A Cyber-Physical Risk Propagation Analysis Method and Device
By employing a cyber-physical integrated risk propagation analysis method, the problem of risk assessment for the control system of refining and chemical enterprises under cyberattacks was solved. This method enables reliable prediction of risk propagation paths and closed-loop security protection, and provides a solution for multi-level node combination explosions in refining and chemical enterprises.
Patent Information
- Application Number
- CN202210773558.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-07-01
- Publication Date
- 2026-11-10
- Estimated Expiration
- 2042-07-01
AI Technical Summary
Risk assessment of refining and chemical enterprise control systems under cyberattacks is difficult to quantify. Traditional methods cannot effectively assess the security risks caused by attacks, and the risk propagation paths are complex and ever-changing, lacking a unified quantitative standard and effective protection measures.
This paper presents a cyber-physical risk propagation analysis method. By establishing an asset assessment system, mining physical risk variables and information security vulnerabilities, constructing a vulnerability-function correlation model, predicting the information-physical risk propagation path, combining propagation entropy and Bayesian attack graphs for path analysis, and using ant colony optimization algorithm to optimize the risk diffusion path.
It enables reliable prediction of risk propagation paths in the control systems of refining and chemical enterprises, reduces the complexity of risk path reasoning, improves the reliability of risk diffusion paths, and provides data support for risk-based closed-loop safety protection.
Smart Images

Figure CN117375868B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of cyber-physical security defense for control systems in refining and chemical enterprises, specifically to a cyber-physical integrated risk propagation analysis method and device. Background Technology
[0002] The Chemical Cyber-Physical System (CCPS) in refining and chemical enterprises is developed based on cyber-physical systems. With the deep integration of OT and IT, the connection between traditional refining process control systems and information systems is becoming increasingly close, forming CCPS. CCPS deeply integrates process, communication, and control, and is a complex system composed of multiple subsystems. The physical and network systems are tightly coupled, making it vulnerable to attacks in both cyberspace and physical processes. Unlike IT systems, attacks against CCPS often spread from the information domain to the physical domain, causing more serious consequences. Therefore, ensuring the safe operation of refining and chemical enterprise control systems is crucial.
[0003] In the past, research on information security protection in refining and chemical enterprises mainly focused on the IT field, with the primary approach being "defense at the perimeter," deploying firewalls and other protective measures at the internal network boundary of the refining and chemical enterprise. However, due to the significant differences between the control systems and IT systems of refining and chemical enterprises, and the fact that firewalls and other protective measures can only provide passive protection against known attacks, assessing the harm of network attacks, classifying security levels, and assisting in security policy decisions are crucial once an attack enters the refining and chemical control system. To accurately assess the security risks caused by attacks, it is necessary to analyze changes in the system's state behavior, model the information attack propagation behavior, and explore the risk diffusion paths triggered by the attack. The feasibility of the above attack behavior modeling presupposes that the asset loss factors caused by the attack must be modeled in a unified formal manner; otherwise, it is impossible to quantify the correlation between changes in the state of variables.
[0004] Risk assessment is a crucial component of risk-based closed-loop security protection for refining and chemical enterprise control systems. However, the processes of refining and chemical enterprise control systems are lengthy, variables are tightly coupled, and operating conditions are complex and variable, making traditional Hazard and Operability (HAZOP) analysis methods inadequate for assessing system risks. This paper uses grey hierarchical analysis to classify the influencing factors of cyberattacks into 12 elements across 4 dimensions, thereby establishing a unified quantitative indicator system for refining and chemical control system assets under cyberattacks. Simultaneously, the CCPS (Combined Cyber-Physical System) information-physical layer risk propagation process is complex and variable, requiring modeling of system behavior states to infer the attack diffusion process. Therefore, risk assessment of refining and chemical enterprise control systems under cyberattacks necessitates a novel risk propagation model for integrated cyber-physical risk propagation analysis, thereby providing data support for risk-based closed-loop security protection for refining and chemical enterprises. Summary of the Invention
[0005] This invention aims to provide a cyber-physical integrated risk propagation analysis method and device for the control system of refining and chemical enterprises under cyberattacks. It addresses the problem of quantifying the harm of attacks on the control system, facilitating subsequent correlation analysis between process variables and information vulnerabilities; it solves the problem of low confidence in causal relationships in the control system by quantitatively calculating the correlation coefficients of process variables and the probability of vulnerability exploitation, thereby quantifying the path generation process and improving the reliability of risk propagation paths; and it addresses the problem of multi-level node explosion in refining and chemical enterprises by modeling the communication behavior and physical processes of the control system, and optimizing the risk propagation path by combining the dual objectives of shortest attack and shortest accident paths, effectively reducing the complexity of risk path reasoning.
[0006] To address the aforementioned technical problems, this invention provides, on the one hand, a cyber-physical integrated risk propagation analysis method, such as... Figure 1 As shown, it includes the following steps: Establish an asset assessment system and provide asset risk scores; Based on the asset assessment system, physical risk variables and information security vulnerabilities are identified. By utilizing the aforementioned physical risk variables and information security vulnerabilities, a "vulnerability-function" correlation model is established; Based on the aforementioned "vulnerability-function" correlation model, predict the information-physical risk propagation path.
[0007] Furthermore, the step of mining physical risk variables and information security vulnerabilities based on the asset assessment system includes: Based on the aforementioned asset assessment system, a risk level determination is made; Calculate the asset point value based on the aforementioned risk level determination; Based on the value of the asset points, physical risk variables and information security vulnerabilities are identified.
[0008] The step of mining physical risk variables and information security vulnerabilities based on the asset value includes: The asset values are sorted and filtered to form a set of key assets; The key asset set is divided into physical risk variables and information security vulnerabilities.
[0009] Furthermore, the asset assessment system includes: Personal injury, property damage, environmental pollution, and equipment malfunction; Operational compliance, personnel protection measures, emergency escape plans, equipment costs, production slowdown rate, social impact, equipment reliability design, mean time between failures (MTBF), redundancy design, media release hazards, environmental risk control level, and external environmental resilience.
[0010] The risk level determination includes: Establish an asset safety risk level matrix to measure the risk level of the asset elements from two dimensions: the probability of an accident and the severity of the consequences of the accident.
[0011] Specifically, the formula for calculating the risk level is as follows: , in, L The risk level corresponding to the asset element. The probability of an accident occurring. The severity of the consequences of the accident.
[0012] Furthermore, the calculation of asset point value based on the risk level determination includes: Analyze the asset elements, establish a numerical weighting coefficient calculation method, and construct a judgment matrix; The judgment matrix is normalized, the largest eigenvalue of the normalized matrix is calculated, and the consistency of the elements of the judgment matrix is checked. The asset point value is calculated by combining the asset risk score, the element safety risk level, and the element weight coefficients that have passed the maximum eigenvalue consistency test.
[0013] Specifically, the formula for calculating the weighting coefficient is as follows: , in, For risk factor weighting coefficients, n For the number of elements, To determine matrix elements, For correction factor, Risk factors Risk factors The relative importance, r ji Risk factors Risk factors The relative importance of.
[0014] Specifically, the largest eigenvalue λ max The calculation formula is: , in, λ max It is the largest eigenvalue. n For the number of elements, To determine the matrix, T As a risk factor, Risk factors i eigenvectors.
[0015] Specifically, the formula for calculating the value of the asset points is as follows: , in, For asset point value, For the first i The risk level of each risk element For the first i The scores of each risk factor, For the first i The weighting coefficients of each risk factor.
[0016] Furthermore, the establishment of a "vulnerability-function" correlation model using the physical risk variables and information security vulnerabilities includes: The risk variables of the physical layer are divided, and a variable association topology graph is constructed; The degree of association and propagation direction of each pair of variables in the variable association topology graph are solved based on the propagation entropy. Analyze the security vulnerabilities of the information layer and establish an attack graph model. This generates redundant penetration paths; The exploitation and consequences of the aforementioned security vulnerabilities were analyzed, and a "vulnerability-function" correlation model was established to uncover the propagation path of redundant information and physical risks.
[0017] Specifically, the formula for the transfer entropy is: , Among them, TE x→y To transfer entropy; This represents the value of the process variable y at time t+h, where h represents the time interval. express A set of historical values with a time length of (l), Let x represent the value of x at time t, and y represent the value of x at time t. t (k) Indicate A set of historical values with a time length of (k). Let (k) represent the value of Y at time t, and (l) and (k) represent the orders of the causal variable and the outcome variable, respectively. It is a random vector , It is a conditional probability density function.
[0018] Specifically, the attack graph model The generation includes: Bayesian attack graph node attribute set , wherein This is the initial set of vulnerable nodes for attackers. This is a collection of vulnerable nodes in the attack process. The set of control nodes of the attack target; It is the set of atomic attacks in the attack graph; A collection of unusual evidence; The relationship between child nodes and parent nodes can be categorized into two types: AND and OR. The prior probability of attacking a node's attribute in the graph can be calculated by multiplying the conditional probabilities of the current node and its parent node.
[0019] Furthermore, the prediction of the information-physical risk propagation path based on the "vulnerability-function" association model includes: Based on the risk propagation path and risk penetration path, the correlation coefficient between physical layer asset nodes is solved, and the propagation probability between information layer asset nodes is calculated. Based on the causal relationships between the node state changes, a complex information-physical network model of a refining and chemical enterprise is constructed. By combining the aforementioned abnormal evidence set to predict the diffusion path of the information-physical network model, an information-physical diffusion path set is formed; By combining the aforementioned "vulnerability-function" model, the ant colony algorithm is used to solve the information-physical diffusion path set to obtain the path with the greatest risk.
[0020] Specifically, the formula for the correlation coefficient between the physical layer asset nodes is as follows: , in, It is the maximum correlation coefficient between x and y across the entire interval. It is a data set The maximum mutual information entropy on, max xy<B(m) It is a dataset The maximum correlation coefficient in B(m) It is the maximum number of grid blocks into which a two-dimensional dataset can be divided. It is a two-dimensional planar dataset consisting of a time series of variables x and y. m Represents a two-dimensional dataset Total amount of data.
[0021] Specifically, the formula for the propagation probability between the information layer asset nodes is: , in, This is current evidence of the attack. The following vulnerabilities The conditional probability of being exploited. It is a known vulnerability. Evidence of an attack was detected while the system was being exploited. The conditional probability of this distribution This is current evidence of the attack. The probability of being detected; It's a vulnerability. The probability of being exploited.
[0022] Specifically, the maximum risk diffusion path The calculation formula is: , Where s is an abnormal evidence node, defined as the starting point of the risk path propagation path; d is an important asset point, defined as the target point of the cyberattack; and p is a feasible path from s to d. This is the path through which the risk spreads the most. It attacks the shortest path. It is the shortest path for the accident.
[0023] On the other hand, the present invention provides a cyber-physical integrated risk propagation device, comprising: Establish a module to build an asset assessment system and provide asset risk scores; The mining module is used to mine physical risk variables and information security vulnerabilities based on the asset assessment system. The association module is used to establish a "vulnerability-function" association model by utilizing the physical risk variables and information security vulnerabilities. The prediction module is used to predict the information-physical risk propagation path based on the "vulnerability-function" association model.
[0024] Furthermore, the asset assessment system includes: Personal injury, property damage, environmental pollution, and equipment malfunction; Operational compliance, personnel protection measures, emergency escape plans, equipment costs, production slowdown rate, social impact, equipment reliability design, mean time between failures (MTBF), redundancy design, media release hazards, environmental risk control level, and external environmental resilience.
[0025] This invention comprehensively considers the harmful impact of network attacks on the control systems of refining and chemical enterprises, and studies the generation of risk propagation paths at the information and physical layers of CCPS. It solves the problems of the lack of a unified quantitative standard for multi-dimensional assets in refining and chemical enterprises and the inability to effectively predict risk paths due to the large scale of multi-level nodes in these enterprises, thereby achieving risk-based closed-loop protection. In summary, compared with existing technologies, the above-described technical solution of this invention can achieve the following beneficial effects: The unified modeling of refining and chemical assets takes into account the characteristics of long cycle, low latency and multiple processes of refining and chemical assets. It classifies the threat impact of network attacks on assets into levels, and then establishes a multi-level and quantifiable asset value assessment index system, which reduces the subjective influence of traditional expert scoring. The process variable causal relationship analysis method based on transfer entropy proposed in this invention can quantify the causal correlation between adjacent process variables in time-series changes, improve the prediction accuracy of the propagation path of subsequent risks at the physical layer, and characterize the process of the harmfulness of network attacks spreading with the material flow. The Bayesian attack graph proposed in this invention can model the multi-step attack penetration process that utilizes complex vulnerabilities, and update the reachability probability of each attribute state node by combining abnormal evidence of risk variables, so as to ensure the prediction accuracy of the information layer risk penetration path and depict the process of network attack penetration and diffusion with information flow. In view of the close coupling between information communication and physical reaction in the control system of refining and chemical enterprises, a dual-objective optimization method based on ant colony algorithm is proposed to generate redundant risk diffusion paths caused by network attacks, so as to obtain the risk maximum diffusion path. This path serves as the system risk diffusion path of the refining and chemical control system, which facilitates subsequent global security risk assessment and closed-loop security protection based on this path.
[0026] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures pointed out in the description, claims and drawings. Attached Figure Description
[0027] Figure 1This is a schematic diagram of the cyber-physical integrated risk propagation analysis method of the present invention; Figure 2 This is a schematic diagram of the overall process of the cyber-physical integrated risk propagation analysis method of the present invention; Figure 3 This is a schematic diagram of the asset assessment system of the present invention; Figure 4 This is a schematic diagram of the risk matrix of the present invention; Figure 5 This is a schematic diagram of the hierarchical structure of the control system of the present invention; Figure 6 This is a schematic diagram of the multi-vulnerability association modeling of the present invention; Figure 7 This is a schematic diagram of the "vulnerability-function" modeling of the present invention; Figure 8 This is a schematic diagram of the dual-objective optimized cyber-physical risk diffusion path prediction method of the present invention. Detailed Implementation
[0028] 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, and 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.
[0029] To address the shortcomings of existing technologies, this invention discloses a cyber-physical integrated risk propagation analysis method, such as... Figure 1 As shown, it includes the following steps: Establish an asset assessment system and provide asset risk scores; Based on the asset assessment system, physical risk variables and information security vulnerabilities are identified. By utilizing the aforementioned physical risk variables and information security vulnerabilities, a "vulnerability-function" correlation model is established; Based on the aforementioned "vulnerability-function" correlation model, predict the information-physical risk propagation path.
[0030] Specifically, the overall process of a cyber-physical integrated risk propagation analysis method is as follows: Figure 2 As shown, it includes the following steps: Step 1: Combining knowledge of the refining and chemical industry, establish a multi-domain asset assessment system for the refining and chemical enterprise control system, covering four dimensions: personnel casualties, property losses, environmental pollution, and equipment functions, and 12 elements: operational compliance, personnel protection measures, emergency escape plans, equipment costs, production slowdown rate, social impact, equipment reliability design, mean time between failures, redundancy design, media leakage hazards, environmental risk control level, and external environmental risk resistance. Expert scores are then provided based on the characteristics of each asset.
[0031] Furthermore, step 1 above includes steps 1.1-1.2: Step 1.1: Establish an asset assessment system covering the industrial control system of refining and chemical enterprises based on knowledge of the industry. This system encompasses four dimensions: personnel casualties, property damage, environmental pollution, and equipment functionality. From the personnel casualty dimension, it further subdivides into three elements: operational compliance, personnel protection measures, and emergency escape plans, quantifying the potential threat posed by a cyberattack on personnel within the refining and chemical plant. From the equipment cost dimension, it subdivides into three elements: equipment cost, production slowdown rate, and social impact, quantifying the potential threat posed by a cyberattack on the production, property, and reputation of the refining and chemical enterprise. From the equipment functionality dimension, it subdivides into three elements: equipment reliability design, mean time between failures (MTBF), and redundancy design, quantifying the potential threat posed by a cyberattack on the entire operation of the refining and chemical enterprise. From the environmental pollution dimension, it subdivides into three elements: media leakage hazards, environmental risk control level, and external environmental resilience, quantifying the potential threat posed by a cyberattack on the surrounding natural environment of the refining and chemical enterprise. A multi-domain asset assessment system for refining and chemical enterprise control systems is as follows: Figure 3 .
[0032] Step 1.2: Measure the asset risk score by combining the characteristics of the asset and the risk assessment criteria. The asset assessment system involves 12 elements, therefore, it is necessary to develop corresponding scoring standards based on industry standards to evaluate them. The grading standards for equipment cost elements adopted in this invention are shown in Table 1.
[0033] Table 1
[0034] Step 2: Measure the risk level of the 12 asset elements from two dimensions: the probability of an accident and the severity of its consequences.
[0035] Furthermore, step 2 above includes steps 2.1-2.2: Step 2.1: Based on the asset identification system for multiple refining fields, extract 12 asset elements.
[0036] Step 2.2: The probability of an accident and the severity of its consequences are the foundation for determining the risk level of each element. According to industry standards, each is divided into five levels. In the probability of accident level, levels 1 to 5 represent increasing probability of the accident. Similarly, in the severity of consequences level, levels 1 to 5 represent increasing severity of the consequences. Based on this, a risk matrix of 12 elements is established to classify the risk of asset elements from two dimensions and calculate the risk level of each element. The risk matrix is as follows: Figure 4 As shown, the formula for calculating the risk level is as follows: (1) Where L represents the risk level of the asset element, r represents the probability of an accident occurring, and h represents the severity of the consequences of the accident.
[0037] Step 3: Based on the different weights of the 12 elements in the asset evaluation process, calculate the weight of each element using the gray-scale hierarchical analysis method, and combine the asset point score and element risk level to comprehensively calculate the asset point value.
[0038] Furthermore, step 3 above includes steps 3.1-3.5: Step 3.1: The relative importance of the 12 risk factors of refining assets varies, and a numerical weighting coefficient calculation method needs to be established through grey hierarchical analysis.
[0039] Step 3.2: Analyze the pairwise importance of the 12 risk factors and construct a judgment matrix from the comparison results.
[0040] Typically, the scoring of refining assets is conducted by a panel of n experts, who will base their assessments on the 12 elements established in step 1. The importance of each pair of objects is compared, and the results form a gray-positive-inverse matrix for judgment: (2) Step 3.3: Normalize the judgment matrix and calculate the largest eigenvalue of the normalized matrix.
[0041] First, the judgment matrix... Perform whitening to obtain its whitening matrix. Then, normalize it to obtain the largest eigenvalue.
[0042] (3) in, Let n be the largest eigenvalue and n be the number of features. To determine the matrix, Let i be the feature vector of risk element i.
[0043] Step 3.4: Calculate the weight coefficient of each element in asset determination based on the largest eigenvalue.
[0044] Specifically, to ensure that the importance of a certain element changes when it deviates significantly from its normal state during the asset assessment process, a correction coefficient is introduced for balance adjustment. The formula for calculating the weight coefficient of each element is as follows: (4) in, Here, n represents the weighting coefficients for risk factors, and n is the number of factors. To determine matrix elements, For correction factor, Risk factors Risk factors The relative importance of.
[0045] Step 3.5: Calculate the asset point value by combining the asset risk score, the factor safety risk level, and the factor weighting coefficient. (5) in, For asset point value, Let i be the risk level of the i-th risk element. Let i be the score for the i-th risk element. is the weighting coefficient of the i-th risk element.
[0046] Step 4: Analyze the asset value of refining enterprises and identify the physical risk variables and information security vulnerabilities that pose the greatest attack threat.
[0047] Furthermore, step 4 above includes steps 4.1-4.2: Step 4.1: Sort the asset values of the refining and chemical enterprises within the scope of analysis, and select asset points that are higher than the important asset threshold to form the key asset set K.
[0048] Specifically, within the scope of the analysis, assets are ranked according to the point values calculated in step 3, and thresholds for significant assets are established based on acceptable asset loss criteria. After comparison, the key asset set K is clustered together.
[0049] Step 4.2: Based on the location of the asset points in the refining and chemical control system, divide the key asset set into two sets: physical risk variables. Information security vulnerability collection .
[0050] Step 5: Based on the knowledge of process topology and vulnerability information, establish a causal relationship model and a vulnerability exploitation model between the state changes of process variables, respectively. On this basis, establish a "vulnerability-function" correlation model.
[0051] Furthermore, step 5 includes the following steps: Step 5.1: For the risk variables of the physical layer, the overall process is initially divided based on the process knowledge constraints in the refining process (e.g., reactor, condenser, reaction unit, crystallization unit, etc.), and variables that belong only to a certain sub-module are directly assigned to that sub-module. On this basis, the topological relationship between variables is further constructed.
[0052] The process control systems of refining and chemical enterprises are generally characterized by numerous procedures, wide distribution, many variables, and complex relationships. Directly performing correlation calculations on variables within the analysis scope would waste significant resources on redundant and useless calculations. Therefore, before performing causal correlation calculations, it is necessary to leverage knowledge from the refining and chemical industry to divide the highly correlated variables within the analysis scope into regions, thereby improving the efficiency of correlation calculations. First, process knowledge constraints within the industrial process are selected to initially divide the overall process, and variables belonging only to a specific sub-module are directly assigned to that sub-module. Variables not assigned to a sub-module are then further classified: one type has no parent variable in the previous sub-module; another type has no child variable in the subsequent sub-module; and a third type has a parent variable in the previous sub-module or a child variable in the subsequent sub-module. If, after the second classification, a variable falls into the third category, these variables are defined as correlated variables for module division and integration. If, after the second classification, a variable falls into the first two categories, these variables are defined as uncorrelated variables. If a variable has a parent variable in a previous module, it is assigned to that previous sub-module. If a variable has a child variable in a subsequent sub-module, it is assigned to that subsequent sub-module. If a variable does not meet all of the above conditions, it is selected separately, and after structural integration, a specific transitive relationship is added to it based on process knowledge constraints.
[0053] Step 5.2: Solve for the degree of association and propagation direction of each pair of variables in the variable association topology graph based on the propagation entropy.
[0054] Modeling the relationships between variables in chemical processes is essentially solving for the causal relationships between time-series variables. Transmission entropy establishes causal relationships between upstream and downstream variable responses based on its asymmetry. The causal relationships determined by transmission entropy primarily reflect the uncertainties in the changes between variables. This invention selects the direction of greatest change as the risk propagation direction, assuming... If the elements are in the candidate set of risk-related process variables, the formula for calculating the transit entropy between variables is as follows: (6) Among them, TE x→y To transfer entropy; This represents the value of the process variable y at time t+h, where h represents the time interval. express A set of historical values with a time length of (l), Let x represent the value of x at time t, and y represent the value of x at time t. t (k) Indicate A set of historical values with a time length of (k), Let represent the value of y at time t, and (k) and (l) represent the orders of the causal variable and the outcome variable, respectively. It is a random vector , It is a conditional probability density function.
[0055] Step 5.3: For security vulnerabilities in the information layer, analyze the vulnerability exploitation scenarios and the consequences once exploited based on vulnerability information knowledge, establish an attack graph model, and then generate redundant penetration paths for multiple vulnerabilities to be exploited sequentially or in combination under different security scenarios.
[0056] Appendix Figure 5 This is a schematic diagram of the hierarchical structure of the control system. Figure 5 The system architecture reveals a high degree of integration and close interaction between the CCPS cyberspace and physical processes, enabling real-time perception and control of industrial processes. However, this tight coupling of information, control, and production networks also provides avenues for threat penetration, allowing attacks from the internet to infiltrate the CCPS control network and disrupt the physical layer system's operation. To conduct risk propagation analysis of the information layer in refining and chemical enterprise control systems, the malicious exploitation of information vulnerabilities by attackers should be fully considered. Analysis of past cases shows that attackers' primary objective is to infiltrate information systems, thereby influencing physical processes.
[0057] Based on the information security vulnerability set obtained in step 4, and combined with the knowledge of security experts, this analysis examines the scenarios in which these vulnerabilities are exploited, the conditions required for exploitation in each scenario, and the consequences once exploited. For example, to compromise an administrator's host, an attacker might launch a multi-step attack: under certain conditions, they might exploit vl1 and vl2 to launch an atomic attack, thereby elevating their own privileges. This gain lays the foundation for further exploitation of other vulnerabilities, such as v13. This process reveals two relationships between vulnerability exploits: combined exploitation and sequential exploitation. The former refers to the simultaneous exploitation of multiple vulnerabilities to obtain a certain benefit; the latter refers to a causal relationship between vulnerability exploits, where the benefit gained from exploiting a vulnerability simultaneously becomes a condition for exploiting other vulnerabilities, such as... Figure 6 The diagram shown illustrates the multi-vulnerability association modeling.
[0058] This invention uses attribute attack graphs to depict the multi-step penetration process of network attacks. An attribute attack graph is an acyclic directed graph, which can be defined as a quintuple. The generation specifically includes: ① Bayesian attack graph node attribute set The set formula is: (7) in, Let i be the probability that the vulnerability is successfully transferred to j. This is a collection of vulnerable nodes in the attack process. This is a collection of vulnerable nodes in the attack process. This represents the set of control nodes for the attack target. Node attribute values exhibit bimorphism, and the calculation formula for node attributes is as follows: (8) ② Atomic attacks are a collection of atomic attacks in an attack graph. Atomic attacks reflect the relationships between vulnerabilities and refer to the way in which network vulnerabilities are exploited to transfer from a parent node to a child node. ③ It is a collection of abnormal evidence, referring to the detected abnormal behavior that occurs when a vulnerable node is attacked and exploited; ④ The relationship between child nodes and parent nodes can be divided into two types: AND and OR, depending on the network operation. ⑤ The prior probability of attacking a node's attribute in the graph can be calculated by multiplying the conditional probabilities of the current node and its parent node.
[0059] Step 5.4: By analyzing the vulnerability exploitation and consequences in various security scenarios, a "vulnerability-function" correlation model can be established to uncover all redundant information-physical risk propagation paths within the scope of analysis.
[0060] Based on step 5.3, redundant penetration paths are established between vulnerabilities. By linking information layer vulnerabilities to physical process variables using control loops, it can be analyzed that a network attack can discover a specific information vulnerability. The relationships between vulnerabilities are then used for layered penetration, and the harmful impact is propagated to physical layer devices through system information-physical interaction. A schematic diagram of the "vulnerability-function" association model is shown below. Figure 7 As shown.
[0061] Step 6: Based on the knowledge of process topology and vulnerability information, establish a causal relationship model and a vulnerability exploitation model between the state changes of process variables, respectively. On this basis, establish a "vulnerability-function" correlation model.
[0062] Furthermore, step 6 above includes steps 6.1-6.3: Step 6.1: Based on the redundant risk propagation path and redundant risk penetration path in step (5), combined with the abnormal evidence group of risk variables. ( (representing the state of the i-th asset node), solve for the correlation coefficient between physical layer asset nodes, and calculate the propagation probability between information layer asset nodes using the Bayesian backward algorithm.
[0063] Based on the changes in the state of risk variables obtained from anomalous evidence, the correlation coefficients between variable nodes are calculated using the maximum information transfer entropy algorithm. The specific steps are as follows: ① Take a time series of length l from two adjacent process variables X and Y, denoted as and , forming a two-dimensional dataset A grid with a topological shape of l×l can be calculated by traversing it. The maximum mutual information entropy corresponding to the total interval of x and y The calculation formula is as follows: (9) in, Represents the dataset The maximum mutual information.
[0064] ② A two-dimensional dataset with a data size of m The maximum correlation coefficient can be calculated using the following formula: (10) in, yes x and y The maximum correlation coefficient corresponding to the total interval, It is a data set The maximum mutual information entropy on, It is a variable x and y A two-dimensional planar dataset consisting of a time series. max xy<B(m) It is a dataset The maximum correlation coefficient in B(m) It is the maximum number of grid blocks that a two-dimensional dataset can be divided into, where m represents the two-dimensional dataset. Total amount of data.
[0065] ③ Normalize the correlation coefficients and then analyze the causal correlation between the overall trends of the two time series.
[0066] ④ Calculate the transmission entropy of strong correlations in the MIC network according to formula (6), and regard the path with a large transmission entropy as the risk transmission path, so that the non-directional correlation relationship is transformed into the causal relationship of risk transmission.
[0067] At the same time, by combining the abnormal evidence set to obtain the exploitation status of information layer vulnerabilities, the probability of related vulnerability nodes being exploited is calculated based on the Bayesian backward algorithm. The specific steps are as follows: ① Local conditional probability is introduced to quantify the likelihood of a vulnerability in the information layer being exploited. This probability is related to the attack payoff ratio from the parent node to the child node and the difficulty of exploiting the vulnerability. It represents the probability of a successful attack state transition when a network attack originates from the parent node and moves to the child node. The calculation formula is as follows: (11) in, Let λ be the probability that vulnerable node i will successfully migrate to node j; λ is the attack benefit ratio of vulnerable node j.
[0068] ② Combining the local conditional probabilities of each potentially exploitable vulnerability, calculate the prior probability of a node in the Bayesian network, inferred from its parent nodes. The formula for calculating the prior probability of a node is as follows: (12) in, Let j be the set of parent nodes of the vulnerable node j.
[0069] ③ In a Bayesian diagram, combined with anomaly evidence sets, vulnerable nodes can be divided into attacked nodes and nodes awaiting attack. For child nodes awaiting attack, the probability of attack reachability can be calculated using the following formula: (13) Step 6.2: Construct a complex information-physical network model for refining and chemical enterprises based on the causal relationships between node state changes.
[0070] Specifically, vulnerabilities or variables in the refining process system are first abstracted as nodes in a network. The penetration between vulnerabilities and the relationships between variables are abstracted as edges between nodes in the network, and an adjacency matrix B is defined: (14) Secondly, using the network topology, the physical layer asset point correlation coefficient and the information layer asset point state posterior probability calculated in step 6.1 are used as the edges of the complex network, and the asset points of the refining and chemical enterprise control system are used as the nodes of the complex network.
[0071] Step 6.3: Combine the abnormal evidence set to predict all redundant diffusion paths in the information-physical network, forming an information-physical diffusion path set. .
[0072] Based on steps 6.1 and 6.2, the propagation path of the risks caused by the network attack at both the information and physical layers has been completed. Next, it is necessary to analyze the process by which the information layer penetrates and affects the physical layer. The impact and failure of physical nodes are mainly caused by erroneous commands directly issued by the control system in the information layer.
[0073] Therefore, the complex network edge values of the information-physical association path in the "vulnerability-feature" association model are defined by the adjacency matrix △B: (15) Where m represents the number of edge nodes in the information layer, and n represents the number of edge nodes in the physical layer. To measure the correlation coefficient between adjacent information node i and physical node j, a binary state is established. This leads to the concatenation of redundant paths between the information layer and the physical layer, forming an information-physical diffusion path set. .
[0074] Step 7: Combining the "vulnerability-feature" model, solve the information-physical diffusion path set using the ant colony algorithm to obtain the maximum risk diffusion path. .
[0075] Maximum risk path generation involves predicting further attack actions and reasoning about their impact on cross-domain propagation, ultimately deriving the overall system security situation. Due to the attacker's agency, attackers often have different targets, which in turn generate different attack strategies and steps, resulting in significant variations in the exhibited attack behaviors. Identifying the attack target effectively improves the accuracy of maximum risk propagation path prediction. This invention optimizes attack path prediction based on dual attack targets—maximum benefit and maximum success probability—ensuring that the attacker's attack methods are effective while also ensuring that the attack's impact on the refining enterprise's control system leads to more severe accidents and a wider attack coverage. The above redundant path set is determined using an ant colony algorithm. The most probable risk diffusion path, and the specific implementation steps are as follows: ① Define the shortest attack path Shortest path to the accident The purpose of incident scenario calculation is to find the shortest path from the initial event to the result event. The shortest path under a network attack is equivalent to the path with the highest propagation probability between information nodes. The shortest path in an incident can be equivalent to the path with the highest propagation entropy under the maximum correlation coefficient, which can be transformed into the optimal solution of the following calculation formula: (16) Where i and j are any two nodes in the complex network. When nodes i and j are in a certain state, a propagation path exists; otherwise, it does not. This represents the propagation probability between i and j when they are both information layer nodes, the propagation entropy between i and j when they are both physical layer nodes, and the propagation entropy between i and j when one is in the information layer and the other in the physical layer. ; ②The calculation formula for the dual-objective path optimization model considering both the shortest attack path and the shortest incident path is: (17) Where s is the abnormal evidence node, defined as the starting point of the risk path propagation path; d is the important asset point obtained in step 4, defined as the target point of the network attack; and p is the feasible path from s to d. This is the path through which the risk spreads the most. It attacks the shortest path. It is the shortest path for the accident.
[0076] ③ Solving discrete-space multi-objective optimization problems based on the dual-objective ant colony algorithm. The core is to apply a fitness calculation method to update the global pheromone during the ant search process, so that the solution continuously approaches the optimal solution, thus finding the global optimal solution in the optimization process. The process of the dual-objective optimization method for predicting the cyber-physical risk diffusion path is as follows: Figure 8 As shown.
[0077] ④ This invention uses anomalous evidence nodes as the starting point for ants and key assets as the search target. A feasible path matrix is constructed for each ant, and the optimal path is found under the guidance of pheromones and heuristic information, with pheromones updated accordingly. The formula for the transition probability of an ant selecting a node is as follows: (18) in, Let be the transition probability of ant k choosing node j from node i. For heuristic functions, As an information heuristic factor, As the expected heuristic factor, This is a selectable set of nodes.
[0078] ⑤ Update the non-dominated set: To maintain the diversity of optimal solutions and keep them on the frontier in the solution space as much as possible, the non-dominated set needs to be updated. First, feasible solutions need to be found. non-dominated sets in ,Will Add global optimal solution Calculate the dominance relationships, eliminate dominated solutions, and complete the process. Update.
[0079] ⑥ Niche fitness calculation, if the current external solution set has A non-inferior solution Microhabitat fitness is calculated using the following formula: (19) (20) (twenty one) in, Let i be the sharing function for i and j in the non-dominated solution set. For the radius of the small habitat, This refers to the number of minor living conditions.
[0080] ⑦ Based on the minimum solution of the niche, after the first generation of ants completes one search, the global update formula for pheromones on the trajectory is: (twenty two) (twenty three) in, The global volatile coefficient of pheromones. Let be the increment of pheromone when ant k travels through path (i,j). This is the minimum solution for the niche. The intensity of the pheromone.
[0081] ⑧ After completing a search, the ant updates the path it has traversed locally, using the following formula: (twenty four) in, The local volatile coefficient of pheromones. This is the initial pheromone.
[0082] ⑨ After iterative optimization, the path of maximum risk propagation was determined. .
[0083] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A cyber-physical risk propagation analysis method, characterized in that, The method includes: Establish an asset assessment system and provide asset risk scores; Based on the asset assessment system, physical risk variables and information security vulnerabilities are identified. By utilizing the aforementioned physical risk variables and information security vulnerabilities, a "vulnerability-function" correlation model is established; Based on the aforementioned "vulnerability-function" correlation model, predict the information-physical risk propagation path; The establishment of a "vulnerability-function" correlation model using the physical risk variables and information security vulnerabilities includes: Risk variables at the physical layer are categorized, and a variable correlation topology graph is constructed. The degree of association and propagation direction of each pair of variables in the variable association topology graph are solved based on the propagation entropy. Analyze security vulnerabilities at the information layer and establish an attack graph model. This generates redundant penetration paths; where S is the set of node attributes in the Bayesian attack graph, A is the set of atomic attacks in the attack graph, E represents the set of anomalous evidence, and T represents the relationship between child and parent nodes. This represents the prior probability of attacking a node's attribute in the graph. The exploitation and consequences of the aforementioned security vulnerabilities were analyzed, and a "vulnerability-function" correlation model was established to uncover the propagation path of redundant information and physical risks.
2. The analytical method according to claim 1, characterized in that, The process of identifying physical risk variables and information security vulnerabilities based on the asset assessment system includes: Based on the aforementioned asset assessment system, a risk level determination is made; Calculate the asset point value based on the aforementioned risk level determination; Based on the value of the asset points, identify physical risk variables and information security vulnerabilities. The step of mining physical risk variables and information security vulnerabilities based on the asset value includes: The asset values are sorted and filtered to form a set of key assets; The key asset set is divided into physical risk variables and information security vulnerabilities.
3. The analytical method according to claim 2, characterized in that, The asset assessment system includes: Personal injury, property damage, environmental pollution, and equipment malfunction; The asset elements involved in the personnel casualties include operational compliance, personnel protection measures, and emergency escape plans; The asset elements involved in the property loss include equipment costs, production slowdown rate, and social impact; The asset elements involved in the environmental pollution include the hazards of media leakage, the level of environmental risk control, and the ability to withstand external environmental risks. The asset elements involved in the equipment functionality include equipment reliability design, mean time between failures (MTBF), and redundancy design. The risk level determination includes: Establish an asset safety risk level matrix to measure the risk level of the asset elements from two dimensions: the probability of an accident and the severity of the consequences of the accident.
4. The analytical method according to claim 3, characterized in that, The formula for calculating the risk level is as follows: , in, L Here, r represents the risk level corresponding to the asset element, and r represents the probability of the accident occurring. The severity of the consequences of the accident.
5. The analytical method according to claim 3, characterized in that, The step of calculating the asset point value based on the risk level determination includes: Analyze the asset elements, establish a numerical method for calculating element weight coefficients, and construct a judgment matrix; The judgment matrix is normalized, the largest eigenvalue of the normalized matrix is calculated, and the consistency of the elements of the judgment matrix is checked. The asset point value is calculated by combining the asset risk score, the element safety risk level, and the element weight coefficient that passes the maximum eigenvalue consistency test.
6. The analytical method according to claim 5, characterized in that, The formula for calculating the weighting coefficient is as follows: , in, Here, n represents the weighting coefficients for risk factors, and n is the number of factors. To determine matrix elements, For correction factor, Risk factors Risk factors The relative importance, r ji Risk factors Risk factors The relative importance; The largest eigenvalue λ max The calculation formula is: , in, λ max Let n be the largest eigenvalue and n be the number of features. To determine the matrix, T As a risk factor, Let i be the feature vector of risk element i; The formula for calculating the value of the asset points is as follows: , in, For asset point value, Let i be the risk level of the i-th risk element. Let i be the score for the i-th risk element. is the weighting coefficient of the i-th risk element.
7. The analytical method according to claim 1, characterized in that, The specific formula for the transfer entropy is: , Among them, TE x→y To transfer entropy; This represents the value of the process variable y at time t+h, where h represents the time interval. express A set of historical values with a time length of (l), Let x represent the value of x at time t, and y represent the value of x at time t. t (k) express A set of historical values with a time length of (k). Let represent the value of y at time t, and (k) and (l) represent the orders of the causal variable and the outcome variable, respectively. It is a random vector , It is a conditional probability density function.
8. The analytical method according to claim 1, characterized in that, Attack graph model The generation includes: Bayesian attack graph node attribute set , wherein This is the initial set of vulnerable nodes for attackers. This is a collection of vulnerable points in the attack process. The set of control nodes of the attack target; The dependency relationships between parent and child nodes described in the text are divided into two types: AND and OR. The calculation is performed by multiplying the conditional probabilities of the current node and its parent node.
9. The analytical method according to claim 1, characterized in that, The prediction of information-physical risk propagation paths based on the "vulnerability-function" association model includes: Based on the risk propagation path and redundant penetration path, the correlation coefficient between physical layer asset nodes is solved, and the propagation probability between information layer asset nodes is calculated. Based on the propagation probability between the asset nodes, a complex information-physical network model for refining and chemical enterprises is constructed. By combining the abnormal evidence set to predict the diffusion path of the information-physical network model, an information-physical diffusion path set is formed; By combining the aforementioned "vulnerability-function" model, the ant colony algorithm is used to solve the information-physical diffusion path set to obtain the path with the greatest risk.
10. The analytical method according to claim 9, characterized in that, The formula for the correlation coefficient between the physical layer asset nodes is: , in, It is the maximum correlation coefficient between x and y across the entire interval. It is a data set The maximum mutual information entropy on, It is a two-dimensional planar dataset consisting of a time series of variables x and y. max xy<B(m) It is a dataset The maximum correlation coefficient in B(m) It is the maximum number of grid blocks into which a two-dimensional dataset can be divided. m It is a two-dimensional dataset Total amount of data.
11. The analytical method according to claim 9, characterized in that, The formula for the propagation probability between the information layer asset nodes is: , in, This is current evidence of the attack. The following vulnerabilities The conditional probability of being exploited. It is a known vulnerability. Evidence of an attack was detected while the system was being exploited. The conditional probability of this distribution This is current evidence of the attack. The probability of being detected; It's a vulnerability. The probability of being exploited.
12. The analytical method according to claim 9, characterized in that, The maximum risk diffusion path The calculation formula is: , in, This represents the path of maximum risk propagation. 's' is an abnormal evidence node, defined as the starting point of the risk propagation path. 'd' is a critical asset point, defined as the target point of the cyberattack. 'p' is a feasible path from 's' to 'd'. It attacks the shortest path. It is the shortest path for the accident.
13. A cyber-physical risk propagation device, characterized in that, The device includes: Establish a module to build an asset assessment system and provide asset risk scores; The mining module is used to mine physical risk variables and information security vulnerabilities based on the asset assessment system. The association module is used to establish a "vulnerability-function" association model by utilizing the physical risk variables and information security vulnerabilities. The prediction module is used to predict the information-physical risk propagation path based on the "vulnerability-function" association model. The establishment of a "vulnerability-function" correlation model using the physical risk variables and information security vulnerabilities includes: Risk variables at the physical layer are categorized, and a variable correlation topology graph is constructed. The degree of association and propagation direction of each pair of variables in the variable association topology graph are solved based on the propagation entropy. Analyze security vulnerabilities at the information layer and establish an attack graph model. This generates redundant penetration paths; where S is the set of node attributes in the Bayesian attack graph, A is the set of atomic attacks in the attack graph, E represents the set of anomalous evidence, and T represents the relationship between child and parent nodes. This represents the prior probability of attacking a node's attribute in the graph. The exploitation and consequences of the aforementioned security vulnerabilities were analyzed, and a "vulnerability-function" correlation model was established to uncover the propagation path of redundant information and physical risks.
14. The apparatus according to claim 13, characterized in that, The asset assessment system includes: Personal injury, property damage, environmental pollution, and equipment malfunction; The casualties mentioned include operational compliance, personal protective measures, and emergency escape plans; The property losses include equipment costs, production slowdown rates, and social impacts. The environmental pollution includes the hazards of media leakage, the level of environmental risk control, and the ability to resist risks from the external environment. The equipment functions include equipment reliability design, mean time between failures (MTBF), and redundancy design.
Citation Information
Patent Citations
Global safety detection system for intelligent manufacturing production line and working method thereof
CN112904817A
Integrated risk attack tree generation method and device, electronic equipment and storage medium
CN114666115A
