have the following shortcomings: First, multi-source heterogeneous knowledge lacks a unified executable mechanism, making it difficult to directly drive attack path generation; second, there is a lack of high-order topology models that can express the common triggering relationship of multiple nodes in different domains, which easily generates paths that do not conform to business logic or system constraints; third, the generation of causal paths lacks the dual constraints of cross-domain coupled topology and physical mechanism, resulting in simulation results that do not fit the real scenario. Summary of the Invention

[0005] This invention provides a method, apparatus, device, and medium for cross-domain attack simulation of power cyber-physical systems, in order to solve the problems of difficulty in execution, difficulty in expressing high-order coupling relationships, lack of domain constraints on causal paths, and difficulty in closed-loop calibration of simulation results in the prior art.

[0006] In a first aspect, the present invention provides a method for extrapolating cross-domain attacks on power cyber-physical systems, the method comprising: Obtain knowledge entries corresponding to multi-source knowledge, verify the credibility of the knowledge entries, and convert the verified knowledge entries into executable rule templates; Candidate hyperedges are generated based on the rule template and preset scenario nodes. Preset verification is performed on the candidate hyperedges, and the verified candidate hyperedges are written as verified hyperedges into the cross-domain coupling hypergraph. The cross-domain coupling hypergraph is used to represent the many-to-many coupling relationship between multi-dimensional entities in the power information physical system. Candidate causal edges are generated only between nodes where the superedge support has been verified and the corresponding constraints are satisfied, and a cross-domain causal graph is constructed based on the candidate causal edges; Attack paths are derived from cross-domain causal graphs to assess the security risks of power cyber-physical systems. The physical feasibility of the attack path is verified, and closed-loop calibration is performed based on the attack path deduction results and the physical feasibility verification results.

[0007] This invention provides a method for cross-domain attack inference in power cyber-physical systems. By converting multi-source knowledge into executable rule templates, knowledge can directly drive attack path generation. By constructing a cross-domain coupling hypergraph to represent many-to-many coupling relationships between multi-dimensional entities, the method solves the problem of expressing high-order cross-domain coupling relationships. By generating causal edges under the constraints of the cross-domain coupling hypergraph to construct a cross-domain causal graph, the generation of causal paths is constrained by the cross-domain coupling topology, thereby improving the accuracy and reliability of attack path inference results. This method solves the problems of difficulty in execution, difficulty in expressing high-order coupling relationships, lack of domain constraints on causal paths, and difficulty in closed-loop calibration of inference results in existing technologies.

[0008] In one alternative implementation, the credibility verification of knowledge entries includes: The knowledge entries are verified for field completeness, source credibility, and logical conflict. The verified knowledge entries are used to generate executable rule templates.

[0009] In the above technical solution, by verifying the completeness of fields, the credibility of sources, and the credibility of logical conflicts of knowledge entries, knowledge with complete fields, credible sources, and no logical conflicts is selected for generating rule templates. This ensures that the knowledge entries participating in the subsequent deduction process have completeness and reliability, reduces the possibility of invalid path generation due to incomplete knowledge, unreliable sources, or logical conflicts, and improves the accuracy of the deduction results.

[0010] In one alternative implementation, the knowledge entry includes: preconditions, applicable objects, result objects, constraints, knowledge sources, and verification status; Convert validated knowledge entries into executable rule templates, including: Based on the source of knowledge, assess the credibility of the source; Whether conversion is allowed depends on the verification status and the credibility of the source. If the prerequisites, applicable objects, result objects, and knowledge sources are complete, the verification status is verified, and the credibility of the source meets the preset conditions, then the prerequisites are mapped to the prerequisites of the rule template, the applicable objects are mapped to the node selectors of the rule template, the result objects are mapped to the result node types of the rule template, and the constraints are mapped to the verification strategies of the rule template, thereby generating an executable rule template; otherwise, the knowledge items are included in the knowledge set to be supplemented.

[0011] In the above technical solution, by mapping the prerequisites, applicable objects, result objects, and constraints of knowledge items to the prerequisites, node selectors, result node types, and verification strategies of rule templates, natural language or unstructured knowledge is transformed into structured rules that can directly participate in the deduction process. At the same time, the transformation is only performed when the prerequisites, applicable objects, result objects, and knowledge sources of a knowledge item are complete, the verification status is verified, and the credibility of the source meets the preset conditions; otherwise, it is included in the supplementary set. This ensures that the rule templates entering the deduction process have complete structural support and credible source evidence.

[0012] In one alternative implementation, the cross-domain coupling hypergraph is constructed as follows: The nodes of the cross-domain coupling hypergraph are determined. The nodes include information domain nodes, communication domain nodes, control domain nodes, physical domain nodes, and business impact domain nodes. The nodes are derived from the scene nodes and event state nodes obtained after instantiating the atomic models. The atomic models include physical device class atomic models, information component class atomic models, behavioral subject class atomic models, and business object class atomic models. Determine the hyperedges of the cross-domain coupling hypergraph, where each hyperedge connects two or more nodes, to represent the many-to-many coupling relationships between multi-dimensional entities in the power information physical system; Set hyperedge coupling weights, semantic labels, and domain identifiers for each hyperedge to generate a cross-domain coupled hypergraph.

[0013] In the above technical solution, by dividing nodes into five domains—information domain, communication domain, control domain, physical domain, and business impact domain—and determining hyperedges connecting two or more nodes, the cross-domain coupling hypergraph can directly represent the many-to-many coupling relationships between multi-dimensional entities belonging to different domains in the power information physical system. At the same time, by setting hyperedge coupling weights, semantic labels, and domain identifiers for each hyperedge, the coupling relationships carried by the hyperedge have quantifiable attribute information and domain affiliation information, thereby providing domain constraints and topological support for the subsequent generation of cross-domain causal edges.

[0014] In one optional implementation, candidate hyperedges are generated based on a rule template and preset scene nodes. Preset verification is performed on the candidate hyperedges, and the verified candidate hyperedges are written as verified hyperedges into the cross-domain coupled hypergraph, including: Construct a network attack scenario template based on the preset attack target, attack entry point, target object, attack stage, and scope; Instantiate scene nodes from the atomic model library based on the network attack scenario template; Match the node selector in the rule template with the scene nodes to generate candidate hyperedges; Candidate hyperedges are subject to pre-defined verification. The pre-defined verification includes: logical verification, temporal verification, data verification, defense verification, and physical verification. If a candidate hyperedge connects to nodes in different domains, cross-domain transfer condition verification is also performed. The verified candidate hyperedges are written into the cross-domain coupled hypergraph as verified hyperedges.

[0015] In the above technical solution, scenario nodes are instantiated from the atomic model library according to the network attack scenario template, and the matching result between the node selector in the rule template and the scenario node is used as the basis for generating candidate hyperedges, so that the generation of candidate hyperedges is constrained by the attack target and scope. By performing logical, temporal, data, defense and physical verification on candidate hyperedges, and additionally performing cross-domain transfer condition verification when connecting nodes in different domains, candidate hyperedges are screened from multiple dimensions, which can reduce the number of candidate hyperedges that do not conform to business logic, temporal order, data integrity, defense strategy and physical mechanism from entering the cross-domain coupling hypergraph. The verified candidate hyperedges are written into the hypergraph as verified hyperedges, so that the hyperedges in the hypergraph have multi-dimensional reliability support.

[0016] In an optional implementation, generating candidate hyperedges based on a rule template and preset scene nodes further includes: Calculate the initial weights of the candidate superedges. The initial weights are jointly determined by the prior weights of the rule template and the credibility of the knowledge source.

[0017] In the above technical solution, the initial weight of the candidate hyperedge is determined by the prior weight of the rule template and the credibility of the knowledge source. This allows the weight of the hyperedge to reflect both the support strength of the rule itself and the reliability of the knowledge source, avoiding the bias caused by single-dimensional weighting. As a result, the weight of the candidate hyperedge has a more comprehensive reference basis in the subsequent inference path scoring.

[0018] In one alternative implementation, candidate causal edges are generated only between nodes where the hyperedge support has been verified and the corresponding constraints are satisfied, and a cross-domain causal graph is constructed based on the candidate causal edges, including: Retrieve nodes, verified hyperedges, and associated rule templates from the cross-domain coupled hypergraph; If any two nodes are selected, and the first and second nodes are supported by the same verified superedge, and the rule template supports propagation from the first node to the second node, then a corresponding candidate causal edge is generated; otherwise, no candidate causal edge is generated. When multiple nodes form a reachable relationship through multiple verified hyperedges, candidate causal edges are generated segment by segment according to the adjacent node pairs supported by each verified hyperedge, and candidate causal edges that cross intermediate nodes are not generated. Perform hypergraph support verification, causal type consistency verification, direction consistency verification, and constraint verification on candidate causal edges. Among them, candidate causal edges connecting nodes in different domains must also satisfy the corresponding cross-domain transition conditions. The verified candidate causal edges are used as cross-domain causal edges, and a cross-domain causal graph is constructed based on the cross-domain causal edges and the nodes in the cross-domain coupling hypergraph.

[0019] In the above technical solution, verified hyperedges in the cross-domain coupling hypergraph are used as constraints for generating candidate causal edges. Candidate causal edges are only allowed to be generated when any two nodes simultaneously belong to the same verified hyperedge or are reachable by multiple verified hyperedges. This restricts the generation range of causal edges to the coupling relationships carried by the hypergraph topology. By performing hypergraph support, causal type consistency, direction consistency, and constraint condition admission verification on the candidate causal edges, the candidate causal edges are further screened. This ensures that each causal edge in the final generated cross-domain causal graph has hypergraph topology support and rule-level compliance, thereby ensuring that the causal graph structure does not deviate from the actual connection relationships and business logic constraints of the power information physical system.

[0020] In an alternative implementation, generating candidate causal edges only between nodes where the hyperedge support has been verified and the corresponding constraints are satisfied further includes: Determine the type of candidate causal edge, and classify candidate causal edges into hard-constrained edges or calibrable edges based on the constraint source and allowed update range of the candidate causal edge; The types include at least one of the following: attack propagation edge, control action edge, physical influence edge, defense blocking edge, environmental disturbance edge, and business influence edge; Hard constraint edges originate from electrical physics mechanisms, mandatory safety constraints, protection action logic, equipment technical manuals, or standard specifications. During closed-loop calibration, hard constraint edges must not be deleted, and the conduction direction and hard constraint conditions of hard constraint edges must not be changed. The causal correlation strength, confidence level, and propagation delay of calibrable edges can be updated within a preset range based on the attack path deduction results and physical feasibility verification results. The updated calibrable edges still need to be supported by verified superedges and must not violate hard constraints.

[0021] In the above technical solution, by classifying candidate causal edges into different types such as attack propagation edges, control action edges, physical influence edges, defense blocking edges, environmental disturbance edges, and business influence edges, the causal graph can distinguish attack propagation relationships of different natures and match the corresponding triggering conditions and extension rules respectively. By further classifying causal edges into hard constraint attributes and calibrable attributes, the relationships derived from power physics mechanisms, mandatory safety constraints, protection action logic, equipment technical manuals, or standard specifications are retained as unchangeable hard constraint attributes. At the same time, the causal correlation strength, confidence, and propagation delay of calibrable attributes are allowed to be updated according to the inference results and physical verification results, thereby enabling the model to adapt to new data updates while keeping the basic constraints unchanged.

[0022] In one optional implementation, the attack path is derived from the cross-domain causality graph, including: Based on the network attack scenario template, determine the attack entry node and the attack target node, set the attack entry node as the initial active node, and set the initial active node as the current active node; Starting from the currently active node, retrieve cross-domain causal edges originating from the currently active node; When a cross-domain causal edge is found to connect nodes in different domains, the cross-domain transfer conditions recorded by the verified superedge associated with the cross-domain causal edge are obtained, and the cross-domain transfer conditions are checked before triggering the endpoint node; when all necessary conditions are met, the trigger judgment continues; when any necessary condition is not met, the current attack path is stopped from extending along the cross-domain causal edge, and the unmet conditions and their associated rule templates are recorded. Based on the type of cross-domain causal edge, the strength of causal association, confidence level, constraints, and defense status, determine whether to trigger the endpoint node pointed to by the cross-domain causal edge; If triggered, the cross-domain causal edge and its corresponding endpoint node are recorded, and the endpoint node is set as the currently active node. The deduction continues until the attack target node is reached or the preset stopping condition is met. Attack paths are generated based on the activation nodes and cross-domain causal edges recorded during the simulation process.

[0023] In the above technical solution, the attack path is deduced by starting from the attack entry node and proceeding hop by hop along the cross-domain causal edge. The determination of whether to trigger the next node is based on the type of the cross-domain causal edge, the strength of the causal relationship, the confidence level, the constraints, and the defense status. This ensures that the deduction process of the attack path is constrained by the causal relationships in the cross-domain causal graph. At the same time, by setting the attack target node and the preset stopping condition as the deduction endpoint, the path deduction terminates when the target is reached or the stopping condition is met. Finally, the attack path is generated based on the active nodes and cross-domain causal edges that are actually triggered during the deduction process, so that each step in the generated path has a clear causal basis and triggering condition.

[0024] In one alternative implementation, the physical feasibility of the attack path is verified, including: When there are control action edges or physical influence edges in the attack path, the attack event is mapped to a physical action that can be identified by the power system, and the physical action is verified for power physical feasibility. The physical feasibility of the attack path is marked according to the verification results. The power physical feasibility verification is the verification of physical actions using topology constraints, power flow constraints, protection logic constraints, equipment state constraints and / or power simulation.

[0025] In the above technical solution, when there are control action edges or physical influence edges in the attack path, the attack event is mapped to a physical action that can be identified by the power system. The physical action is verified for power physical feasibility using topology constraints, power flow constraints, protection logic constraints, equipment state constraints and / or power simulation, so that the determination of the physical impact of the attack path has a basis at the level of power system operation mechanism. The physical feasibility of the attack path is marked according to the verification results, so that paths that are reachable from the network but lack physical support are not misjudged as paths with physical impact, thereby improving the accuracy of the attack path evaluation results.

[0026] In one optional implementation, closed-loop calibration is performed based on the attack path deduction results and physical feasibility verification results, including: Based on the consistency between the attack path deduction results and the physical feasibility verification results, the calibrable attributes of the causal edges are updated, while keeping the hard constraints determined by the protocol premises, device binding, physical topology and protection logic unchanged. The rule template, cross-domain coupling hypergraph and cross-domain causal graph are modified according to the update results to perform closed-loop calibration.

[0027] In the above technical solution, the calibrable attributes are updated based on the consistency between the inference results of the attack path and the physical feasibility verification results. The updated results are then fed back to the rule template, cross-domain coupling hypergraph, and cross-domain causal graph for correction. This allows the rule template, hypergraph structure, and causal edge attributes to be corrected in reverse based on the actual results of the inference and verification, thereby providing a more accurate knowledge and model foundation for subsequent inferences.

[0028] Secondly, the present invention provides a cross-domain attack simulation device for power cyber-physical systems, the device comprising: The knowledge entry acquisition module is used to acquire knowledge entries corresponding to knowledge from multiple sources. The causal model construction and management module is used to verify the credibility of knowledge entries and convert verified knowledge entries into executable rule templates. Based on the rule templates and preset scenario nodes, candidate hyperedges are generated, preset verifications are performed on the candidate hyperedges, and verified candidate hyperedges are written as verified hyperedges into the cross-domain coupling hypergraph. The cross-domain coupling hypergraph is used to represent the many-to-many coupling relationships between multi-dimensional entities in the power information physical system. Candidate causal edges are generated only between nodes supported by verified hyperedges and whose corresponding constraints are satisfied, and a cross-domain causal graph is constructed based on the candidate causal edges. The cross-domain attack simulation and deduction module is used to deduce attack paths based on cross-domain causal graphs, and to assess the security risks of the power cyber-physical system based on the attack paths. The physical verification and closed-loop calibration module is also used to verify the physical feasibility of attack paths and perform closed-loop calibration based on the attack path deduction results and physical feasibility verification results.

[0029] In one alternative implementation, the causal model building and management module includes: The credibility verification unit is used to verify the credibility of knowledge entries by checking field integrity, source credibility, and logical conflicts. Knowledge entries that pass the verification are used to generate executable rule templates.

[0030] In one optional implementation, the knowledge entries include: preconditions, applicable objects, result objects, constraints, knowledge sources, and verification status; the causal model construction and management module includes: The credibility reading unit is used to read the credibility of the source based on the knowledge source. The judgment unit is used to determine whether conversion is allowed based on the verification status and source credibility. The rule template generation unit is used to generate an executable rule template if the prerequisites, applicable objects, result objects, and knowledge sources are complete, the verification status is verified, and the source credibility meets the preset conditions. Otherwise, the knowledge items are included in the knowledge set to be supplemented.

[0031] In one alternative embodiment, the device further includes: The cross-domain coupling hypergraph construction module is used to determine the nodes of the cross-domain coupling hypergraph. The nodes are derived from scene nodes and event state nodes obtained by instantiating atomic models. The atomic models include physical device atomic models, information component atomic models, behavioral subject atomic models, and business object atomic models. The nodes include information domain nodes, communication domain nodes, control domain nodes, physical domain nodes, and business influence domain nodes. The module determines the hyperedges of the cross-domain coupling hypergraph. Each hyperedge connects two or more nodes and is used to represent the many-to-many coupling relationship between multi-dimensional entities in the power information physical system. The module sets the hyperedge coupling weight, semantic label, and domain identifier for each hyperedge to generate the cross-domain coupling hypergraph.

[0032] In one alternative implementation, the causal model building and management module further includes: The attack scenario template building unit is used to build network attack scenario templates based on preset attack targets, attack entry points, target objects, attack stages, and scope of action. Instantiation unit, used to instantiate scene nodes from the atomic model library based on the network attack scenario template; The candidate hyperedge generation unit is used to match the node selector in the rule template with the scene nodes to generate candidate hyperedges; The multi-dimensional verification unit is used to perform pre-defined verification on candidate hyperedges. The pre-defined verification includes: logical verification, temporal verification, data verification, defense verification, and physical verification. If the candidate hyperedge connects to nodes in different domains, cross-domain transfer condition verification is also performed. The write unit is used to write the verified candidate hyperedges as verified hyperedges into the cross-domain coupled hypergraph.

[0033] In one alternative implementation, the causal model building and management module further includes: The initial weight calculation unit is used to calculate the initial weights of candidate hyperedges. The initial weights are jointly determined by the prior weights of the rule template and the credibility of the knowledge source.

[0034] In one alternative implementation, the causal model building and management module further includes: The data acquisition unit is used to acquire nodes, verified hyperedges, and associated rule templates in the cross-domain coupled hypergraph. The candidate causal edge judgment unit is used to select any two nodes. If the first node and the second node are supported by the same verified superedge, and the rule template supports the propagation from the first node to the second node, a corresponding candidate causal edge is generated; otherwise, no candidate causal edge is generated. The candidate causal edge generation unit is used to generate candidate causal edges segment by segment according to the adjacent node pairs supported by each verified hyperedge when multiple nodes form a reachable relationship through multiple verified hyperedges, without generating candidate causal edges that cross intermediate nodes; the causal edge admission verification unit is used to perform hypergraph support verification, causal type consistency verification, direction consistency verification and constraint condition verification on candidate causal edges. Among them, candidate causal edges connecting nodes in different domains must also satisfy the corresponding cross-domain transition conditions. The cross-domain causal graph construction unit is used to take the verified candidate causal edges as cross-domain causal edges and construct a cross-domain causal graph based on the cross-domain causal edges and the nodes in the cross-domain coupling hypergraph.

[0035] In one alternative implementation, the causal model building and management module further includes: The candidate causal edge type determination unit is used to determine the type of candidate causal edge and classify the candidate causal edge into hard constraint edge or calibrable edge according to the constraint source and allowed update range of the candidate causal edge. The types include at least one of the following: attack propagation edge, control action edge, physical influence edge, defense blocking edge, environmental disturbance edge, and business influence edge; Hard constraint edges originate from electrical physics mechanisms, mandatory safety constraints, protection action logic, equipment technical manuals, or standard specifications. During closed-loop calibration, hard constraint edges must not be deleted, and the conduction direction and hard constraint conditions of hard constraint edges must not be changed. The causal correlation strength, confidence level, and propagation delay of calibrable edges can be updated within a preset range based on the attack path deduction results and physical feasibility verification results. The updated calibrable edges still need to be supported by verified superedges and must not violate hard constraints.

[0036] In one optional implementation, the cross-domain attack simulation module includes: The node determination unit is used to determine the attack entry node and the attack target node according to the attack scenario template, set the attack entry node as the initial activation node, and set the initial activation node as the current activation node. The retrieval unit is used to retrieve cross-domain causal edges starting from the currently active node. The cross-domain transfer judgment unit is used to obtain the cross-domain transfer conditions recorded by the verified super-edge associated with the cross-domain causal edge when the retrieved cross-domain causal edge connects nodes in different domains, and check the cross-domain transfer conditions before triggering the endpoint node; when all necessary conditions are met, the trigger judgment continues to be executed; when any necessary condition is not met, the current attack path is stopped from extending along the cross-domain causal edge, and the unmet conditions and their associated rule templates are recorded. The trigger judgment unit is used to determine whether to trigger the endpoint node pointed to by the cross-domain causal edge based on the type of the cross-domain causal edge, the strength of the causal relationship, the confidence level, the constraint conditions, and the defense status. The deduction unit is used to record the cross-domain causal edge and the corresponding endpoint node if triggered, and set the endpoint node as the currently active node, and continue the deduction until the attack target node is reached or the preset stopping condition is met. The attack path generation unit is used to generate attack paths based on the active nodes and cross-domain causal edges recorded during the simulation process.

[0037] In one alternative implementation, the physical verification and closed-loop calibration module includes: The physical feasibility verification unit is used to map attack events into physical actions that can be identified by the power system when there are control action edges or physical influence edges in the attack path, and to perform power physical feasibility verification on the physical actions. The physical feasibility of the attack path is marked according to the verification results. Power physical feasibility verification is the verification of physical actions using topology constraints, power flow constraints, protection logic constraints, equipment state constraints and / or power simulation.

[0038] In one optional implementation, the physical verification and closed-loop calibration module further includes: The model calibration and update unit is used to update calibrable attributes based on the consistency between the attack path deduction results and the physical feasibility verification results, keeping the hard constraints determined by the protocol premises, device binding, physical topology and protection logic unchanged, and correcting the rule template, cross-domain coupling hypergraph and cross-domain causal graph based on the update results.

[0039] Thirdly, the present invention provides an electronic device, comprising: a memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the computer instructions to perform the cross-domain attack simulation method for power cyber-physical systems described in the first aspect or any corresponding embodiment.

[0040] Fourthly, the present invention provides a computer-readable storage medium storing computer instructions for causing a computer to execute the cross-domain attack simulation method for power cyber-physical systems described in the first aspect or any corresponding embodiment thereof.

[0041] Fifthly, the present invention provides a computer program product, including computer instructions, which are used to cause a computer to execute the cross-domain attack simulation method for power cyber-physical systems described in the first aspect or any corresponding embodiment. Attached Figure Description

[0042] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0043] Figure 1 This is a structural block diagram of a cross-domain attack simulation system for a power cyber-physical system according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the first type of cross-domain attack simulation method for power cyber-physical systems according to an embodiment of the present invention; Figure 3 This is a schematic diagram of the second type of the cross-domain attack simulation method for power cyber-physical systems according to an embodiment of the present invention; Figure 4 This is a schematic diagram of the third process of the cross-domain attack simulation method for power cyber-physical systems according to an embodiment of the present invention; Figure 5 This is a flowchart of cross-domain coupled hypergraph construction according to an embodiment of the present invention; Figure 6 This is a flowchart illustrating the cross-domain causal graph construction and attack process according to an embodiment of the present invention; Figure 7 This is a structural block diagram of a cross-domain attack simulation device for a power cyber-physical system according to an embodiment of the present invention; Figure 8 This is a schematic diagram of the hardware structure of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0044] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0045] It is understood that before using the technical solutions disclosed in the various embodiments of the present invention, users should be informed of the types, scope of use, and usage scenarios of the personal information involved in the present invention and their authorization should be obtained in accordance with relevant laws and regulations through appropriate means.

[0046] The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0047] As one application scenario of this invention, such as Figure 1 The diagram shown is a structural block diagram of the cross-domain attack simulation system for power cyber-physical systems provided by the present invention. The system includes: The system comprises a power knowledge base, a causal model construction and management module, a cross-domain attack simulation module, a probabilistic perturbation algorithm unit, and a simulation environment interference module. These modules form a closed-loop data flow through knowledge entries in the power knowledge base, atomic models, cross-domain coupling hypergraphs, cross-domain causal graphs, simulation logs, and calibration results.

[0048] The system includes: 1) A power knowledge base: storing knowledge in the power sector, power safety, equipment parameters, business logic, protocol rules, defense strategies, environmental disturbances, and probabilistic disturbance baselines. Each piece of knowledge includes its source, credibility score, version number, and verification status. The power knowledge base provides the causal attribute baseline values, constraint thresholds, and equipment action thresholds to the causal model construction and management module; attack chain rules, network modeling rules, and defense strategy rules to the cross-domain attack simulation and deduction module; and environmental disturbance baseline values ​​and equipment tolerance thresholds to the environmental disturbance simulation and deduction module.

[0049] 2) Causal Model Construction and Management Module: This module includes an atomic model management unit, a cross-domain causal graph construction unit, a model calibration and update unit, and a probabilistic perturbation algorithm unit. The atomic model management unit maintains standardized models of physical devices, information components, behavioral subjects, and business objects; the cross-domain causal graph construction unit generates causal edges under the constraints of a cross-domain coupled hypergraph; the model calibration and update unit updates the attributes of causal edges based on the inference results and drives the hypergraph weight update in reverse; and the probabilistic perturbation algorithm unit performs constrained perturbations on causal strength, confidence, and propagation delay.

[0050] 3) Cross-Domain Attack Simulation Module: This module includes a network attack scenario design unit, a network attack process simulation unit, and a network attack impact calculation and physical verification unit. The network attack scenario design unit generates standardized scenario templates and constructs a cross-domain coupled hypergraph; the network attack process simulation unit performs path expansion, scoring, and blocking judgments under verified hyperedge and cross-domain transfer constraints; and the network attack impact calculation and physical verification unit establishes the mapping between control actions and physical objects, outputting business impact, physical feasibility, and verification basis.

[0051] 4) Environmental Interference Simulation Module: This module is used to inject non-attack-related disturbances, including device-level interference, communication-level interference, data-level interference, business process disturbances, and personnel operation disturbances. This module only changes the existing node state, superedge weights, causal edge delays, or disturbance coefficients; it does not generate new attack logic or new electrical physical relationships.

[0052] The system can also be configured with optional probabilistic perturbation and environmental interference functions. Perturbation parameters are read from the system configuration, and the allowed objects, attributes, value ranges, and time windows are limited; attributes without a configured allowable range are not perturbed. The above functions only adjust the calibrable attributes of verified relationships and do not generate new nodes, hyperedges, causal edges, or attack logic.

[0053] The system also includes: The power simulation and physical verification module is used to verify the feasibility of attack paths derived from cross-domain attack simulations at the power physics level. When there are control action edges or physical impact edges in the attack path, this module maps the attack event to physical actions identifiable by the power system (such as circuit breaker opening and closing, protection tripping, etc.), and uses topology constraints, power flow constraints, protection logic constraints, equipment state constraints, and / or power simulation models to verify the physical actions to determine whether they conform to the actual operating mechanism of the power system. Finally, based on the verification results, the attack path is divided into physically feasible paths, conditionally feasible paths, or physically infeasible paths, and physically infeasible paths are removed from the high-risk output, thereby avoiding misjudging paths that are reachable on the network side but not valid in power physics as attack paths with physical impact.

[0054] The closed-loop calibration and system optimization module feeds back the inference and physical verification results to the model layer, enabling continuous model optimization. This module compares the consistency between the inferred attack path predictions and the actual physical verification results, and accordingly updates the calibrable causal edge attributes (such as causal strength, confidence, propagation delay, and defense residual coefficients), hyperedge weights, and rule confidence. Simultaneously, the updated results are reintegrated into the knowledge verification and inference process, forming a closed loop of inference-verification-calibration-re-inference. This allows the rule templates, cross-domain coupling hypergraphs, and cross-domain causal graphs to be progressively modified with new scenarios, events, and verification results, while hard constraints such as protocol preconditions, device binding, physical topology, protection logic, and hard defense conditions remain unchanged throughout the process.

[0055] This invention aims to solve the technical problems in cross-domain attack simulation of power cyber-physical systems, such as the difficulty in executing knowledge, the difficulty in expressing high-order coupling relationships, the lack of domain constraints on causal paths, and the difficulty in closing-loop calibration of simulation results.

[0056] This invention first converts trusted, verified multi-source knowledge into executable rule templates, then generates a weighted cross-domain coupled hypergraph from the rule templates and scene nodes. Cross-domain causal edges are generated and attack paths are expanded only with the support of verified hyperedges and corresponding cross-domain transition conditions. For paths involving physical influences, a mapping between control actions and physical objects is established, and its feasibility is verified through physical constraints or power simulation. Finally, the inference results and physical verification results are used to calibrate updatable causal edge attributes, hyperedge weights, and rule attributes, while maintaining hard constraints such as protocol premises, device binding, physical topology, and protection logic unchanged. This forms a continuous technical chain of mutual constraints among knowledge, topology, causal inference, physical verification, and feedback calibration.

[0057] In addition, the system includes a probabilistic perturbation algorithm unit and a simulated environmental interference module. The probabilistic perturbation algorithm unit is used to adjust the causal strength, confidence level, and propagation delay within the constraints of the power knowledge base; the simulated environmental interference module is used to simulate non-attack factors such as communication delays, equipment failures, data noise, load fluctuations, and human error. The above perturbations only apply to existing node, superedge, and causal edge attributes and do not change attack logic not supported by the rule template.

[0058] This invention sets up attack propagation edges, control action edges, physical influence edges, defense blocking edges, environmental disturbance edges, and service influence edges in a cross-domain causal graph, and configures corresponding generation conditions, constraint attributes, and calibration strategies for different types of causal edges. This allows for the differentiation of attack propagation, control action triggering, physical state changes, and service influence formation in the information and communication domains, and identifies which relationships are hard constraints and which attributes can be calibrated based on events and simulation results.

[0059] This invention categorizes cross-domain coupling hyperedges and causal edges into hard-constrained objects and calibrable objects. Hard-constrained objects express relationships in protocol premises, physical topology, device binding, defense blocking, and business processes that cannot be arbitrarily rewritten by statistical data. Calibrable objects express attributes that can be updated with the inference results, such as causal strength, confidence level, propagation delay, defense residual coefficient, and perturbation coefficient. This avoids pure data-driven methods mistaking accidental co-occurrences for executable attack logic, while preserving the model's ability to evolve with new evidence.

[0060] Unlike related technologies that use knowledge graphs, attack graphs, causal graphs, or power simulations respectively, this invention establishes a collaborative processing chain with sequential dependencies and feedback relationships: the power knowledge base generates rule templates through reliable verification; the rule templates drive the generation of cross-domain coupled hypergraphs; the cross-domain coupled hypergraph constrains the generation of cross-domain causal edges and path expansion; the path results output by the cross-domain causal graph and the physical verification results are used to back-calibrate the updatable hyperedge weights, rule confidence, and causal edge attributes. This reduces candidate paths lacking topological, rule, or physical support while preserving the necessary physical relationships within the power system.

[0061] Cyber-Physical Systems (CPS) are a type of system where the power CPS is a composite system formed by coupling the primary power system with communication, control, computing, and business systems.

[0062] According to an embodiment of the present invention, a method for cross-domain attack simulation of a power cyber-physical system is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0063] This embodiment provides a method for extrapolating cross-domain attacks on power cyber-physical systems, which can be used in the aforementioned electronic devices. Figure 2 This is a flowchart of a cross-domain attack deduction method for power cyber-physical systems according to an embodiment of the present invention, such as... Figure 2 As shown, the process includes the following steps: Step S201: Obtain the knowledge entries corresponding to the multi-source knowledge, verify the credibility of the knowledge entries, and convert the verified knowledge entries into executable rule templates.

[0064] Multi-source knowledge refers to heterogeneous knowledge derived from various channels, such as power industry standards, equipment manuals, security knowledge bases, historical event data, expert rules, and simulation results. Examples include power standard IEC 61850, equipment technical manuals, the ATT&CK for ICS attack knowledge base, power simulation results, expert rules of experience, and historical fault and event data.

[0065] Credibility verification of knowledge entries refers to checking whether the fields of the knowledge entries are complete, whether the source is reliable, and whether there are logical conflicts between the entries, so as to filter out knowledge with complete fields, reliable sources and logical consistency for generating rule templates.

[0066] Multi-source knowledge is converted into structured knowledge items, defined as: (1); Among them, the prerequisites This describes the attack conditions, operational conditions, or device status that must be met before a rule can be established; applicable objects. This describes the device, communication object, service object, or event object to which the knowledge applies; the result object. Used to describe the events, state changes, or physical consequences that may occur after a rule is triggered; constraints. Used to describe protocols, timing, defenses, topological or physical constraints; knowledge sources This is used to record that the knowledge comes from standards, equipment manuals, ATT&CK for ICS (Adversarial Tactics, Techniques, and Common Knowledge for Industrial Control Systems, a knowledge system of adversarial tactics and techniques compiled by MITRE for industrial control systems, which this invention uses as one of the sources of attack behavior knowledge), simulation results, expert rules, or event data; verifying the status. This is used to indicate that the knowledge is in a verified, pending review, conflicting, expired, or unverified state.

[0067] Executable rule templates are used to transform knowledge entries into executable hyperedge generation rules. Each rule template... Defined as: (2); in, For rule identification, As a source of knowledge, For runtime superedge type, For semantic tags, For a set of node selectors, For prior weights, For source confidence level, To verify the policy set, The minimum number of matching nodes required to generate candidate superedges.

[0068] rule template collection It is used to store rule templates derived from verified or available knowledge entries. The set of knowledge items to be supplemented is denoted as the rule template. It is used to store knowledge entries that are missing, fail validation, or have conflicts. This is a knowledge entry to be added.

[0069] Specifically, structured knowledge entries are first extracted from heterogeneous knowledge sources such as power standards, equipment manuals, ATT&CK for ICS, expert rules, historical event data, and simulation results. Each knowledge entry includes at least the following fields: preconditions, applicable objects, result objects, constraints, knowledge source, and verification status. Then, the source credibility is read according to the knowledge source, and the verification status is used to determine whether the entry meets the conversion conditions. For knowledge entries with complete fields, a verified status, and source credibility that meets the preset conditions, the preconditions are mapped to the preconditions of the rule template, the applicable objects are mapped to node selectors, the result objects are mapped to result node types, and the constraints are mapped to verification strategies, thereby generating rule templates that can be directly called for subsequent node matching and hyperedge generation. Knowledge entries that do not meet the conditions are included in the set of knowledge entries to be supplemented and do not participate in the trusted inference process.

[0070] The preset conditions are a set of rules configured by the system during initialization based on security policies and scenario requirements. The core logic is that only reliable and fully verified knowledge can enter the trusted deduction process, ensuring that the subsequently generated rule templates and attack paths have sufficient credibility.

[0071] Step S202: Generate candidate hyperedges according to the rule template and preset scenario nodes, perform preset verification on the candidate hyperedges, and write the verified candidate hyperedges as verified hyperedges into the cross-domain coupling hypergraph. The cross-domain coupling hypergraph is used to represent the many-to-many coupling relationship between multi-dimensional entities in the power information physical system.

[0072] Among them, candidate hyperedges refer to temporary hyperedges generated by matching rule templates with scene nodes during the construction of cross-domain coupled hypergraphs, which have not yet passed multi-dimensional verification.

[0073] Performing preset verification on candidate hyperedges refers to performing multi-dimensional compliance checks on candidate hyperedges generated by matching rule templates with scene nodes, including logic, timing, data, defense, physics, and cross-domain transfer conditions, in order to select candidate hyperedges that meet system constraints and write them into the cross-domain coupled hypergraph.

[0074] Cross-domain coupling hypergraphs are used to characterize many-to-many coupling relationships between multi-dimensional entities in a cyber-physical system. They are topological models, and the cross-domain coupling hypergraph model is defined as follows: (3); in, This represents a cross-domain coupled hypergraph. The components are represented sequentially as follows: node set, hyperedge set, hyperedge coupling weight, hyperedge semantic label, and domain identifier. The node set consists of atomic model instances and event state nodes; each hyperedge connects two or more nodes; the hyperedge coupling weight represents the support strength or credibility of higher-order relationships; the hyperedge semantic label includes structure, protocol, attack, causality, defense, and physical semantics; and the domain identifier is used to distinguish between the information domain, communication domain, control domain, physical domain, and business impact domain.

[0075] Construction of the cross-domain coupling hypergraph: First, scenario nodes and event state nodes are instantiated from the atomic model library based on the attack scenario template, including five types of nodes: information domain, communication domain, control domain, physical domain, and business impact domain. Then, the node selector in the rule template is matched with the scenario nodes to generate candidate hyperedges that connect two or more nodes. Next, the candidate hyperedges are verified in multiple dimensions, including logic, timing, data, defense, and physical aspects, and the cross-domain transfer conditions are additionally verified when connecting nodes from different domains. After the verification is passed, a hyperedge coupling weight, semantic label, and domain identifier are set for each hyperedge, and finally, a cross-domain coupling hypergraph is generated to represent the many-to-many coupling relationship between multi-dimensional entities in the power information physical system.

[0076] Specifically, a network attack scenario template is constructed based on preset attack targets, attack entry points, target objects, attack stages, and scope of action. Based on this, five types of scenario nodes—information domain, communication domain, control domain, physical domain, and business impact domain—are instantiated from the atomic model library. Then, the node selector in the rule template is matched with the scenario nodes to generate candidate hyperedges. Logical, temporal, data, defense, and physical verifications are performed on the candidate hyperedges in sequence. If they connect to nodes in different domains, cross-domain transfer condition verification is performed additionally. After all verifications are passed, the candidate hyperedge is written as a verified hyperedge into the cross-domain coupling hypergraph, making it the topological constraint basis for subsequent causal edge generation and path deduction.

[0077] Step S203: Generate candidate causal edges only between nodes that have been verified to support superedges and whose corresponding constraints are satisfied, and construct a cross-domain causal graph based on the candidate causal edges.

[0078] Among them, candidate causal edges are directed edges generated between nodes supported by the cross-domain coupled hypergraph during the construction phase of the cross-domain causal graph, representing potential causal transmission relationships between events or states. They must undergo causal edge admission verification before they can be written into the cross-domain causal graph as formal causal edges.

[0079] Generating candidate causal edges only between nodes supported by verified superedges and whose corresponding constraints are satisfied means that for any two nodes, a directed edge representing a potential causal relationship between events or states is allowed to be generated only if they belong to the same verified superedge or are reachable by multiple verified superedges, and satisfy temporal, protocol, and physical constraints.

[0080] Cross-domain causal graphs are used to characterize directional transmission relationships supported by domain knowledge, rule constraints, system topology, event sequences, and physical verification. They do not necessarily require the automatic discovery of true causal structures solely from observational data. Supported by power knowledge constraints and verified hyperedges, and integrating mechanistic attributes with statistical attributes formed from event or simulation data, a cross-domain causal graph is constructed, denoted as: (4); in, A set of nodes representing events or states, contained in node set V. It is a set of directed causal edges.

[0081] Each causal edge for: (5); in, The attributes are represented in the following order: cause node, effect node, causal edge type, causal relationship strength, confidence level, propagation delay, evidence count, and constraint attributes. The initial attributes of the causal edge are derived from the power knowledge base, corresponding verified superedges, attack scenarios, and event sequences. Subsequent updates only update calibrable attributes.

[0082] Causal edge type It includes at least six categories: attack propagation edges, used to represent the propagation relationship between attack actions such as reconnaissance, access, execution, and lateral movement; control action edges, used to represent the relationship between information events triggering control commands, protection actions, or switching actions; physical impact edges, used to represent the relationship between equipment actions leading to power flow changes, voltage overruns, or load losses; defense blocking edges, used to represent the weakening or blocking relationship between defense measures such as detection, isolation, verification, and interlocking on attack propagation; environmental disturbance edges, used to represent the impact of equipment failures, communication delays, data noise, load fluctuations, and human error on existing relationship attributes; and business impact edges, used to represent the impact of physical consequences or control anomalies on scheduling operations, maintenance operations, and operational risks.

[0083] constraint attributes This includes hard constraints and calibrable constraints. Hard constraint properties are generated only when the corresponding cross-domain coupling hyperedge, rule template premise, and verification strategy are all satisfied simultaneously. The calibration process must not modify them to violate protocols, topologies, defenses, or physical mechanisms. Calibrable properties allow updating the causal correlation strength based on event sequences, inference logs, and physical simulation results. Confidence level Propagation delay Alternatively, residual coefficients can be defended, but the updated edges still need to retain cross-domain coupled hypergraph support.

[0084] Specifically, nodes and verified hyperedges in the cross-domain coupled hypergraph are obtained. For any two nodes, candidate causal edges are only allowed to be generated if they belong to the same verified hyperedge or are reachable by multiple verified hyperedges. These candidate causal edges are then assigned attributes such as type, strength, confidence, and latency. Candidate causal edges must undergo admission verification, including hypergraph support, causal type consistency, direction consistency, and constraints. Once verified, they are used as edges in the cross-domain causal graph and together with the nodes, they constitute the cross-domain causal graph.

[0085] Step S204: The attack path is derived from the cross-domain causal graph to assess the security risks of the power cyber-physical system based on the attack path.

[0086] The attack path, derived from a cross-domain causal graph, is a directed propagation sequence consisting of candidate causal edges connecting the attack entry node to the attack target node. It characterizes the complete process of an attack event propagating from the information domain to the physical domain within the power cyber-physical system. In this embodiment, the attack path is a Top-k candidate path, representing the top k candidate paths output after sorting by path propagation support score. This score is used for sorting and is not equivalent to the actual attack success probability.

[0087] The attack path in this embodiment has the following characteristics: each step in the attack path is connected by candidate causal edges, and each candidate causal edge carries quantifiable attributes such as causal strength, confidence, propagation delay, and defense residual coefficient; the attack path is sorted and output through a scoring mechanism (the product of the propagation support scores of all causal edges in the path); the attack path derivation process is constrained by cross-domain transfer conditions, and paths that do not meet the conditions are stopped from expansion and the reason for stopping is recorded; after the attack path has been physically feasible, conditionally feasible, or physically infeasible, it is used to support the security risk assessment and defense strategy verification of the power information physical system.

[0088] Specifically, after determining the attack entry node and attack target node based on the attack scenario template, the system starts from the attack entry node and advances hop by hop along the candidate causal edges in the cross-domain causal graph, with the current active node as the starting point. At each step, the system comprehensively judges whether to trigger the endpoint node based on the type, causal strength, confidence, constraints, and defense status of the candidate causal edges, and checks whether the transition conditions are met when crossing domains. If triggered, the endpoint node is set as the new current active node and the causal edges traversed are recorded. This process is repeated until the attack target node is reached or the preset stopping condition is met. Finally, the attack path is generated based on the actual sequence of active nodes and candidate causal edges triggered during the simulation.

[0089] Step S205: Perform physical feasibility verification on the attack path, and perform closed-loop calibration based on the attack path deduction results and physical feasibility verification results.

[0090] Physical feasibility verification refers to verifying whether the physical impact of the attack path is consistent with the power system's operating mechanism by mapping the attack event to a physical action identifiable by the power system and using topology constraints, power flow constraints, protection logic constraints, equipment state constraints, and / or power simulation. This is done to determine whether the attack path is truly feasible at the power physical level.

[0091] Closed-loop calibration refers to comparing the results of attack path deduction with the results of physical feasibility verification. Based on this, adjustable parameters such as causal edge attributes, hyperedge weights, and rule confidence are corrected in reverse. The corrected results are then incorporated back into the knowledge verification and deduction process, so that the model is continuously optimized as deduction and verification are repeatedly executed, while keeping hard constraints such as protocol premises, device binding, physical topology, and protection logic unchanged.

[0092] Specifically, for attack paths involving physical impacts, attack events are mapped to physical actions identifiable by the power system, and the physical feasibility of these physical actions is verified using topology constraints, power flow constraints, protection logic constraints, equipment state constraints, and / or power simulation. The physical feasibility of the attack path is then marked based on the verification results.

[0093] Based on the consistency between the attack path deduction results and the physical feasibility verification results, the calibrable edges are updated, and the rule templates, cross-domain coupling hypergraphs, and cross-domain causal graphs are modified according to the update results, while the hard constraint relationships determined by protocol premises, device binding, physical topology, protection logic, and hard defense conditions remain unchanged.

[0094] The cross-domain attack inference method for power cyber-physical systems provided in this embodiment converts multi-source knowledge into executable rule templates, enabling knowledge to directly drive attack path generation. By constructing a cross-domain coupling hypergraph to represent many-to-many coupling relationships between multi-dimensional entities, it solves the problem of the difficulty in expressing high-order cross-domain coupling relationships. By generating causal edges under the constraints of the cross-domain coupling hypergraph to construct a cross-domain causal graph, the generation of causal paths is constrained by the cross-domain coupling topology, thereby improving the accuracy and reliability of attack path inference results. This solves the problems in the prior art where multi-source knowledge is difficult to directly drive path generation, high-order cross-domain coupling relationships are difficult to represent, and causal paths lack both topological and physical constraints.

[0095] This embodiment provides a method for extrapolating cross-domain attacks on power cyber-physical systems, which can be used in the aforementioned electronic devices. Figure 3 This is a flowchart of a cross-domain attack deduction method for power cyber-physical systems according to an embodiment of the present invention, such as... Figure 3 As shown, the process includes the following steps: Step S301: Obtain the knowledge entries corresponding to the multi-source knowledge, verify the credibility of the knowledge entries, and convert the verified knowledge entries into executable rule templates.

[0096] Specifically, this step includes an initialization step, which is used to initialize the power knowledge base, atomic model library, rule template set, and inference-related parameters, providing a unified data foundation and rule source for subsequent cross-domain coupled hypergraph construction (including network attack scenario planning) and cross-domain causal graph construction.

[0097] Knowledge items include: prerequisites, applicable objects, result objects, constraints, knowledge sources, and verification status; the above step S301 includes: Step S3011: Verify the integrity of the knowledge entries in terms of field completeness, source credibility, and logical conflict. The knowledge entries that pass the verification are used to generate executable rule templates.

[0098] Import knowledge entries from the Multi-Source Power CPS Knowledge Base After importation, the system performs field integrity verification, source credibility verification, and logical conflict verification on the knowledge entries. If a knowledge entry lacks preconditions, applicable objects, result objects, or knowledge sources, or if the verification status is unverified, the knowledge entry will not directly enter the trusted inference process, but will instead be added to the set of knowledge entries to be supplemented.

[0099] Step S3012: Read the source credibility based on the knowledge source; determine whether conversion is allowed based on the verification status and source credibility; if the prerequisites, applicable objects, result objects, and knowledge sources are complete, the verification status is verified, and the source credibility meets the preset conditions, then map the prerequisites to the prerequisites of the rule template, map the applicable objects to the node selectors of the rule template, map the result objects to the result node types of the rule template, and map the constraints to the verification strategies of the rule template to generate an executable rule template; otherwise, classify the knowledge items into the knowledge set to be supplemented.

[0100] Specifically, after knowledge initialization is complete, the system uses a rule template generation method based on knowledge credibility and verification status to convert knowledge items into executable rule templates. .

[0101] During this transformation process, the system executes a rule template generation algorithm driven by knowledge entry trust verification. This algorithm reads each knowledge entry. The attributes, and output a set of rule templates accordingly. and a collection of knowledge items to be supplemented The system generates a rule template only when the preconditions, applicable objects, result objects, and knowledge sources all exist, and the verification status is not unverified. Otherwise, the knowledge entry will not participate in the generation of trustworthy hyperedges, but will instead enter... .

[0102] It should be noted that this embodiment also determines the source credibility of the rule template. The source credibility of the rule template is read from the knowledge source credibility configuration and the knowledge item verification status. The system obtains the source credibility according to the pre-configured source category mapping, and determines whether to allow the generation of a credible rule template according to the verified, pending review, conflict, expired, or unverified status. When the source configuration or verification basis is missing, empirical values ​​are not used instead, and the corresponding knowledge items are written into the supplementary or pending review set.

[0103] If a knowledge entry has conflicts but has not yet passed manual review or batch verification, the knowledge entry can only enter the candidate knowledge set and will not generate a trusted rule template. Through this process, the knowledge in the power knowledge base is first converted into executable rules, and then participates in subsequent hyperedge generation and causal inference, avoiding the direct use of natural language descriptions to replace inference rules.

[0104] When generating executable rule templates, establish a mapping from knowledge fields to inference objects: To ensure that the rule template is executable, the system also establishes a mapping between the rule template fields and the inference object fields during the initialization phase. Mapping as a prerequisite for attack or operation. Mapped to node selector Matchable atomic model types, The resulting node is mapped to the candidate hyperedge connection. Mapped to verification strategy Fields that fail to be mapped are not replaced with natural language interpretations of the rules, but are instead added as missing fields to the set of knowledge entries to be supplemented.

[0105] More specifically, the algorithm for knowledge item validation and rule template generation is as follows: Input: Set of knowledge items K, source credibility configuration and verification status configuration .

[0106] Output: Collection of rule templates and a collection of knowledge items to be supplemented .

[0107] Step 1: Read the preconditions one by one Applicable objects Result object Constraints Sources of knowledge and verification status Fields; write when necessary fields are missing .

[0108] Step 2: Based on the source of knowledge Read source trustworthiness; write when source configuration is missing. It is not made up for by experience points.

[0109] Step 3: Check the verification status Conflicting, expired, or unverified entries must not generate trusted rule templates.

[0110] Step 4: Apply to the target audience Mapping to a node selector, setting the preconditions and constraints Mapping to prerequisites and validation strategies, the result object Mapped to the result node type.

[0111] Step 5: Generate a rule template when the mapping is complete and the verification passes. And write Otherwise, record the missing fields and the reason for the failure.

[0112] Step S302: Generate candidate hyperedges according to the rule template and preset scene nodes, perform preset verification on the candidate hyperedges, and write the verified candidate hyperedges as verified hyperedges into the cross-domain coupling hypergraph. The cross-domain coupling hypergraph is used to represent the many-to-many coupling relationship between multi-dimensional entities in the power information physical system.

[0113] In one alternative implementation, the initialization phase further includes: Initialize the atomic model library: Configure physical devices, information components, behavioral agents, and business objects in the power CPS as standardized atomic models. Each atomic model includes an input interface, an output interface, state transition rules, and a cross-domain interaction interface.

[0114] The atomic models for physical devices can correspond to circuit breakers, busbars, lines, transformers, loads, and protection devices; the atomic models for information components can correspond to hosts, communication gateways, switches, protocol services, and databases; the atomic models for behavioral agents can correspond to attackers, maintenance personnel, schedulers, or automated control programs; and the atomic models for business objects can correspond to remote control processes, operation ticket processes, alarm handling processes, or state estimation tasks. The atomic model library provides a unified source of nodes for subsequent node instantiation, composite model construction, cross-domain coupled hypergraph construction, and cross-domain causal graph construction.

[0115] In one alternative implementation, such as Figure 5 As shown, the cross-domain coupled hypergraph is constructed in the following manner: The nodes of the cross-domain coupling hypergraph are determined, including nodes from the information domain, communication domain, control domain, physical domain, and business influence domain. These nodes originate from scenario nodes and event state nodes obtained by instantiating atomic models. The atomic models include physical device atomic models, information component atomic models, behavioral subject atomic models, and business object atomic models. The hyperedges of the cross-domain coupling hypergraph are determined, with each hyperedge connecting two or more nodes to represent many-to-many coupling relationships between multi-dimensional entities in the power information physical system. Each hyperedge is assigned a coupling weight, semantic label, and domain identifier to generate the cross-domain coupling hypergraph.

[0116] Specifically, such as Figure 5 As shown, step S302 above includes: Step S3021: Construct a network attack scenario template based on the preset attack target, attack entry point, target object, attack stage, and scope of effect.

[0117] Specifically, this step is to construct a network attack scenario template. First, based on the attack target, attack entry point, target object, attack stage, scope of effect, defense conditions, operational constraints, and simulation time window set manually or input by the system, a network attack scenario template is constructed.

[0118] (6); in, For the intended identifier, As the target of attack, As an entry point for attacks, For the target of the attack, For the attack phase set, For the scope of application, Given the known defense conditions, For operational constraints, This is the time window for simulation.

[0119] For example, in a scenario where abnormal access to the remote maintenance entry point might affect the remote control operation of substation monitoring and control devices, the system constructs a network attack scenario template based on manually set attack targets, attack entry points, target objects, attack stages, scope of impact, defense conditions, operational constraints, and simulation time windows.

[0120] in, This is the intended identifier; The target of the attack is to assess whether abnormal remote operation and maintenance behavior may affect the substation remote control operation. The entry point for the attack is either the remote maintenance host or the remote access link; The target of the attack is the station control layer's measurement and control device and its associated circuit breakers; This is a set of attack phases, including abnormal access, lateral access, control service triggering, and physical impact verification. The scope of application includes the remote operation and maintenance area, the station control layer network, measurement and control devices, circuit breakers, and remote control business processes; Known defense conditions include access control, whitelists, firewalls, and anomaly alert rules; Operational constraints include equipment controllable status, remote control process verification requirements, and protection interlocking conditions; This template defines the simulation time window. It specifies that the simulation starts from the remote operation and maintenance entry point, goes through stages such as abnormal access, lateral access, and control service triggering, and analyzes whether these may affect the remote control services of the monitoring and control devices and circuit breakers.

[0121] The network attack scenario template is used to define where this simulation begins, which targets it targets, which stages it goes through, and what defenses and operational conditions it is subject to.

[0122] Verify the network attack scenario template: Perform integrity and consistency checks on the network attack scenario template. Integrity checks are used to determine whether the attack target, attack entry point, attack target object, scope of action, and operational constraints are complete; consistency checks are used to determine whether the attack stages satisfy the dependencies between them, whether there is a modelable cross-domain path between the attack entry point and the attack target object, and whether the scope of action includes the necessary devices, communication links, business objects, defense objects, and physical objects.

[0123] If the attack target, attack entry point, target object, or operational constraints are missing, the system will not generate a trusted candidate hyperedge and will mark the scenario template as pending completion. If the attack phase sequence is unreasonable or the necessary objects are missing from the scope of action, the system will also not enter the trusted simulation process, but will record the missing fields and reasons for inconsistencies for later supplementation or review.

[0124] Step S3022: Instantiate scene nodes from the atomic model library based on the network attack scenario template.

[0125] Specifically, scenario nodes include at least information domain nodes, communication domain nodes, control domain nodes, physical domain nodes, and business impact domain nodes; defense devices, defense strategies, and protection interlocking conditions are instantiated as defense constraint objects that act on nodes or cross-domain transfer relationships.

[0126] For example, remote operation and maintenance hosts, station control layer hosts, or attack source hosts can be instantiated as information domain nodes; communication gateways, switches, or protocol services can be instantiated as communication domain nodes; protection devices, measurement and control devices, or control command interfaces can be instantiated as control domain nodes; circuit breakers, buses, lines, or loads can be instantiated as physical domain nodes; scheduling processes, operation tickets, or business tasks can be instantiated as business impact domain nodes. Firewalls, whitelists, access control policies, or anomaly detection rules participate in candidate hyperedge verification as defense constraint objects.

[0127] Step S3023: Match the node selector in the rule template with the scene nodes to generate candidate hyperedges.

[0128] Specifically, after instantiating the nodes, the system will combine the scene nodes with the generated set of rule templates. Perform matching. The node selector in the rule template is used to select objects that meet the conditions from the nodes of the current scene.

[0129] When the preconditions, applicable objects, constraints, and minimum number of matching nodes are all met, the system generates candidate hyperedges. This candidate hyperedge represents a higher-order coupling relationship formed by multiple cross-domain objects under the current attack scenario, rather than a normal connection relationship between two nodes. Candidate hyperedges can correspond to attack propagation relationships, control action relationships, physical influence relationships, defensive constraint relationships, business dependency relationships, or environmental disturbance relationships.

[0130] Step S3024: Calculate the initial weights of the candidate superedges. The initial weights are jointly determined by the prior weights of the rule template and the credibility of the knowledge source.

[0131] Specifically, the initial weights of candidate hyperedges are calculated simultaneously when generating them. For those generated from rule templates... Generated candidate superedges Its initial weight is denoted as The initial weights are determined jointly by the prior weights of the rule template and the confidence level of the knowledge source. (7); in, The prior support weights configured in the rule template, The credibility of the knowledge sources cited in this rule is limited to a range of 0 to 1; Used to limit the output range. The initial weights only represent the degree of prior support that the rule and its source have for the candidate superedge, but do not include the severity of business impact or physical consequences.

[0132] The initial weights of candidate superedges are derived from the rule template and source credibility configuration. Knowledge from power standards, equipment manuals, or validated simulation mappings can have higher source credibility in the configuration; expert experience to be verified or event data that has not yet been fully verified has lower source credibility and is entered into the trusted model or the set to be verified according to its verification status. The potential business or physical impact of candidate relationships is calculated separately in the physical feasibility verification and is not used to increase the credibility weight of the relationship itself.

[0133] Step S3025: Perform preset verification on the candidate hyperedge; the preset verification includes: logical verification, timing verification, data verification, defense verification and physical verification. If the candidate hyperedge connects to nodes in different domains, cross-domain transfer condition verification is also performed.

[0134] Specifically, after generating candidate hyperedges, the system performs pre-defined verifications on them. These verifications include logical verification, temporal verification, data verification, defense verification, and physical verification.

[0135] Among them, logical verification is used to determine whether the causal sequence relationship is satisfied between the attack phases; temporal verification is used to determine whether the event sequence, duration and response delay are reasonable; data verification is used to determine whether the required fields, state quantities, quantity measurements and object identifiers are complete; defense verification is used to determine whether the existing defense strategy blocks or weakens the candidate hyperedge; physical verification is used to determine whether the consequences described by the candidate hyperedge are in line with the power system operation mechanism and physical constraints.

[0136] For candidate hyperedges connecting nodes in different domains, the system also checks the corresponding transfer conditions according to the cross-domain transfer type recorded in the rule template. Candidate transfers from the information domain to the communication domain check asset association, network reachability, and service matching conditions; candidate transfers from the communication domain to the control domain check the correspondence between communication status, control permissions, command semantics, and the target control object; candidate transfers from the control domain to the physical domain check control object binding, physical preconditions, and physical action mapping. The actual fields checked are based on the current rule template, system configuration, and verifiable data; transfer conditions are not presumed to be valid based on missing information.

[0137] (8); in, For candidate superedges in the current attack scenario The comprehensive verification results are as follows. When hard constraint verification fails... Candidate hyperedges do not enter trusted cross-domain coupled hypergraphs; when all applicable verification projects pass. When only partial soft evidence supports the claim, Take the value between 0 and 1 given by the corresponding verification configuration, and retain the verification source. For candidate hyperedges The final weights, after verification, are used to write into the cross-domain coupled hypergraph.

[0138] If any validation fails, the candidate hyperedge is not written into the trusted cross-domain coupled hypergraph. Instead, it is written into the set to be reviewed along with its initial weight, failure condition, stopping domain, and association rule template. This result is used in subsequent path deduction to restrict candidate paths from expanding to the next domain.

[0139] Step S3026: The verified candidate hyperedges are written as verified hyperedges into the cross-domain coupled hypergraph.

[0140] Specifically, for the verified candidate hyperedges, the system writes them into the cross-domain coupled hypergraph and sets the hyperedge weights to... At the same time, the rule source, scenario identifier, various verification results, and hyperedge type are retained. Candidate hyperedges that fail the hard constraint verification are not included in the trusted cross-domain coupled hypergraph, but only in the set to be reviewed.

[0141] After the above processing, step S302 outputs the verified cross-domain coupling hypergraph, the verified network attack scenario template, the candidate hyperedges that failed verification but need to be retained, and the reasons for their failure. Among them, the cross-domain coupling hypergraph is the main input provided to step S303.

[0142] The subsequent step S303 generates a cross-domain causal graph under the constraints of the cross-domain coupled hypergraph. Only combinations of nodes supported by the hyperedge are allowed to generate candidate causal edges, thereby preventing attack deduction from deviating from the real system connection relationships, business constraints and electrical physical constraints.

[0143] The algorithm for generating, verifying, and writing candidate hyperedges is as follows: Input: Collection of rule templates Node set and cyberattack scenario template .

[0144] Output: Trusted cross-domain coupled hypergraph and the set of candidate superedges to be reviewed .

[0145] Step 1: Use the node selector in the rule template to match scene nodes; do not generate candidate hyperedges if necessary nodes or prerequisites are missing.

[0146] Step 2: Calculate the initial weights of candidate hyperedges And record the rules and sources.

[0147] Step 3: Execute logic, timing, data, defense, physical, and cross-domain transfer condition verification.

[0148] Step 4: When hard constraint verification fails, add the candidate hyperedges and the reason for failure to the set of candidate hyperedges to be reviewed. .

[0149] Step 5: Calculate the hyperedge weights if the verification passes. Write candidate hyperedges into the cross-domain coupled hypergraph And save all verification records.

[0150] Step S303: Generate candidate causal edges only between nodes that have been verified to support superedges and whose corresponding constraints are satisfied, and construct a cross-domain causal graph based on the candidate causal edges.

[0151] Specifically, such as Figure 6 As shown, after completing step S302, which involves simulating a network attack scenario, the system obtains a verified cross-domain coupling hypergraph, a verified network attack scenario template, and candidate hyperedge verification results. Step S303 is used to construct a cross-domain causal graph under the constraints of the cross-domain coupling hypergraph, and to perform attack process deduction based on the cross-domain causal graph, obtaining attack paths, key nodes, propagation support scores, latency results, and defense impact results.

[0152] Step S303 above includes: Step S3031: Obtain the nodes, verified hyperedges, and associated rule templates in the cross-domain coupled hypergraph.

[0153] First, the system receives the cross-domain coupling hypergraph output in step S302. The attack propagation nodes in the cross-domain coupling hypergraph include information domain nodes, communication domain nodes, control domain nodes, physical domain nodes, and business impact domain nodes. Defense targets are attached as constraint information to relevant nodes, hyperedges, or causal edges. The system then selects nodes that are relevant to the current attack scenario and have been verified, forming a node set in the cross-domain causal graph.

[0154] A collection of nodes for an event or state Instead of being generated independently from step S302, the nodes are derived from nodes in the cross-domain coupled hypergraph that have already passed rule matching and multi-dimensional verification. This ensures that subsequent causal graphs will not generate nodes unrelated to the current attack scenario, nor will they include unverified objects in the trusted inference process.

[0155] Step S3032: Select any two nodes. When the first node and the second node are supported by the same verified superedge and the rule template supports propagation from the first node to the second node, generate the corresponding candidate causal edge; otherwise, do not generate a candidate causal edge. When multiple nodes form a reachable relationship through multiple verified superedges, generate candidate causal edges segment by segment according to the adjacent node pairs supported by each verified superedge, and do not generate candidate causal edges that cross intermediate nodes.

[0156] In this context, reachability refers to a connection between two nodes in a cross-domain coupled hypergraph that do not directly belong to the same verified hyperedge but can be connected through one or more verified hyperedges and their shared intermediate nodes.

[0157] Based on the hyperedge in the cross-domain coupled hypergraph Generate candidate causal edges. For any two nodes... and The system allows the generation of candidate causal edges only when both belong to the same verified hyperedge, or when there is a reachable coupling relationship between them formed by multiple hyperedges.

[0158] That is, the cross-domain causal graph is not generated by connecting all nodes pairwise, but rather under the constraints of the cross-domain coupled hypergraph. This constraint... Represents a node and Whether they are supported by a cross-domain coupling hypergraph. If If the system does not generate a reliable causal edge between the two, then the system will not generate a reliable causal edge between them; if Then, the candidate causal edge generation process begins. In this way, attack chain experience, log statistics, or manually set relationships cannot directly form causal edges. They must first be supported by trusted superedges, thereby preventing the causal graph from deviating from the actual system topology, business processes, and electrical physical constraints.

[0159] Step S3033: Determine the type of candidate causal edge, and classify the candidate causal edge into hard constraint edge or calibrable edge according to the constraint source and allowable update range of the candidate causal edge; the type includes at least one of attack propagation edge, control action edge, physical influence edge, defense blocking edge, environmental disturbance edge, and business influence edge; hard constraint edge comes from power physics mechanism, mandatory safety constraint, protection action logic, equipment technical manual or standard specification. During the closed-loop calibration process, hard constraint edge shall not be deleted, and the propagation direction and hard constraint condition of hard constraint edge shall not be changed; the causal correlation strength, confidence and propagation delay of calibrable edge may be updated within a preset range according to the attack path inference results and physical feasibility verification results. The updated calibrable edge still needs to be supported by the verified superedge and shall not violate the hard constraint condition.

[0160] Specifically, for node combinations supported by cross-domain coupled hypergraphs, the system determines candidate causal edge types based on rule templates, attack phases, node types, and hyperedge types. These candidate causal edges include at least attack propagation edges, control action edges, physical influence edges, defense blocking edges, environmental disturbance edges, and business influence edges.

[0161] Among them, the attack propagation edge describes the process of an attack spreading from one information object or communication object to another; the control action edge describes the impact of control commands, protection actions, or operating procedures on equipment status; the physical impact edge describes the impact of control actions, equipment status changes, or fault events on power grid physical quantities and physical equipment; the defense blocking edge describes the blocking or weakening of attack propagation by firewalls, whitelists, access control, anomaly detection, or protection interlocking; the environmental disturbance edge describes the impact of equipment failure, communication delay, data noise, or operational disturbances on the simulation process; and the service impact edge describes the impact of attack events on scheduling processes, monitoring services, operation ticket processes, or operational decisions.

[0162] By introducing multiple types of causal edges, the system can distinguish how attacks propagate, how commands take effect, how physical consequences occur, how defenses block attacks, and how business is affected, thus avoiding simplifying all relationships into a single attack path.

[0163] Generate causal edge attributes: Generate quantifiable attributes for each candidate causal edge. The structure of the candidate causal edge is represented by formula (5). Causal correlation strength Used to indicate events or states Transmitted to Degree of influence; causal confidence level Used to indicate the degree of confidence in the causal relationship; propagation delay Used to represent the time required for causal transmission; constraints This is used to record the protocol constraints, timing constraints, defense constraints, business constraints, or physical constraints that must be satisfied for the causal edge to be valid.

[0164] The causal edge attribute is jointly determined by the mechanistic causal attribute and the statistical causal attribute. The mechanistic causal attribute is derived from the power knowledge base, equipment manuals, standards and specifications, attack mechanism rules, protection logic and physical constraints; the statistical causal attribute is derived from historical event sequences, simulation events, log data or experimental data.

[0165] For candidate edges supported by both mechanistic evidence and event data, the system does not use the same weight to uniformly average the strength, confidence, and delay. Mechanism rules are used to determine whether causal edges are allowed to exist, their propagation direction, and necessary constraints; event sequences or simulation data are only estimated for calibrable attributes that have been verified by the hypergraph and rules. Causal strength, confidence, and propagation delay are calculated using their respective estimation methods and update coefficients. When valid samples or timestamp evidence are lacking, the mechanistic values ​​are retained, and the corresponding attributes are marked as requiring calibration.

[0166] Through the above-mentioned hierarchical processing, statistical data cannot change the existence and direction of hard constraint relationships. It can only calibrate the attributes that are allowed to be updated when there is sufficient evidence, thereby avoiding the direct writing of accidental co-occurrence into the credible causal graph.

[0167] Furthermore, causal edges are divided into hard-constrained edges and calibrable edges. Hard-constrained edges originate from electrical physics mechanisms, mandatory safety constraints, protection action logic, equipment technical manuals, or standards and specifications, and their existence should not be deleted due to insufficient short-term data. For example, the impact of circuit breaker opening and closing states on line power flow topology, and the impact of protection actions on circuit breaker tripping, can be treated as hard-constrained edges.

[0168] Calibrable edges are primarily derived from historical events, simulation data, statistical correlations, or expert experience. Their causal strength, confidence level, and latency can be dynamically updated based on subsequent extrapolation and physical simulation results. Through this hierarchical mechanism, the system can both preserve the inviolable physical and business logic of power CPS and adapt to new attack samples, changes in equipment status, and changes in the operating environment.

[0169] Step S3034: Perform hypergraph support verification, causal type consistency verification, direction consistency verification, and constraint condition verification on the candidate causal edges. Among them, the candidate causal edges connecting nodes in different domains must also meet the corresponding cross-domain transfer conditions.

[0170] Specifically, candidate causal edges undergo hypergraph support verification, causal type verification, direction consistency verification, constraint verification, and physical feasibility verification. Only when a candidate causal edge receives verified hyperedge support, its edge type matches the node type, its direction conforms to the attack phase or business process sequence, and its constraints satisfy the current state, is it written into the cross-domain causal graph. Each trusted causal edge retains its corresponding hyperedge source, rule source, edge type, strength, confidence level, latency, and constraints.

[0171] Cross-domain transfer conditions refer to a series of prerequisites that must be met simultaneously when an attack path moves from one domain (information domain, communication domain, control domain, physical domain, or business impact domain) to another domain in a cross-domain coupling hypergraph, including asset association, network reachability, control permissions, device binding, physical preconditions, and the correspondence between command semantics and target objects.

[0172] Step S3035: The verified candidate causal edges are used as cross-domain causal edges, and a cross-domain causal graph is constructed based on the cross-domain causal edges and the nodes in the cross-domain coupling hypergraph.

[0173] Step S304: The attack path is derived from the cross-domain causal graph to assess the security risks of the power cyber-physical system based on the attack path.

[0174] Specifically, such as Figure 6 As shown, step S304 above includes: Step S3041: Determine the attack entry node and attack target node according to the attack scenario template, set the attack entry node as the initial activation node, and set the initial activation node as the current activation node.

[0175] Specifically, after the cross-domain causal graph is constructed, the system initializes the attack process simulation state according to the network attack scenario template. The system sets the node corresponding to the attack entry point as the initial active node, sets the attack target object as the target node or a set of target nodes, and determines the simulation order according to the attack stage set.

[0176] The simulation status includes at least the currently active node, triggered events, candidate causal edges, node status, cumulative path score, cumulative latency, and defense effectiveness status. If the attack entry point cannot be mapped to a node in the cross-domain causal graph, or if the target object is not in the current causal node set, the system terminates the trusted simulation and returns the reason for the scenario template needing to be supplemented or the node instantiation failure.

[0177] Step S3042: Starting from the currently active node, retrieve cross-domain causal edges originating from the currently active node; when the retrieved cross-domain causal edge connects nodes in different domains, obtain the cross-domain transfer conditions recorded by the verified superedge associated with the cross-domain causal edge, and check the cross-domain transfer conditions before triggering the endpoint node; when all necessary conditions are met, continue to execute the trigger judgment; when any necessary condition is not met, stop the current attack path from extending along the cross-domain causal edge, and record the unmet conditions and their associated rule templates.

[0178] Specifically, all necessary conditions refer to all the specific conditions that must be met simultaneously in the cross-domain transfer conditions, including asset association, network reachability, service matching, communication status, control permissions, the correspondence between command semantics and target objects, control object binding, and physical preconditions, etc. If any one of them is not met, cross-domain extension will be prevented.

[0179] Starting from the currently active node, retrieve cross-domain causal edges originating from that node. When a retrieved cross-domain causal edge connects nodes in different domains, obtain the cross-domain transfer conditions recorded by the verified superedge associated with that causal edge. Before triggering the endpoint node, check all necessary conditions such as asset association, network reachability, control permissions, device binding, and physical prerequisites item by item. If all conditions are met, continue to execute the trigger judgment. If any condition is not met, stop the current path from extending along that edge and record the unmet conditions and their associated rule templates to clarify the reason for the path termination and the source of the constraints.

[0180] Step S3043: Based on the type of cross-domain causal edge, the strength of causal association, confidence level, constraint conditions, and defense status, determine whether the endpoint node pointed to by the cross-domain causal edge is triggered; if triggered, record the cross-domain causal edge and the corresponding endpoint node, set the endpoint node as the currently active node, and continue the deduction until the attack target node is reached or the preset stopping condition is met.

[0181] Specifically, starting from the initial activated node, the attack process is progressively extended along the causal edges in the cross-domain causal graph. For the current causal node... The system search is based on Candidate causal edges starting from the given edge are selected, and the decision on whether to trigger the next result node is made based on the edge type, causal strength, confidence level, delay, constraints, and defense status. .

[0182] If the causal edge is an attack propagation edge, the system determines whether the attack conditions, permission conditions, and communication reachability conditions are met; if the causal edge is a control action edge, the system determines whether the control command, business process, and device status are met; if the causal edge is a physical influence edge, the system determines whether the corresponding physical action or state change conforms to the operational constraints; if the causal edge is a defense blocking edge, the system determines whether the attack propagation is blocked or weakened according to the defense strategy.

[0183] When a candidate causal edge connects nodes in different domains, the system reads and checks the cross-domain transition conditions of the associated superedge before triggering the next node. Only when all necessary conditions are met can the candidate path enter the next domain; if any necessary condition is missing or not met, the system stops the current path from expanding and records the stopped domain, the unmet condition, the corresponding superedge, and the source of the rule.

[0184] The system calculates a propagation support score for each triggerable causal edge: (9); in, Score the transmission support of the causal edge under the current assumption. For the strength of causal relationship, For the confidence level of evidence, To protect against residual coefficients, all four are limited to the range of 0 to 1. This is only used for ranking candidate paths and does not represent the statistical probability of a successful attack in a real system. For edges where the defense is partially effective, let... For defensive effectiveness, To determine the extent to which the defense covers the current edge, the defense residual coefficient is: (10); in, Read from verified defense effectiveness configurations or defense verification results. The determination is based on the attack phase, target, and time window covered by the defense measure. If no corresponding defense measure is available... When partial defense is effective, the calculation is performed according to the above formula; when the hard blocking condition is met. The related attack edges cease to expand. If there is a lack of evidence regarding the effectiveness or coverage of the defense, the defense is not presumed to be fully effective; instead, it is recorded as a defense pending review.

[0185] Step S3044: Generate an attack path based on the activated nodes and cross-domain causal edges recorded during the deduction process.

[0186] During the simulation, the attack path from the attack entry point to the target node is recorded and denoted as: (11); for The path score is calculated based on the propagation support score of each causal edge in the path: (12); The total latency of the entire attack path is also calculated based on the propagation latency of each causal edge in the path: (13); in, As the attack path, All are nodes. This indicates the total number of nodes in the attack path. Rate the path For nodes To the node The transmission of causal edges between them supports the score. The total latency of the entire attack path. For nodes To the node The propagation delay of the causal edges between them.

[0187] Path scores represent the degree of continuous propagation support a candidate path receives under current knowledge, hypergraph, and defense conditions, but do not represent the probability of a real attack occurring. The product form causes any low-support step to reduce the overall path score, with natural decay for longer paths. The system first sorts paths by score; when scores are the same or the difference is less than a preset threshold, a secondary sort is performed based on total path delay, the output target impact level, and defense residual coefficients. The system outputs Top-k candidate paths along with their rule sources, supporting hyperedges, stopping or blocking points, and residual risk descriptions.

[0188] The system starts from the attack source node and reaches the maximum depth. and maximum cumulative delay Search for the target node within the constraints. and Read from the current attack scenario or path search configuration; if the configuration is missing, use the default value explicitly recorded by the system. The system... Before output There are several candidate paths. If... If the simulation time window or defense response window is exceeded, the path is stopped, downgraded, or marked as an out-of-window path, and the corresponding reason is recorded.

[0189] When encountering a defensive blocking edge during the simulation, the system does not simply delete the attack path. Instead, it assesses the impact based on the defense type and its effectiveness. If the defense is fully effective, the relevant attack propagation edge will no longer participate in the current path expansion. If the defense is partially effective, the system reduces the propagation support score of the causal edge based on the defense residual coefficient and continues to determine whether there are any residual attack paths.

[0190] Through this process, the system can compare the number of attack paths, path scores, and changes in the impact of key nodes and targets before and after defense, thereby supporting the verification of defense strategies. For example, when an access control policy blocks the attack propagation edge from a remote operations and maintenance host to a site control layer host, the system can determine whether the policy has cut off the main attack path, or whether the attack can still form an alternative path through other communication objects, business processes, or weakly defended nodes.

[0191] After the attack process simulation is completed, the system outputs the cross-domain attack simulation results. The simulation results include at least the cross-domain causal graph, the attack path set, the top-k high-risk paths, the key node set, the triggered causal edge set, the blocked causal edge set, the path score, the path latency, the defense impact results, and the reasons for the candidate paths that were not successfully triggered.

[0192] Among them, the cross-domain causal graph is used to illustrate the causal structure of attack propagation; the attack path set is used to illustrate the possible propagation process of the attack from the entry point to the target object; the critical node set is used to locate the nodes that contribute significantly to the path connectivity or target impact during the attack process; the blocked causal edge set is used to illustrate the location of the actual impact of the defense strategy; and the reasons for the unsuccessfully triggered candidate paths are used to illustrate at which conditions, constraints, or physical mechanisms the attack process is terminated.

[0193] The above steps output cross-domain causality graphs and attack process deduction results. The cross-domain causality graphs serve as the basis for subsequent model calibration and path interpretation; the attack process deduction results serve as inputs for subsequent physical simulation verification, environmental perturbation analysis, defense strategy evaluation, and closed-loop calibration.

[0194] The system further transforms the generated weighted cross-domain coupled hypergraph into an attack process model with causal direction and inferable properties, so that cross-domain attack inference can not only explain the objects with coupled relationships, but also explain: how the attack propagates along the causal relationship, where it is blocked by defense, and when it has business or physical impact.

[0195] It should be noted that this embodiment also provides a process for handling probabilistic perturbations and inferred environmental disturbances: After completing the basic attack simulation, the system can introduce probabilistic perturbations and environmental disturbances within the existing trusted hypergraph and causal graph to analyze the sensitivity of paths to uncertain operating conditions. The perturbations only adjust the strength, confidence, latency, node state, or defense residual coefficients of calibrable causal edges, without generating new nodes, hyperedges, causal edges, or attack logic.

[0196] The system determines the attributes and scope that can be adjusted based on the causal edge type, the amount of evidence, the rule confidence level, and the hard constraint identifier. Relationships with sufficient evidence or high source credibility are subject to a smaller adjustment scope; hard constraint relationships must not have their existence or direction changed. Environmental interference events must at least record the event type, affected object, time window, scope, and severity, and only affect credible relationships where the object, time, and type match.

[0197] The system can perform multiple rounds of perturbation simulations on the same cross-domain causal graph, recalculating edge propagation support scores, path scores, and path delays respectively, and outputting path score distributions, delay distributions, conditionally triggered paths, sensitive causal edges, and sensitive environmental factors. Paths that only hold true under specific perturbation conditions are marked as conditionally triggered paths and are not directly promoted to unconditionally trusted paths.

[0198] Step S305: Perform physical feasibility verification on the attack path, and perform closed-loop calibration based on the attack path deduction results and physical feasibility verification results.

[0199] Specifically, this step involves calculating and verifying the impact of network attacks. After constructing the cross-domain causal graph, simulating the attack process, and implementing optional probabilistic perturbations and simulation environment interference, the system obtains the attack path set, path scores, path delays, triggered causal edges, blocked causal edges, and target trigger support. The calculation and verification of the impact of network attacks further maps the attack simulation results to the impact on power operations and the physical impact on the power grid, calculates the degree of attack impact, and determines whether the attack path is feasible in a power system sense through power simulation or physical constraint verification.

[0200] Step S305 above includes: Step S3051: When there is a control action edge or physical influence edge in the attack path, the attack event is mapped to a physical action that can be identified by the power system, and the physical action is verified for power physical feasibility. The physical feasibility of the attack path is marked according to the verification result. The power physical feasibility verification is the verification of physical actions using topology constraints, power flow constraints, protection logic constraints, equipment state constraints and / or power simulation.

[0201] In an optional implementation, step S3051 includes: Step a1, Receive Attack Simulation Results: First, receive the attack simulation results output in step S304. The attack simulation results include cross-domain causal graph, attack path set, Top-k high-risk attack paths, key node set, path score, path latency, target trigger support, defense blocking results, and perturbation simulation results.

[0202] For each attack path The system reads the starting node, ending node, intermediate nodes, causal edge type, control actions, physically affecting nodes, business affecting nodes, and constraints of the attack path. If the attack path only stays in the information domain or communication domain and does not trigger control actions, business anomalies, or physical state changes, the system can mark it as a network-side affecting path. If the attack path further triggers changes in the state of control devices, business processes, or physical equipment, the system enters the attack impact calculation and physical verification process.

[0203] Step a2, Identify the objects affected by business operations and physical impacts: Identify the affected objects based on the node types and causal edge types in the attack path. Objects affected by business operations may include monitoring services, scheduling processes, status estimation, alarm judgment, protection interlocking logic, or operational decision-making processes; objects affected by physical impacts may include circuit breakers, lines, transformers, loads, generators, protection devices, or measurement and control devices.

[0204] The system focuses on identifying the control action edge, business impact edge, and physical impact edge in the attack path. If events such as abnormal remote control commands, malfunctioning protection, measurement offset, equipment failure to operate, communication interruption, or data tampering exist in the attack path, the system determines the business objects and physical objects that may be affected based on the event type.

[0205] Step a3: Establish a mapping from attack events to physical actions: For attack paths that can trigger physical effects, the system maps attack events to physical actions or changes in operating status that can be identified by the power system. This mapping does not directly modify the power grid simulation model; instead, it first generates descriptions of physical actions for subsequent physical verification.

[0206] For example, abnormal remote control commands can be mapped to circuit breaker opening or closing actions; malfunctioning protection can be mapped to protection tripping events; measurement data offsets can be mapped to abnormal state estimation inputs or control decision deviations; communication interruptions can be mapped to monitoring unreachability, control command delays, or missing protection information; and equipment failures can be mapped to equipment out of operation or abnormal parameters.

[0207] A physical action description must include at least the action object, action type, action time, duration, action amplitude, preconditions, and constraints. The system generates a physical action description only when the control action edge or physical influence edge in the path has been triggered, there is a valid mapping between the control action and the target physical object supported by rule templates or model configurations, and the action object belongs to the current physical model or physical constraint set; otherwise, the path is reserved as a candidate path on the network or control side, and its physical influence is not directly presumed.

[0208] Step a4, Calculate the impact of the attack on services: The system calculates the degree of service impact based on the affected objects, service importance, path score, target trigger support, and defense blocking results. Service impact can be calculated from dimensions such as availability, integrity, real-time performance, and observability. A rule-based hierarchical approach can also be used to calculate service impact. For example, paths affecting scheduling decisions, protection actions, or critical remote control links are classified as high service impact; paths affecting only log recording, ordinary alarms, or low-priority communication are classified as low service impact.

[0209] Step a5, Perform power physical constraint verification: The system performs power physical constraint verification based on the physical action description. Physical constraint verification is used to determine whether the physical actions caused by the attack path conform to the operating mechanism of the power system, rather than judging whether the path is valid only from the perspective of network connection or attack chain.

[0210] Physical constraints can include topology constraints, power flow constraints, protection logic constraints, device state constraints, operational boundary constraints, and time response constraints. Topology constraints are used to determine whether the attacked device is actually located in the current operating network; power flow constraints are used to determine whether the device's action will cause line overload, voltage exceeding limits, or power imbalance; protection logic constraints are used to determine whether the protection action meets the set values, action conditions, and blocking conditions; device state constraints are used to determine whether the device is in a state capable of performing opening and closing, enabling and disabling, or regulating actions; operational boundary constraints are used to determine whether the system operation exceeds the safety boundary after the attack; and time response constraints are used to determine whether the time relationship between control actions, protection actions, and physical responses is reasonable.

[0211] Step a6: Verify by connecting to the power simulation model: In a preferred embodiment, the system inputs physical action descriptions into a power simulation model for verification. The power simulation model can be a power flow calculation model, a transient simulation model, a protection action simulation model, or a power CPS co-simulation model.

[0212] The system converts the actions and their types within the attack path into input events recognizable by the simulation model. For example, it converts abnormal circuit breaker tripping into line or load disconnection events; protection maloperation into protection tripping events; measurement offsets into state estimation input deviations; and communication delays into control command delays or protection information delays. After the simulation model runs, the system reads outputs such as voltage, current, frequency, power flow, load loss, equipment status, protection action results, and over-limit information.

[0213] Step a7, calculate the degree of physical impact: The system calculates the degree of physical impact based on the output of physical verification or simulation. Physical impacts may include line overload, voltage exceeding limits, frequency deviation, load loss, equipment outage, protection malfunction range, and changes in system stability.

[0214] If the attack path cannot produce physical actions consistent with the power operation mechanism, or if the simulation results show that the path will not lead to observable physical effects, the system will mark the path as having no physical effects or low physical effects.

[0215] Step a8, calculate the overall attack impact: The system calculates the overall attack impact by combining path scoring, target trigger support, business impact scoring, and physical impact scoring. The overall impact represents the total risk level of the attack path under the current scenario, rather than just network reachability. Furthermore, a tiered rule can be used to output the overall impact level, such as three levels (high, medium, and low) or five levels (one to five).

[0216] Step a9, mark physically feasible and infeasible paths: Based on the physical constraint verification and simulation results, the system divides the attack path into physically feasible paths, conditionally feasible paths, and physically infeasible paths.

[0217] A physically feasible path is one that is reachable in a cross-domain causal graph and possesses effective control actions, controlled object mappings, and physical preconditions, with its physical constraint checks or simulation results meeting preset validity criteria. A conditionally feasible path is one that only meets the above conditions under specific operating modes, defense states, environmental conditions, or equipment states. A physically infeasible path is one that, while reachable in the information or communication domain, lacks effective control actions, controlled object mappings, or physical preconditions, or whose physical verification results do not meet validity criteria. Physically infeasible paths must not be output as trusted high-risk paths with physical impact, but their stopping points, unmet conditions, and sources of associated rules are retained for review and calibration.

[0218] This classification helps the system avoid misclassifying paths that may occur on the network but are not electrically physical as high-risk paths.

[0219] Step a10, output the attack impact and physical verification results: The output includes the business impact results of the attack path, the physical impact results, the comprehensive impact score, the physical verification conclusion, the simulation output summary, the set of physically feasible paths, the set of conditionally feasible paths, and the set of physically infeasible paths.

[0220] Among them, the set of physically feasible paths can be used as high-confidence attack paths to enter subsequent closed-loop calibration; the set of conditionally feasible paths is used to indicate the operating mode, defense conditions or environmental disturbances that need attention; and the set of physically infeasible paths is used to reverse correct rule templates, causal edge attributes or hyperedge weights.

[0221] Through the above steps, the system further applies the attack simulation results obtained in step S304 to the levels of power business impact and power grid physical impact, enabling the present invention to not only determine whether the attack path is reachable on the network or causal graph, but also to determine whether the attack path is likely to cause consequences to the operation of the power system, thereby improving the credibility and engineering applicability of the cross-domain attack simulation results.

[0222] Step S3052: Based on the consistency between the attack path deduction results and the physical feasibility verification results, update the calibrable attributes of the causal edge, keep the hard constraints determined by the protocol premise, device binding, physical topology and protection logic unchanged, and modify the rule template, cross-domain coupling hypergraph and cross-domain causal graph according to the update results to perform closed-loop calibration.

[0223] After completing the calculation and physical verification of the network attack impact, the system feeds back the path results, verification conclusions, and reasons for failure to the knowledge base, rule templates, cross-domain coupling hypergraph, and cross-domain causal graph. Closed-loop calibration only adjusts calibrable attributes such as causal strength, confidence, propagation delay, defense residual coefficient, hyperedge weight, and rule confidence, without changing the hard constraints determined by protocol premises, device binding, physical topology, protection logic, and hard defense conditions.

[0224] In an optional implementation, step S3052 includes: Step b1, receive physical verification and deduction feedback: The system first receives the simulation feedback information output in step S304. The simulation feedback information includes at least the attack path set, Top-k high-risk paths, triggered causal edges, blocked causal edges, path score, path latency, business impact score, physical impact score, comprehensive attack impact, physical verification conclusion, simulation output summary, and reasons for failure to pass verification.

[0225] For each attack path, the system records the rule templates, cross-domain coupling hyperedges, causal edges, node states, defense conditions, environmental interference conditions, and physical verification results involved. This record is used to determine whether the model prediction results are consistent with the actual deduction or simulation results.

[0226] Step b2, calculate the causal matching degree: The system calculates the causal matching degree based on the consistency between the cross-domain causal graph prediction results, attack inference results, and physical verification results. The causal matching degree is used to measure the explanatory power of the current causal graph for attack propagation and physical consequences.

[0227] The degree of causal fit can be expressed as: (14); in, , and These are the normalized errors for event triggering, physical impact, and propagation delay, respectively. , and For the corresponding weights, all are not less than 0 and satisfy the following conditions: + + =1. Weights are read from the calibration configuration and recorded along with the calibration target; if no weights are configured, an explicitly recorded equal weight setting is used. It is only used to locate model objects that need to be reviewed or calibrated, and is not used to override hard constraint verification conclusions.

[0228] If the causal matching degree is higher than the preset threshold, it means that the current cross-domain causal graph and cross-domain coupling hypergraph can explain the attack process well; if the causal matching degree is lower than the preset threshold, the system enters the calibration process.

[0229] Step b3, Locate the calibration object: When the causal matching degree is below the threshold, or when there are situations such as the physical feasible path being underestimated, the physical infeasible path being overestimated, or the defense blocking effect being predicted incorrectly, the system locates the object that needs to be calibrated.

[0230] The calibration targets include knowledge entries, rule templates, cross-domain coupling hyperedges, causal edges, defense blocking edges, perturbation parameters, and environmental interference parameters. The system prioritizes checking causal edges and hyperedges directly associated with the deviation path, and then checks the corresponding rule templates and knowledge sources. For paths that fail physical verification, the system prioritizes checking physical influence edges, physical constraint hyperedges, and related rule templates; for paths with incorrect defense effect predictions, the system prioritizes checking defense blocking edges, defense strategy mapping relationships, and defense residual coefficients; for paths with large time delay deviations, the system prioritizes checking conducted time delay parameters and environmental interference parameters.

[0231] By using the above positioning sequence, the system avoids making indiscriminate adjustments to the entire model, but instead calibrates only the local model objects that are relevant to the deviation results.

[0232] Step b4, update the causal edge properties: The system updates the attributes of calibrable causal edges based on inference feedback and physical verification results. Updatable attributes include causal strength, causal confidence, propagation delay, amount of evidence, and constraints.

[0233] For causal edges that are triggered multiple times and pass physical verification, the system increases their causal confidence or causal strength; for causal edges that are frequently triggered by deduction but fail physical verification, the system decreases their causal strength or adds constraints; for attack propagation edges that are not triggered due to defense blocking, the system adjusts the defense residual coefficient according to the effectiveness of the defense; for causal edges with significant delay deviations, the system updates the propagation delay according to the actual deduction delay or simulation delay.

[0234] Causality strength can be updated incrementally: (15); in, For the updated causal strength, The causal strength before the update. Based on the strength of observational support obtained from this round of simulations or physical verification, The update coefficient is between 0 and 1. This update is not performed if observations are missing, validation fails, or it conflicts with hard constraints. Limiting function. Used to restrict update results to Within the range.

[0235] For hard-constrained edges, the system does not delete their existence, but only allows the time delay, triggering conditions or explanatory information to be corrected within the constraint range; for calibrable edges, the system can adjust their strength, confidence and constraint conditions based on long-term feedback.

[0236] Step b5, update the weights of cross-domain coupled hyperedges: The system updates the weights of cross-domain coupling hyperedges in reverse based on the causal edge calibration results and physical verification results. If multiple causal edges supported by a certain hyperedge are repeatedly triggered in the inference and pass physical verification, the weight of the hyperedge is increased; if the causal edges supported by a certain hyperedge are not triggered multiple times, are blocked by defense, or fail physical verification, the weight of the hyperedge is decreased or it is marked as pending review.

[0237] (16); in, and These are the superedge weights before and after the update, respectively. This is the observational verification result of the hyperedge in this round of simulation and physical verification. The update coefficient is between 0 and 1. The weight must be calculated from the verification record associated with the hyperedge; the hyperedge weight is not updated if an associated verification record is missing.

[0238] Through this process, a two-way linkage is formed between the cross-domain coupled hypergraph and the cross-domain causal graph: the hypergraph constrains the generation of the causal graph, and the results of causal graph deduction and physical verification are used to back-calibrate the hypergraph weights.

[0239] Step b6, update knowledge entries and rule templates: When model bias arises from missing knowledge entries, incomplete rule templates, knowledge conflicts, or outdated knowledge, the system updates the power knowledge base and rule template set.

[0240] For knowledge entries that are missing, of unknown origin, have insufficient verification status, or conflict with simulation results, the system marks them as pending supplementation, pending review, or in conflict status. For rule templates that have been confirmed to be valid through multiple rounds of deduction and physical verification, the system increases their source confidence or verification status. For rule templates that generate erroneous superedges or erroneous causal edges, the system modifies their preconditions, applicable objects, constraints, node selectors, or verification strategies.

[0241] The updated knowledge entries will not be immediately and unconditionally incorporated into the trusted inference model. Instead, they need to be re-processed through the knowledge verification and rule template generation process in step S301 to ensure that the knowledge base, rule templates, hypergraph, and causal graph remain consistent.

[0242] Step b7, optimize defense strategy and simulation parameters: The system optimizes defense strategies and simulation parameters based on attack simulations and physical verification results. For defense strategies that can effectively block high-risk paths, the system increases their effectiveness evaluation; for defense strategies that fail to block critical paths or only cause minor path detours, the system decreases their effectiveness evaluation and marks the locations that need adjustment.

[0243] The system can compare the Top-k attack paths, changes in key nodes, changes in path scores, changes in target trigger support, and changes in physical impact before and after defense to determine whether the defense strategy has truly reduced the attack risk. If the defense strategy only reduces the network path score but does not reduce the physical or business impact, the system will mark the strategy as insufficiently effective.

[0244] Meanwhile, the system adjusts the probability perturbation amplitude, environmental interference trigger probability, path search threshold, and the number of Top-k outputs based on the results of multiple rounds of perturbation simulations, so that the simulation results are closer to the real operating environment while maintaining interpretability.

[0245] Step b8, output calibration results and optimized model: The output includes the updated knowledge entry status, rule template set, cross-domain coupling hypergraph weights, cross-domain causal graph attributes, defense strategy evaluation results, parameter optimization results, and model version information.

[0246] The system generates a version record for each calibration, documenting the attack path used in this calibration, physical verification results, the updated object, parameters before the update, parameters after the update, and the reason for the update. This version record is used for model backtracking, auditing, and subsequent reproduction experiments.

[0247] Through the above steps, the system forms a closed-loop technical chain: knowledge base generation of rule templates, generation of cross-domain coupling hypergraphs from rule templates, constraint of cross-domain causal graphs by cross-domain coupling hypergraphs, attack inference driven by cross-domain causal graphs, and back-calibration of the knowledge base and model based on the inference and physical verification results. Compared with static attack graphs or one-time simulation methods, this embodiment enables the model to be continuously optimized based on new event data, simulation results, and defense effectiveness, improving the credibility, interpretability, and engineering applicability of cross-domain attack inference.

[0248] The algorithm for constrained path derivation, physical verification, and feedback calibration provided in this embodiment is as follows: Input: Trusted cross-domain coupling hypergraph Cross-domain causal graph Network attack scenario template Physical mapping and verification rules.

[0249] Output: Top-k candidate paths, reasons for path stopping or blocking, physical verification results, and calibration records.

[0250] Step 1: Generate or select candidate causal edges only among node combinations that have been verified to support superedges.

[0251] Step 2: Before extending the path across domains, check the transfer conditions item by item; if any necessary condition is not met, stop the extension and record the reason.

[0252] Step 3: Calculate the propagation support score for edges that meet the conditions. Calculate candidate paths and .

[0253] Step 4: Paths involving physical effects must pass control action mapping, physical preconditions, and physical verification; failed paths must not be output as trusted physical high-risk paths.

[0254] Step 5: Update calibrable properties based on verification records; do not perform updates if there is a lack of observation data or if there is a conflict with hard constraints.

[0255] It should be noted that the following alternative solutions can also be used in this embodiment: 1. The cross-domain coupling model can be implemented using a single hypergraph or a hypergraph layered according to the information domain, communication domain, control domain, physical domain, and service impact domain; as long as the multi-node joint premise, cross-domain transfer conditions, and verification status are retained, they can all be used as alternative implementations of this invention.

[0256] 2. Candidate causal edges can be generated from attack scenarios, mechanism rules, historical event sequences, or a combination of the three. Regardless of the source, candidate edges must be supported by verified superedges and meet the corresponding cross-domain transfer conditions before entering the trusted causal graph.

[0257] 3. Path search can employ depth-first search, breadth-first search, shortest path search, heuristic search, or k-optimal path search algorithms. The search algorithm itself is replaceable, but hyperedge support, hard constraints, defensive blocking, and time windows must be checked before each cross-domain expansion.

[0258] 4. Physical verification can be achieved through power flow calculation, transient simulation, protection action simulation, digital twin or power CPS co-simulation; when an external simulation engine is not available, the physical feasibility can be determined using verified topology, equipment status, protection logic and operating boundary rules.

[0259] 5. Closed-loop calibration can use incremental weighting, sliding window, Bayesian update, or classification threshold update, but updates are only allowed for attributes marked as calibrable; updates will not be performed if there is a lack of observational evidence, validation fails, or there is a conflict with hard constraints.

[0260] The cross-domain attack simulation method for power cyber-physical systems provided in this embodiment is improved not by using knowledge bases, hypergraphs, causal graphs, or power simulations alone, but by organizing these objects into a continuous link according to the simulation process, enabling mutual constraints and feedback between knowledge, topology, causality, simulation results, and physical verification results. Its beneficial effects are mainly reflected in the following aspects: (1) This invention first organizes knowledge in the power field, safety knowledge, protocol rules, physical mechanisms, defense strategies, and expert rules into structured knowledge items; then, after completeness, credibility, and conflict verification, it is converted into rule templates. The rule templates contain node selectors, prior weights, source confidence, and verification strategies, which can be directly used for subsequent node matching, candidate hyperedge generation, and candidate causal edge generation. After this processing, knowledge is no longer just static storage content, but a rule basis that can enter the deduction process.

[0261] (2) This invention utilizes a cross-domain coupled hypergraph to describe the multi-node coupling relationships in a power information physical system. A single hyperedge can simultaneously associate multiple objects in the information domain, communication domain, control domain, physical domain, and business impact domain, and record defense constraints, weights, semantic labels, and verification states. Compared to using only binary edges to represent local relationships, this structure can directly express the high-order dependencies of a cross-domain state transition jointly constrained by attack events, communication states, control devices, defense conditions, and physical topology.

[0262] (3) This invention does not directly generate causal paths based on event co-occurrence or statistical correlation, but generates cross-domain causal graphs under cross-domain coupling hypergraph constraints. Candidate causal edges need to be supported by verified hyperedges and satisfy causal type, direction, and constraint conditions before entering the trusted causal graph. Different types of causal edges are also subject to protocol, timing, defense, business, and physical constraints, thus reducing invalid paths that do not conform to the actual system connection relationships or power operation mechanisms.

[0263] (4) This invention sets causal strength, confidence level, propagation delay and constraint attributes for causal edges, and introduces defense blocking edges and defense residual coefficients to calculate path scores. As a result, the system can not only determine whether an attack path exists, but also compare the risk levels of different paths, output high-risk paths, blocked links and residual risks, and provide a basis for the verification of defense strategies.

[0264] (5) This invention further maps the attack simulation results into identifiable physical actions or changes in operating status of the power system, and verifies them in conjunction with topology constraints, power flow constraints, protection logic constraints, equipment status constraints, or power simulation results. This can distinguish between physically feasible paths, conditionally feasible paths, and physically infeasible paths, avoiding outputting paths that are reachable from the network side but not physically feasible as high-risk paths.

[0265] (6) This invention uses attack simulation results and physical verification results to calculate causal matching degree, and updates causal edge attributes, hyperedge weights, rule template confidence and knowledge item status accordingly. Through this closed-loop process, the model can be gradually corrected with new events, new scenarios and simulation results, rather than remaining on a one-time modeling result.

[0266] Furthermore, this invention sets hard constraint edges and calibrable edges to prevent statistical data from arbitrarily overriding the inherent relationships in power physics. Through probabilistic perturbations and environmental disturbances during the simulation, it simulates uncertainties such as communication delays, equipment failures, data noise, and human error without adding unverified attack logic, thereby improving the adaptability of the simulation results to the real operating environment. This preserves the constraints of power system mechanisms while giving the model a certain degree of self-adaptability. As one or more specific application embodiments of this invention, combined with... Figure 4The cross-domain attack simulation method for power cyber-physical systems provided by this invention will be further described in detail, such as... Figure 4 As shown, the specific process is as follows: Step S1, Knowledge Validation and Rule Initialization: Import knowledge entries and atomic models, perform field integrity, source credibility, and logical conflict validation on the knowledge entries, convert the validated knowledge entries into executable rule templates, and output the rule template set, the knowledge set to be supplemented, and the atomic model library.

[0267] Step S2, Network Attack Scheme Scenario and Cross-Domain Coupling Hypergraph Construction: Generate network attack scenario template, generate candidate hyperedges based on rule template and atomic model and calculate initial weights, perform logical, temporal, data, defense, physical and cross-domain transfer condition verification on candidate hyperedges, and output the verified weighted cross-domain coupling hypergraph and candidate hyperedges to be verified.

[0268] Step S3, Cross-domain causal graph construction and attack process deduction: Generate multi-type candidate causal edges only between node combinations that have been verified to support superedges; check the corresponding transition conditions before expanding the candidate path across domains, stop the expansion and record the reason for stopping if the conditions are not met; calculate the score and latency of the path that meets the conditions, and output the Top-k candidate paths and their evidence sources.

[0269] Step S4, Calculation and Physical Verification of Network Attack Impact: For candidate paths involving physical impact, check the effective control actions, control object mapping and physical preconditions, convert events that meet the mapping conditions into physical action descriptions, and use physical constraints or power simulation results to determine physical feasibility and impact level.

[0270] Step S5, Closed-loop calibration and system optimization: Calculate the matching degree based on the event sequence, inference results and physical verification results, update the calibrable causal edge attributes, hyperedge weights and rule confidence; the updated results re-enter the knowledge verification process, and must not cover the hard constraints determined by the protocol premise, device binding, physical topology, protection logic and hard defense conditions.

[0271] The execution process of this invention is illustrated using a scenario where abnormal access to the remote operation and maintenance entry point may affect the substation remote control operation. Inputs include the attack entry point, the target monitoring and control device and its associated physical objects, the simulation time window, the current access control and whitelist status, and asset association, communication reachability, control permissions, device binding, and physical constraint rules provided and verified by the applicant. Conditions not provided or not verified are not presumed to be true by the system.

[0272] In step S1, the system maps the aforementioned knowledge entries to prerequisites, applicable objects, result objects, constraints, knowledge sources, and verification status, respectively. Knowledge entries with complete fields and available verification status generate rule templates; knowledge entries lacking control object binding or physical prerequisites are added to the supplementary set and do not participate in trusted hyperedge generation.

[0273] For example, asset-related knowledge entries Record remote operation and maintenance objects, communication objects, and their verified associations; the status is "verified"; control objects are bound to knowledge entries. Record the binding relationship between the measurement and control object and the target physical object. Generate structural or communication constraint rule templates when fields are complete. ;like If there is a lack of verifiable binding information for the target object, then Enter It does not generate trusted rule templates from the control domain to the physical domain.

[0274] like Figure 5 As shown, in step S2, the system instantiates remote operation and maintenance objects, communication objects, measurement and control objects, associated physical objects, and remote control service objects according to the attack scenario. After the rule template matches the scenario nodes, candidate hyperedges are generated and initial weights are calculated. For candidate hyperedges from the information domain to the communication domain, from the communication domain to the control domain, and from the control domain to the physical domain, the system checks the transfer conditions explicitly recorded in their association rules. Candidate hyperedges that pass all necessary verifications are written into the trusted cross-domain coupling hypergraph; candidate hyperedges lacking control permissions, device binding, or physical preconditions are written into the set to be reviewed, and the stopping domain is recorded.

[0275] Rule template Candidate superedges are formed after matching scene nodes. Its initial weights are from Calculation of prior weights and source credibility; After passing the appropriate verification, it is written into the trusted hypergraph. Another candidate hyperedge. Although the communication object and the control object are connected, verified control permissions or target object mappings are lacking, therefore... , Only enter the set to be reviewed.

[0276] like Figure 6As shown, in step S3, the system generates candidate causal edges only between node combinations supported by the trusted cross-domain coupling hypergraph. The resulting path one sequentially reaches the communication object, the measurement and control object, and the remote control service object from the remote operation and maintenance entry point; while path two can reach the communication object, its subsequent control actions lack verified control permissions or target object mappings. Before cross-domain expansion, the system checks the transfer conditions item by item, allowing path one to continue participating in the simulation, stopping path two from expanding into the control domain or physical domain, and writing the missing conditions and rule sources into the simulation results.

[0277] For path one that can continue to expand, the system calculates the path score and total delay based on the strength, confidence, propagation delay, and defense residual coefficient of each causal edge. When the defense measures fully meet the hard blocking condition, the relevant edges will no longer expand; when the defense is partially effective, only the defense residual coefficient is adjusted. The system outputs the sorted candidate paths, blocked links, reasons for stopping, and sources of evidence.

[0278] In step S4, the system does not directly generate physical consequences based on communication anomalies. Instead, it first checks whether control actions have been triggered, whether there is a valid mapping between control actions and the target physical object, and whether the physical preconditions are met. Paths that meet the above conditions are converted into physical action descriptions and enter physical constraints or simulation verification; paths lacking any necessary conditions are retained as candidate paths on the network or control side and are not output as credible high-risk paths with physical impact.

[0279] The physical verification results categorize paths into three types: physically feasible, conditionally feasible, and physically infeasible. Paths that pass physical constraint or simulation verification and meet the validity criteria are included in the physically feasible path set; paths that are valid only under specific operating modes or equipment states are included in the conditionally feasible path set; and paths that fail due to topology, equipment state, protection logic, or other explicit physical conditions not being met are included in the physically infeasible path set, with the reasons for verification failure retained.

[0280] In step S5, the system locates relevant causal edges, superedges, and rule templates based on the aforementioned path results. Verified calibrable relationships can have their confidence or weight increased; calibrable relationships that repeatedly fail physical verification have their corresponding attributes reduced or constraints increased; hard constraints such as protocol premises, device binding, physical topology, and protection logic are not deleted or reverse-written based on a single simulation result. Updated rules still need to be verified again in step S1 before proceeding to the next round of trusted simulation.

[0281] In contrast, without verified hyperedge constraints, statistical correlations in events or logs may directly enter candidate paths; with this invention, relationships lacking rule, topology, or cross-domain transfer support are stopped or transferred to the verification set. Without physical verification, network-side reachable paths may be mistakenly identified as physically high-risk paths; with this invention, only paths with effective control action mappings and passing physical feasibility checks are output as physically influential trusted paths. This embodiment illustrates the complete technical chain of this invention, encompassing knowledge rule generation, cross-domain coupling modeling, constrained path deduction, physical verification, and feedback calibration.

[0282] Supplementary verification results: To further verify the candidate relationship screening, physical impact determination, and closed-loop calibration capabilities of this invention, cross-domain attack scenarios and protection action impact scenarios were selected for verification. Communication-side events were constructed based on publicly available data samples, and control actions, equipment state changes, and physical impacts were determined based on rule templates and power simulation results. Simulation results are not considered as field measurement results.

[0283] (1) Candidate Relation Construction and Verification. In the cross-domain attack scenario of the control channel, the system instantiated 18 scenario nodes. After matching the rule template with the scenario nodes, 41 candidate hyperedges were generated. After verification based on logic, timing, data, defense, physical, and cross-domain transfer conditions, 15 candidate hyperedges passed the verification and were written into the trusted cross-domain coupling hypergraph. The other 26 were not included in the trusted hypergraph due to the lack of necessary nodes, rule premises, or cross-domain transfer conditions. Based on the verified hyperedges, the system further generated 20 causal edges with hyperedge support records. This result shows that candidate relations are only included in the trusted inference model after being screened by rules and constraints.

[0284] (2) Physical Impact Verification. In another protection action impact scenario, the system maps the control actions in the candidate paths to the IEEE 14-node power flow model for verification. The simulation calculation converges normally, with the lowest node voltage being 0.9935 pu (pu represents per unit value) and the highest node voltage being 1.09 pu. In this embodiment, the allowable range of node voltage is set to 0.95 pu to 1.05 pu; when the node voltage is below 0.95 pu or above 1.05 pu, it is determined to be a voltage over-limit. This condition is determined to have a voltage over-limit, but no line overload occurs. Therefore, the system only outputs the voltage impact results supported by the simulation and does not directly interpret the network-side path as a line overload. This result shows that physical verification can distinguish different types of physical impacts and filter impact claims that lack physical basis.

[0285] (3) Closed-loop calibration verification. After completing the physical verification of the protection action impact scenario, the system performs bounded updates to the causal edge attributes based on the verification results. This round of calibration involves 5 causal edges, of which the causal strength and confidence of 3 calibrable edges are adjusted according to preset update coefficients, while the 2 hard constraint edges determined by device binding and physical mechanisms remain unchanged. After calibration, each causal edge still retains the corresponding superedge support record, and no new nodes, superedges, or attack logic are generated.

[0286] This result demonstrates that the present invention can correct calibrable properties based on deduction and physical verification results without rewriting hard constraints.

[0287] This embodiment also provides a cross-domain attack simulation device for power cyber-physical systems. This device is used to implement the above embodiments and preferred embodiments, and details already described will not be repeated. As used below, the term "module" can be a combination of software and / or hardware that implements a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0288] This embodiment provides a cross-domain attack simulation device for power cyber-physical systems, such as... Figure 7 As shown, it includes: The knowledge entry acquisition module 701 is used to acquire knowledge entries corresponding to multi-source knowledge. The causal model construction and management module 702 is used to verify the credibility of knowledge entries and convert the verified knowledge entries into executable rule templates. Based on the rule templates and preset scenario nodes, candidate hyperedges are generated, preset verification is performed on the candidate hyperedges, and the verified candidate hyperedges are written as verified hyperedges into the cross-domain coupling hypergraph. The cross-domain coupling hypergraph is used to represent the many-to-many coupling relationship between multi-dimensional entities in the power information physical system. Candidate causal edges are generated only between nodes supported by verified hyperedges and whose corresponding constraints are satisfied, and a cross-domain causal graph is constructed based on the candidate causal edges.

[0289] The cross-domain attack simulation and deduction module 703 is used to deduce the attack path based on the cross-domain causal graph, so as to assess the security risks of the power cyber-physical system based on the attack path.

[0290] The physical verification and closed-loop calibration module 704 is also used to verify the physical feasibility of the attack path and perform closed-loop calibration based on the attack path deduction results and the attack path verified by physical feasibility.

[0291] In one alternative implementation, the causal model building and management module includes: The credibility verification unit is used to verify the credibility of knowledge entries by checking field integrity, source credibility, and logical conflicts. Knowledge entries that pass the verification are used to generate executable rule templates.

[0292] In one optional implementation, the knowledge entries include: preconditions, applicable objects, result objects, constraints, knowledge sources, and verification status; the causal model construction and management module 702 includes: The credibility reading unit is used to read the credibility of the source based on the knowledge source. The judgment unit is used to determine whether conversion is allowed based on the verification status and source credibility. The rule template generation unit is used to generate an executable rule template if the prerequisites, applicable objects, result objects, and knowledge sources are complete, the verification status is verified, and the source credibility meets the preset conditions. Otherwise, the knowledge items are included in the knowledge set to be supplemented.

[0293] In one alternative embodiment, the device further includes: The cross-domain coupling hypergraph construction module is used to determine the nodes of the cross-domain coupling hypergraph. The nodes are derived from scene nodes and event state nodes obtained by instantiating atomic models. The atomic models include physical device atomic models, information component atomic models, behavioral subject atomic models, and business object atomic models. The nodes include information domain nodes, communication domain nodes, control domain nodes, physical domain nodes, and business influence domain nodes. The module determines the hyperedges of the cross-domain coupling hypergraph. Each hyperedge connects two or more nodes and is used to represent the many-to-many coupling relationship between multi-dimensional entities in the power information physical system. The module sets the hyperedge coupling weight, semantic label, and domain identifier for each hyperedge to generate the cross-domain coupling hypergraph.

[0294] In an optional implementation, the causal model building and management module 702 further includes: The attack scenario template building unit is used to build network attack scenario templates based on preset attack targets, attack entry points, target objects, attack stages, and scope of action. Instantiation unit, used to instantiate scene nodes from the atomic model library based on the network attack scenario template; The candidate hyperedge generation unit is used to match the node selector in the rule template with the scene nodes to generate candidate hyperedges; The multi-dimensional verification unit is used to perform pre-defined verification on candidate hyperedges. The pre-defined verification includes: logical verification, temporal verification, data verification, defense verification, and physical verification. If the candidate hyperedge connects to nodes in different domains, cross-domain transfer condition verification is also performed. The write unit is used to write the verified candidate hyperedges as verified hyperedges into the cross-domain coupled hypergraph.

[0295] In an optional implementation, the causal model building and management module 702 further includes: The initial weight calculation unit is used to calculate the initial weights of candidate hyperedges. The initial weights are jointly determined by the prior weights of the rule template and the credibility of the knowledge source.

[0296] In an optional implementation, the causal model building and management module 702 further includes: The data acquisition unit is used to acquire nodes, verified hyperedges, and associated rule templates in the cross-domain coupled hypergraph. The candidate causal edge judgment unit is used to select any two nodes. If the first node and the second node are supported by the same verified superedge, and the rule template supports the propagation from the first node to the second node, a corresponding candidate causal edge is generated; otherwise, no candidate causal edge is generated. The candidate causal edge generation unit is used to generate candidate causal edges segment by segment according to the adjacent node pairs supported by each verified hyperedge when multiple nodes form a reachable relationship through multiple verified hyperedges, without generating candidate causal edges that cross intermediate nodes; the causal edge admission verification unit is used to perform hypergraph support verification, causal type consistency verification, direction consistency verification and constraint condition verification on candidate causal edges. Among them, candidate causal edges connecting nodes in different domains must also satisfy the corresponding cross-domain transition conditions. The cross-domain causal graph construction unit is used to take the verified candidate causal edges as cross-domain causal edges and construct a cross-domain causal graph based on the cross-domain causal edges and the nodes in the cross-domain coupling hypergraph.

[0297] In an optional implementation, the causal model building and management module 702 further includes: The candidate causal edge type determination unit is used to determine the type of candidate causal edge and classify the candidate causal edge into hard constraint edge or calibrable edge according to the constraint source and allowed update range of the candidate causal edge. The types include at least one of the following: attack propagation edge, control action edge, physical influence edge, defense blocking edge, environmental disturbance edge, and business influence edge; Hard constraint edges originate from electrical physics mechanisms, mandatory safety constraints, protection action logic, equipment technical manuals, or standard specifications. During closed-loop calibration, hard constraint edges must not be deleted, and the conduction direction and hard constraint conditions of hard constraint edges must not be changed. The causal correlation strength, confidence level, and propagation delay of calibrable edges can be updated within a preset range based on the attack path deduction results and physical feasibility verification results. The updated calibrable edges still need to be supported by verified superedges and must not violate hard constraints.

[0298] In one optional implementation, the cross-domain attack simulation module 703 includes: The node determination unit is used to determine the attack entry node and the attack target node according to the attack scenario template, set the attack entry node as the initial activation node, and set the initial activation node as the current activation node. The retrieval unit is used to retrieve cross-domain causal edges starting from the currently active node. The cross-domain transfer judgment unit is used to obtain the cross-domain transfer conditions recorded by the verified super-edge associated with the cross-domain causal edge when the retrieved cross-domain causal edge connects nodes in different domains, and check the cross-domain transfer conditions before triggering the endpoint node; when all necessary conditions are met, the trigger judgment continues to be executed; when any necessary condition is not met, the current attack path is stopped from extending along the cross-domain causal edge, and the unmet conditions and their associated rule templates are recorded. The trigger judgment unit is used to determine whether to trigger the endpoint node pointed to by the cross-domain causal edge based on the type of the cross-domain causal edge, the strength of the causal relationship, the confidence level, the constraint conditions, and the defense status. The deduction unit is used to record the cross-domain causal edge and the corresponding endpoint node if triggered, and set the endpoint node as the currently active node, and continue the deduction until the attack target node is reached or the preset stopping condition is met. The attack path generation unit is used to generate attack paths based on the active nodes and cross-domain causal edges recorded during the simulation process.

[0299] In one alternative implementation, the physical verification and closed-loop calibration module 704 includes: The physical feasibility verification unit is used to map attack events into physical actions that can be identified by the power system when there are control action edges or physical influence edges in the attack path, and to perform power physical feasibility verification on the physical actions. The physical feasibility of the attack path is marked according to the verification results. Power physical feasibility verification is the verification of physical actions using topology constraints, power flow constraints, protection logic constraints, equipment state constraints and / or power simulation.

[0300] In an optional implementation, the physical verification and closed-loop calibration module 704 further includes: The model calibration and update unit is used to update calibrable attributes based on the consistency between the attack path deduction results and the physical feasibility verification results, keeping the hard constraints determined by the protocol premises, device binding, physical topology and protection logic unchanged, and correcting the rule template, cross-domain coupling hypergraph and cross-domain causal graph based on the update results.

[0301] The cross-domain attack simulation device for power cyber-physical systems provided in this embodiment of the invention can execute the cross-domain attack simulation method for power cyber-physical systems provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects for executing the method. Further functional descriptions of the above modules and units are the same as in the corresponding embodiments described above, and will not be repeated here.

[0302] Figure 8 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention.

[0303] The following is a detailed reference. Figure 8 This diagram illustrates a suitable structural schematic for implementing an electronic device according to embodiments of the present invention. The electronic device may include a processor (e.g., a central processing unit, graphics processor, etc.) 801, which can perform various appropriate actions and processes based on a program stored in read-only memory (ROM) 802 or a program loaded from memory 808 into random access memory (RAM) 803. The RAM 803 also stores various programs and data required for the operation of the electronic device. The processor 801, ROM 802, and RAM 803 are interconnected via a bus 804. An input / output (I / O) interface 805 is also connected to the bus 804.

[0304] Typically, the following devices can be connected to I / O interface 805: input devices 806 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 807 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; memory devices 808 including, for example, magnetic tapes, hard disks, etc.; and communication devices 809. Communication device 809 allows electronic devices to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 8 Electronic devices with various devices are shown, but it should be understood that it is not required to implement or have all of the devices shown, and more or fewer devices may be implemented or have instead.

[0305] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device 809, or installed from a memory 808, or installed from a ROM 802. When the computer program is executed by the processor 801, it performs the functions defined in the cross-domain attack simulation method for power cyber-physical systems according to embodiments of the present invention.

[0306] Figure 8 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments of the present invention.

[0307] This invention also provides a computer-readable storage medium. The methods described above according to embodiments of the invention can be implemented in hardware or firmware, or implemented as computer code that can be recorded on a storage medium, or implemented as computer code downloaded over a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and then stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code. When the software or computer code is accessed and executed by the computer, processor, or hardware, the cross-domain attack simulation method for power cyber-physical systems shown in the above embodiments is implemented.

[0308] A portion of this invention can be applied as a computer program product, such as computer program instructions, which, when executed by a computer, can invoke or provide the methods and / or technical solutions according to the invention through the operation of the computer. Those skilled in the art will understand that the forms in which computer program instructions exist in a computer-readable medium include, but are not limited to, source files, executable files, installation package files, etc. Correspondingly, the ways in which computer program instructions are executed by a computer include, but are not limited to: the computer directly executing the instructions, or the computer compiling the instructions and then executing the corresponding compiled program, or the computer reading and executing the instructions, or the computer reading and installing the instructions and then executing the corresponding installed program. Here, the computer-readable medium can be any available computer-readable storage medium or communication medium accessible to a computer.

[0309] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A method for extrapolating cross-domain attacks on power cyber-physical systems, characterized in that, The method includes: Obtain knowledge entries corresponding to multi-source knowledge, verify the credibility of the knowledge entries, and convert the verified knowledge entries into executable rule templates; Candidate hyperedges are generated according to the rule template and preset scene nodes. Preset verification is performed on the candidate hyperedges. The verified candidate hyperedges are written as verified hyperedges into the cross-domain coupling hypergraph. The cross-domain coupling hypergraph is used to represent the many-to-many coupling relationship between multi-dimensional entities in the power information physical system. Candidate causal edges are generated only between nodes where the superedge support has been verified and the corresponding constraints are satisfied, and a cross-domain causal graph is constructed based on the candidate causal edges; Attack paths are derived from the cross-domain causal graph, and corresponding cross-domain transfer conditions are checked before the candidate attack paths are extended across domains. When the cross-domain transfer conditions are not met, the candidate attack paths are stopped from being extended to the next domain. The security risks of the power cyber-physical system are assessed based on the attack paths. The physical feasibility of the attack path is verified, the attributes of the calibrable edges are updated within a preset range, and the existence, direction and hard constraint conditions of the hard constraint edges remain unchanged; and closed-loop calibration is performed based on the attack path deduction results and the physical feasibility verification results.

2. The method according to claim 1, characterized in that, The credibility verification of knowledge entries includes: The knowledge entries are verified for field integrity, source credibility, and logical conflict. The verified knowledge entries are used to generate executable rule templates.

3. The method according to claim 1, characterized in that, The knowledge entries include: prerequisites, applicable objects, result objects, constraints, knowledge sources, and verification status; The process of converting verified knowledge entries into executable rule templates includes: The credibility of the source is determined based on the aforementioned knowledge source; Whether conversion is allowed is determined based on the verification status and the source credibility. If the prerequisites, applicable objects, result objects, and knowledge sources are complete, the verification status is verified, and the credibility of the source meets the preset conditions, then the prerequisites are mapped to the prerequisites of the rule template, the applicable objects are mapped to the node selectors of the rule template, the result objects are mapped to the result node types of the rule template, and the constraints are mapped to the verification strategies of the rule template, thereby generating an executable rule template; otherwise, the knowledge items are included in the knowledge set to be supplemented.

4. The method according to claim 3, characterized in that, The cross-domain coupling hypergraph is constructed in the following manner: The nodes of the cross-domain coupling hypergraph are determined. The nodes include information domain nodes, communication domain nodes, control domain nodes, physical domain nodes, and business impact domain nodes. The nodes are derived from scene nodes and event state nodes obtained after instantiating the atomic model. The atomic model includes physical device class atomic model, information component class atomic model, behavior subject class atomic model, and business object class atomic model. The hyperedges of the cross-domain coupling hypergraph are determined, and each hyperedge connects two or more nodes to represent the many-to-many coupling relationship between multi-dimensional entities in the power information physical system. Set hyperedge coupling weights, semantic labels, and domain identifiers for each hyperedge to generate a cross-domain coupled hypergraph.

5. The method according to claim 4, characterized in that, The step of generating candidate hyperedges based on the rule template and preset scene nodes, performing preset verification on the candidate hyperedges, and writing the verified candidate hyperedges as verified hyperedges into the cross-domain coupled hypergraph includes: Construct a network attack scenario template based on the preset attack target, attack entry point, target object, attack stage, and scope; Based on the network attack scenario template, instantiate scene nodes from the atomic model library; The node selector in the rule template is matched with the scene node to generate candidate hyperedges; The candidate hyperedges are subjected to preset verification; the preset verification includes: logical verification, timing verification, data verification, defense verification and physical verification. If the candidate hyperedges connect to nodes in different domains, cross-domain transfer condition verification is also performed. The verified candidate hyperedges are written into the cross-domain coupled hypergraph as verified hyperedges.

6. The method according to claim 5, characterized in that, The step of generating candidate hyperedges based on the rule template and preset scene nodes also includes: Calculate the initial weights of the candidate hyperedges, which are determined by the prior weights of the rule template and the credibility of the knowledge source.

7. The method according to claim 5, characterized in that, The step of generating candidate causal edges only between nodes whose hyperedge support has been verified and whose corresponding constraints are satisfied, and constructing a cross-domain causal graph based on the candidate causal edges, includes: Obtain the nodes, verified hyperedges, and associated rule templates in the cross-domain coupled hypergraph; Select any two nodes. If the first node and the second node are supported by the same verified superedge, and the rule template supports propagation from the first node to the second node, generate the corresponding candidate causal edge; otherwise, do not generate the candidate causal edge. When multiple nodes form a reachable relationship through multiple verified hyperedges, candidate causal edges are generated segment by segment according to the adjacent node pairs supported by each verified hyperedge, and candidate causal edges that cross intermediate nodes are not generated. The candidate causal edges are subjected to hypergraph support verification, causal type consistency verification, direction consistency verification, and constraint verification. In addition, the candidate causal edges connecting nodes in different domains must also satisfy the corresponding cross-domain transition conditions. The verified candidate causal edges are used as cross-domain causal edges, and a cross-domain causal graph is constructed based on the cross-domain causal edges and the nodes in the cross-domain coupling hypergraph.

8. The method according to claim 7, characterized in that, The step of generating candidate causal edges only between nodes whose hyperedge support has been verified and whose corresponding constraints are satisfied also includes: The type of the candidate causal edge is determined, and the candidate causal edge is classified into hard-constrained edges or calibrable edges according to the constraint source and allowed update range of the candidate causal edge. The types include at least one of attack propagation edge, control action edge, physical influence edge, defense blocking edge, environmental disturbance edge, and business influence edge; The hard constraint edges are derived from electrical physics mechanisms, mandatory safety constraints, protection action logic, equipment technical manuals or standard specifications. During the closed-loop calibration process, the hard constraint edges must not be deleted, and the conduction direction and hard constraint conditions of the hard constraint edges must not be changed. The causal correlation strength, confidence level, and propagation delay of the calibrable edge can be updated within a preset range based on the attack path deduction results and physical feasibility verification results. The updated calibrable edge still needs to be supported by the verified superedge and must not violate the hard constraints.

9. The method according to claim 8, characterized in that, The attack path derived from the cross-domain causality graph includes: The attack entry node and the attack target node are determined according to the attack scenario template. The attack entry node is set as the initial activation node and the initial activation node is set as the current activation node. Starting from the currently active node, retrieve cross-domain causal edges originating from the currently active node; When a cross-domain causal edge is found to connect nodes in different domains, the cross-domain transfer conditions recorded by the verified superedge associated with the cross-domain causal edge are obtained, and the cross-domain transfer conditions are checked before triggering the endpoint node; when all necessary conditions are met, the trigger judgment continues; when any necessary condition is not met, the current attack path is stopped from extending along the cross-domain causal edge, and the unmet conditions and their associated rule templates are recorded. Based on the type, causal correlation strength, confidence level, constraint conditions, and defense status of the cross-domain causal edge, determine whether to trigger the endpoint node pointed to by the cross-domain causal edge; If triggered, the cross-domain causal edge and its corresponding endpoint node are recorded, and the endpoint node is set as the currently active node. The deduction continues until the target node is reached or the preset stopping condition is met. Attack paths are generated based on the activation nodes and cross-domain causal edges recorded during the simulation process.

10. The method according to claim 9, characterized in that, The physical feasibility verification of the attack path includes: When there are control action edges or physical influence edges in the attack path, the attack event is mapped to a physical action that can be identified by the power system, and the physical feasibility of the physical action is verified. The physical feasibility of the attack path is marked according to the verification results. The physical feasibility verification is performed by using topology constraints, power flow constraints, protection logic constraints, equipment state constraints and / or power simulation to verify the physical action.

11. The method according to claim 10, characterized in that, The closed-loop calibration based on the attack path deduction results and physical feasibility verification results includes: Based on the consistency between the attack path deduction results and the physical feasibility verification results, the calibrable attributes are updated, while keeping the hard constraints determined by the protocol premises, device binding, physical topology and protection logic unchanged. Based on the update results, the rule template, the cross-domain coupling hypergraph and the cross-domain causal graph are modified to perform closed-loop calibration.

12. A cross-domain attack simulation device for power cyber-physical systems, characterized in that, The device includes: The knowledge entry acquisition module is used to acquire knowledge entries corresponding to knowledge from multiple sources. The causal model construction and management module is used to verify the credibility of knowledge entries and convert the verified knowledge entries into executable rule templates. Based on the rule templates and preset scenario nodes, candidate hyperedges are generated, preset verifications are performed on the candidate hyperedges, and the verified candidate hyperedges are written as verified hyperedges into a cross-domain coupling hypergraph. The cross-domain coupling hypergraph is used to represent the many-to-many coupling relationships between multi-dimensional entities in the power information physical system. Candidate causal edges are generated only between nodes supported by verified hyperedges and where the corresponding constraints are satisfied, and a cross-domain causal graph is constructed based on these candidate causal edges. The cross-domain attack simulation and deduction module is used to deduce the attack path based on the cross-domain causal graph, so as to assess the security risks of the power cyber-physical system based on the attack path; The physical verification and closed-loop calibration module is used to verify the physical feasibility of the attack path and perform closed-loop calibration based on the attack path deduction results and the physical feasibility verification results.

13. The apparatus according to claim 12, characterized in that, The causal model construction and management module includes: The credibility verification unit is used to verify the credibility of the knowledge entries in terms of field integrity, source credibility, and logical conflict. The knowledge entries that pass the verification are used to generate executable rule templates.

14. The apparatus according to claim 12, characterized in that, The knowledge entries include: preconditions, applicable objects, result objects, constraints, knowledge sources, and verification status; the causal model construction and management module includes: A credibility reading unit is used to read the source credibility based on the knowledge source. The judgment unit is used to determine whether conversion is allowed based on the verification status and the source credibility. The rule template generation unit is used to map the prerequisites to the prerequisites of the rule template, the applicable objects to the node selectors of the rule template, the result objects to the result node types of the rule template, and the constraints to the verification strategies of the rule template if the prerequisites, applicable objects, result objects, and knowledge sources are complete, the verification status is verified, and the credibility of the sources meets preset conditions, thereby generating an executable rule template; otherwise, the knowledge items are included in the knowledge set to be supplemented.

15. The apparatus according to claim 14, characterized in that, The device further includes: A cross-domain coupling hypergraph construction module is used to determine the nodes of the cross-domain coupling hypergraph. The nodes are derived from scene nodes and event state nodes obtained by instantiating atomic models. The atomic models include physical device atomic models, information component atomic models, behavioral subject atomic models, and business object atomic models. The nodes include information domain nodes, communication domain nodes, control domain nodes, physical domain nodes, and business influence domain nodes. The module determines the hyperedges of the cross-domain coupling hypergraph, with each hyperedge connecting two or more nodes to represent many-to-many coupling relationships between multi-dimensional entities in the power information physical system. The module sets hyperedge coupling weights, semantic labels, and domain identifiers for each hyperedge to generate the cross-domain coupling hypergraph.

16. The apparatus according to claim 15, characterized in that, The causal model construction and management module also includes: The attack scenario template building unit is used to build network attack scenario templates based on preset attack targets, attack entry points, target objects, attack stages, and scope of action. An instantiation unit is used to instantiate scene nodes from an atomic model library according to the network attack scenario template; The candidate hyperedge generation unit is used to match the node selector in the rule template with the scene nodes to generate candidate hyperedges; A multi-dimensional verification unit is used to perform preset verification on the candidate hyperedge; the preset verification includes: logical verification, timing verification, data verification, defense verification and physical verification. If the candidate hyperedge connects to nodes in different domains, cross-domain transfer condition verification is also performed. The writing unit is used to write the verified candidate hyperedges as verified hyperedges into the cross-domain coupled hypergraph.

17. The apparatus according to claim 16, characterized in that, The causal model construction and management module also includes: An initial weight calculation unit is used to calculate the initial weight of the candidate hyperedge, which is determined by the prior weight of the rule template and the credibility of the knowledge source.

18. The apparatus according to claim 16, characterized in that, The causal model construction and management module also includes: The data acquisition unit is used to acquire the nodes, verified hyperedges, and associated rule templates in the cross-domain coupled hypergraph. The candidate causal edge judgment unit is used to select any two nodes. When the first node and the second node are supported by the same verified superedge, and the rule template supports the propagation from the first node to the second node, a corresponding candidate causal edge is generated; otherwise, the candidate causal edge is not generated. The candidate causal edge generation unit is used to generate candidate causal edges segment by segment according to the adjacent node pairs supported by each verified superedge when multiple nodes form a reachable relationship through multiple verified superedges, without generating candidate causal edges that cross intermediate nodes. The causal edge admission verification unit is used to perform hypergraph support verification, causal type consistency verification, direction consistency verification and constraint condition verification on the candidate causal edges. Among them, the candidate causal edges connecting nodes in different domains must also satisfy the corresponding cross-domain transfer conditions. The cross-domain causal graph construction unit is used to take the verified candidate causal edges as cross-domain causal edges and construct a cross-domain causal graph based on the cross-domain causal edges and the nodes in the cross-domain coupling hypergraph.

19. The apparatus according to claim 18, characterized in that, The causal model construction and management module also includes: A candidate causal edge type determination unit is used to determine the type of the candidate causal edge and classify the candidate causal edge into hard constraint edges or calibrable edges according to the constraint source and allowed update range of the candidate causal edge. The types include at least one of attack propagation edge, control action edge, physical influence edge, defense blocking edge, environmental disturbance edge, and business influence edge; The hard constraint edges are derived from electrical physics mechanisms, mandatory safety constraints, protection action logic, equipment technical manuals or standard specifications. During the closed-loop calibration process, the hard constraint edges must not be deleted, and the conduction direction and hard constraint conditions of the hard constraint edges must not be changed. The causal correlation strength, confidence level, and propagation delay of the calibrable edge can be updated within a preset range based on the attack path deduction results and physical feasibility verification results. The updated calibrable edge still needs to be supported by the verified superedge and must not violate the hard constraints.

20. The apparatus according to claim 19, characterized in that, The cross-domain attack simulation module includes: The node determination unit is used to determine the attack entry node and the attack target node according to the attack scenario template, set the attack entry node as the initial activation node, and set the initial activation node as the current activation node; The retrieval unit is used to retrieve cross-domain causal edges starting from the currently active node. The cross-domain transfer judgment unit is used to obtain the cross-domain transfer conditions recorded by the verified super-edge associated with the cross-domain causal edge when the retrieved cross-domain causal edge connects nodes of different domains, and check the cross-domain transfer conditions before triggering the endpoint node; when all necessary conditions are met, the trigger judgment continues to be executed; when any necessary condition is not met, the current attack path is stopped from extending along the cross-domain causal edge, and the unmet conditions and their associated rule templates are recorded. The triggering judgment unit is used to determine whether to trigger the endpoint node pointed to by the cross-domain causal edge based on the type, causal correlation strength, confidence level, constraint conditions and defense status of the cross-domain causal edge. The deduction unit is used to record the cross-domain causal edge and the corresponding endpoint node if triggered, and set the endpoint node as the currently active node, and continue the deduction until the attack target node is reached or the preset stopping condition is met. The attack path generation unit is used to generate attack paths based on the active nodes and cross-domain causal edges recorded during the simulation process.

21. The apparatus according to claim 20, characterized in that, The physical verification and closed-loop calibration module includes: The physical feasibility verification unit is used to map attack events into physical actions that can be identified by the power system when there are control action edges or physical influence edges in the attack path, and to perform power physical feasibility verification on the physical actions, and mark the physical feasibility of the attack path according to the verification results; the power physical feasibility verification is to verify the physical actions using topology constraints, power flow constraints, protection logic constraints, equipment state constraints and / or power simulation.

22. The apparatus according to claim 21, characterized in that, The physical verification and closed-loop calibration module also includes: The model calibration and update unit is used to update the calibrable attributes of causal edges based on the consistency between the attack path deduction results and the physical feasibility verification results, keeping the hard constraints determined by the protocol premises, device binding, physical topology and protection logic unchanged, and correcting the rule template, the cross-domain coupling hypergraph and the cross-domain causal graph based on the update results.

23. An electronic device, characterized in that, include: A memory and a processor are interconnected, the memory stores computer instructions, and the processor executes the computer instructions to perform the cross-domain attack simulation method for power cyber-physical systems as described in any one of claims 1 to 11.

24. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing the computer to execute the cross-domain attack simulation method for power cyber-physical systems as described in any one of claims 1 to 11.

25. A computer program product, characterized in that, It includes computer instructions for causing a computer to execute the cross-domain attack simulation method for power cyber-physical systems as described in any one of claims 1 to 11.