A Hypergame-Based Computational Network Attack-Defense Game Model

By constructing a hypergame model, simplifying the computing power network into a three-level node, building a subgame set and calculating the payoff matrix of attack and defense strategies, the security problem of the computing power network under distributed attacks is solved, and effective defense strategy selection and network security improvement are achieved.

CN118821937BActive Publication Date: 2026-11-17SUN YAT SEN UNIV
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202410634794.0
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-05-22
Publication Date
2026-11-17
Estimated Expiration
2044-05-22

AI Technical Summary

Technical Problem

Due to their unique architecture and purpose, computing networks exhibit vulnerabilities different from traditional networks when facing distributed attacks. Existing network security defense strategies are unable to effectively deal with high-risk attacks on computing networks, and the game theory model assumes that the participants have complete information and the same cognition, leading to inaccurate decision-making.

Method used

A hypergame-based computing power network attack and defense adversarial game model is constructed. By simplifying it into a three-level node distributed network, a subgame set is constructed, the payoff matrix of attack and defense strategies is calculated, and a hypergame paradigm matrix is ​​constructed by combining perception parameters. Considering the incomplete information and limited cognition of the participants, the strategy that maximizes the expected utility is selected to take action.

Benefits of technology

It achieves effective modeling of the attack and defense game process in computing power networks, introduces defensive deception strategies, prolongs network crash time, reduces attacker gains, improves network security, and adapts to decision-making methods under incomplete information.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118821937B_ABST
    Figure CN118821937B_ABST
Patent Text Reader

Abstract

This invention provides a hypergame-based attack-defense adversarial game model for computing power networks, including simplifying the computing power network into a distributed network of three-level nodes; constructing a set of possible subgames; calculating the payoff matrix and constructing a hypergame paradigm matrix by combining perception parameters; calculating the expected hypergame utility of each strategy; updating the network node state according to the strategy choices of both participants; and repeating the above steps until the network system collapses. This invention models the attack-defense game process under computing power networks and incorporates defensive deception strategies into the set of defensive strategies. Under different initial perceptions of the game by attackers and defenders, it realizes multi-stage attack-defense game theory, verifying the effectiveness of the decision-making method and deception strategy based on the hypergame paradigm. This invention also considers mixed Nash equilibrium decision-making under complete information and hypergame decision-making methods under incomplete information, verifying the impact of participants' incomplete perception of the game on the game outcome and process.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of attack and defense game theory algorithms, and in particular to a computing power network attack and defense adversarial game model based on hypergame theory. Background Technology

[0002] With the continuous development of 5G, cloud computing, edge computing, and mobile internet technologies, and the increasing maturity of new network services and applications such as AR / VR and vehicle-to-everything (V2X), computing resources are showing a trend towards ubiquitous deployment. To more efficiently utilize massive distributed computing resources, meet urgent computing power demands, and promote the deep integration of edge computing and networks, computing power networks have received widespread attention. A computing power network is a distributed network structure encompassing the edge, device, and cloud, aiming to connect distributed computing nodes through dynamic integration and sharing of computing and network resources, forming a computing power-aware and allocable network system.

[0003] Compared to traditional cybersecurity threats, computing networks possess characteristics such as high generalization, cross-regional integration, computing-network integration, and flexible connectivity. This results in a wider security exposure surface and a higher frequency of data connections, significantly increasing the risk of resource attacks and posing new challenges to cybersecurity strategies. Due to its unique architecture and purpose, computing networks exhibit different vulnerabilities and maintenance challenges compared to traditional networks when facing distributed attacks and botnet attacks. Because they typically use simple processors and operating systems, edge devices in computing networks often cannot support complex security defense schemes, making them more vulnerable to becoming entry points and botnets for attacks. Furthermore, attacks on the cloud control plane can lead to the failure of the entire network.

[0004] For defense strategies in computing power networks, existing network security defense strategies mainly include firewalls, patch management, and node maintenance. In addition, deceptive defense strategies are also included. Deceptive defense strategies include honeypots and honey messages. A honeypot strategy involves deploying honeypot nodes in a given network topology to build a honeypot network and lure attackers to attack the honeypot nodes. Honeypot nodes are primarily responsible for identifying and interfering with attackers' actions; destroying a honeypot node does not affect the integrity of the network, and attackers cannot reach regular nodes from honeypot nodes. Honey messages involve spreading false information to interfere with attackers' actions.

[0005] Game theory is an effective tool for analyzing attacker-defender models, designed to help understand behavioral patterns and decision dynamics in situations involving the interaction of multiple offensive and defensive decision-makers. Existing game theory assumptions presuppose that both sides have the same perspective and possess identical and complete common knowledge and global understanding. However, in real-world conflicts, limited knowledge and incomplete information about the game may lead to decision-makers' inaccurate understanding of relevant game information. Furthermore, each decision-maker's subjective cognitive preferences regarding their game state will further influence game decisions and the final outcome, thus severely impacting the defensive outcome. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides a hypergame-based attack and defense adversarial game model for computing power networks. This invention constructs a multi-stage game-based attack and defense model for computing power networks based on computing power network nodes, realizes the attack and defense game process under a specific computing power network topology, and achieves effective decision-making at each game stage.

[0007] The technical solution of this invention is: a hypergame-based computing power network attack and defense adversarial game model, the construction of which includes the following steps:

[0008] S1) Design a computing power network architecture, simplifying the computing power network into a distributed network with three levels of nodes;

[0009] S2) Based on the participants' understanding of the offensive and defensive game, construct a set of possible subgames, and initialize the participants' beliefs about the occurrence of possible subgames and the uncertainty of the participants' current beliefs.

[0010] S3) Based on a specific computing power network state, calculate the payoff matrix for all attack and defense strategies in the current possible subgame, and construct the hypergame paradigm matrix by combining the perception parameters;

[0011] S4) Based on the hypergame paradigm matrix, the hypergame expected utility of each strategy is calculated by comprehensively considering the participants' expected strategic payoffs and the worst possible outcome.

[0012] S5) Participants choose strategies that maximize expected utility, and update the network node states and participant uncertainties based on the strategy choices of both participants.

[0013] S6) In each stage of the game, participants choose strategies that maximize the expected utility function and act accordingly, affecting the state of the computing network until the computing network system collapses.

[0014] Preferably, in step S1), the distributed network of the three-level nodes is a distributed network with three-level node hierarchical interconnection from edge to end to cloud, and the defense strategy is uniformly deployed in the cloud, and fake nodes are introduced to realize the honeypot network.

[0015] Preferably, in step S2), the subgame set includes four possible subgames, namely:

[0016] Subgame 1: The defender will not use a deceptive strategy set;

[0017] Subgame 2: The defender uses a deceptive strategy – a honeypot;

[0018] Subgame 3: The defender uses a deceptive strategy – the honey information;

[0019] Subgame 4: The defender uses a deception strategy – honeypot, honey information.

[0020] Preferably, in step S2), the deception strategy used by the defender is a deceptive defense strategy that includes honeypot network and honey information.

[0021] Preferably, in step S3), the attack and defense benefit matrix under a specific attack strategy and defense strategy is calculated as follows:

[0022] S31) Define the system's loss function:

[0023]

[0024] In the formula, AV represents the minimum computing power required for the defined computing network, and power... i The initial computing power of node i is av. i nd represents the available state of a node. i val represents the degree of damage to node i. i p2 is the node importance parameter, p2 is the acceptable network damage ratio, and max is defined as a maximum value as the penalty for network unavailability caused by system damage.

[0025] S32) Profit function: The model uses the overall damage level of the network to assess the impact of the strategy on the computing network, and represents the profit brought by the player to exercise each strategy;

[0026] impact(D i S j A q ) = SF * -SF;

[0027] In the formula, D i Conventional defense strategies i, S adopted by the defender j Deception strategy j, A adopted by the defender q The attack strategy q adopted by the attacker, SF * SF indicates the degree of damage to the system after the policy is executed, while SF indicates the degree of damage to the system before the policy is executed.

[0028] S33) Define the defender's utility function:

[0029] U D (D i S j A q ) = -cost D (D i S j )+impact(D i S j A q )*μ+aff(t)*θ;

[0030] In the formula, D i Conventional defense strategies i, S adopted by the defender j Deception strategy j, A adopted by the defender q The attack strategy adopted by the attacker, q, impact(D) i S j A q ) represents the impact of the strategy combination on the structural integrity of the system, μ is the impact weight, aff(t) is the long-term payoff of the deception strategy, and θ is the long-term payoff weight;

[0031] S34) Define the attacker's utility function

[0032]

[0033] Where D i Conventional defense strategies i, S adopted by the defender j Deception strategy j, A adopted by the defender q The attack strategy adopted by the attacker, q, impact(D) i S j A q This indicates the impact of strategy combinations on the structural integrity of the system. Here, aff(t) represents the long-term payoff of the deception strategy, and v represents the long-term payoff weight.

[0034] S35) Define the cost function for implementing the defense strategy.

[0035]

[0036] In the formula, when S is taken j When S1 is reached, the defender uses the deceptive defense strategy S1 for the first time. The cost of the deceptive strategy is defined based on the designed number of honeypot nodes. The HoneyNode set represents the set of honeypot nodes, and stage represents the game stage, i.e., the cost of strategy S1 decreases as the game progresses; for S... j =S2, the honey information only applies to the current stage of the game, defined as follows: Where c Di c s1 c s2 All of these are derived from prior public knowledge;

[0037] S36) For attackers, define the attacker's chosen attack strategy A. q , strategy A q The policy is applied to the set of nodes Anode, c Aq Based on prior public knowledge, the cost function for executing the attack strategy is defined as:

[0038]

[0039] Preferably, in step S4), the expression for the expected utility of the hypergame is as follows:

[0040] HEU(R x )=(1-g D )×EU(R x A all )+g D ×EU worst (R x A all );

[0041] HEU(A q )=(1-g A )×EU(R all A q )+g A ×EU worst (R all A q );

[0042] In the formula, R x =(D i S j To treat the classical defense strategy and the defensive deception strategy as a single strategy combination, the strategy space of both the attacker and defender is confined to a two-dimensional plane, g. D EU(R) represents the uncertainty perceived by participants in the current game. x A all ) represents the strategy combination R x The expected utility, EU(R) all A q ) represents attack strategy A q Expected utility, g AEU represents the attacker's uncertainty about the game. worst (R x A all ) indicates that the participant adopts the worst-case scenario from the set of strategies.

[0043] Preferably, in step S4), the hypergame expected utility is used to evaluate the expected utility of the row player when choosing strategy i, given that the column player chooses one strategy from all available strategies:

[0044]

[0045] EU worst (R x A all )=n*min{U D (R x A q ), A q ∈A set}*w min ;

[0046]

[0047] EU worst (R all A q )=m*min{U A (R x A q ), R x ∈R set}*w min ;

[0048] In the formula, w q U represents the probability that attack strategy q is selected. D (R x A q ) represents the strategy combination U D (R x A q The defensive utility of ); w min w represents the probability that the worst-case strategy will be chosen. x For the defensive strategy combination R x The probability of being selected, U A (R x A q ) represents the strategy combination U D (R x A q The attacker's utility; n and m are the number of defense strategies and the number of attack strategies, respectively.

[0049] Preferably, in step S4), based on the hypergame paradigm, the probability expression for a row participant choosing strategy j for a column participant is as follows:

[0050]

[0051] In the formula, P K Let col be the probability that the participants believe subgame k will proceed. kj Let w be the probability that a row participant believes a column participant will adopt strategy j in subgame k. j In a hypergame, let the probability that a row participant believes a column participant will adopt strategy j be given.

[0052] Based on the hypergame definition of participants' perception of possible subgames, for participants and column participants in the strategy payoff matrix, the probability that a column participant will use strategy i in subgame k is defined in the perception of a row participant as follows:

[0053]

[0054] In the formula, CMS k In a subgame k, the row participant's preference vector for the column participant's strategy selection;

[0055] Define the preference vector of the row participants in subgame k as:

[0056]

[0057] In the formula, RMS k In subgame k, row represents the preference vector for the participants choosing their strategies. ki Let be the probability that a participant chooses strategy i in subgame k;

[0058] Define the row participant's belief about the hypergame, that is, the probability that the column participant chooses to play subgame j, as follows:

[0059]

[0060] In the formula, P k Let $k$ be the probability that the participants believe subgame k will proceed.

[0061] Preferably, in step S5), the defender's strategy and the attacker's strategy will directly lead to changes in the damaged state of the node, where the changes in the state of the computing network node are as follows:

[0062]

[0063] Among them, nd i Indicates the degree of damage to the current node i, e(D) i ,S j Aq ) represents the defensive strategy combination (D i ,S j and attack strategy A q The impact on the degree of node damage This represents the degree of damage to node i after the update.

[0064] The beneficial effects of this invention are as follows:

[0065] 1. This invention realizes the modeling of the attack and defense game process under computing power network, and incorporates defensive deception strategy into the set of defense strategies. Under the different initial perceptions of the game by the attacker and the defender, a multi-stage attack and defense game is realized, and the effectiveness of the decision-making method and deception strategy based on the hypergame paradigm is verified.

[0066] 2. This invention constructs a computing power network model based on a specific network topology that can be used for game reasoning. It considers hybrid Nash equilibrium decision-making under complete information and hypergame decision-making methods under incomplete information, and verifies the impact of participants' incomplete perception of the game on the game results and process. Attached Figure Description

[0067] Figure 1 This is a flowchart illustrating the method of an embodiment of the present invention;

[0068] Figure 2 This is an example diagram of a small-scale three-level computing network topology including honeypot nodes in an embodiment of the present invention;

[0069] Figure 3 This is a diagram showing the participant utility and system integrity of a complete information game without deception in an embodiment of the present invention.

[0070] Figure 4 This is a diagram showing the participant utility and system integrity of a complete game under a deception strategy in an embodiment of the present invention.

[0071] Figure 5 This is a diagram showing the participant utility and system integrity of a hypergame under a deception strategy in an embodiment of the present invention.

[0072] Figure 6 This is a graph showing participant utility and system integrity under different initial beliefs of the attacker in this embodiment of the invention;

[0073] Figure 7 This is a graph showing the participant utility and network integrity of different deception strategies in a game of complete information according to an embodiment of the present invention.

[0074] Figure 8 This is a graph showing the participant utility and system integrity under different uncertainties of the attacker in a hypergame according to an embodiment of the present invention;

[0075] Figure 9This is a graph showing the impact of the attacker's subgame uncertainty on the attack cost in a hypergame according to an embodiment of the present invention.

[0076] Figure 10 This is a model architecture diagram constructed for an embodiment of the present invention. Detailed Implementation

[0077] The specific embodiments of the present invention will be further described below with reference to the accompanying drawings:

[0078] like Figure 1 , Figure 10 As shown, this embodiment provides a hypergame-based computing power network attack and defense adversarial game model, including the following steps:

[0079] S1) Design a computing power network architecture, simplifying the computing power network into a three-level distributed network; wherein, in this embodiment, the three-level distributed network is a distributed network with three-level node hierarchical interconnection from edge to terminal to cloud, such as... Figure 2 As shown.

[0080] Attackers can implement a set of advanced persistent attack strategies on a computing network, and the characteristics of each attack strategy can be modeled as follows: 1) Attack cost, representing the cost required for the attacker to adopt the strategy; 2) Expected attack utility, representing the expected utility of the strategy on the node; 3) Attack availability conditions, which evaluate the availability of the attack strategy on the node based on the node state and the probability of strategy success, as shown in Table 1:

[0081] Table 1. Set of attacker strategies under computing power network attack and defense tasks

[0082]

[0083] Defenders can deploy different types of defense strategies to counter attackers. This embodiment considers three classic defense strategies and two popular defensive deception strategies. The classic defense strategies include firewalls, network management, and node repair. The two defensive deception strategies include honeypots and honey messages. The characteristics of each defense strategy can be modeled as follows: 1) Defense cost, representing the cost required for the defender to implement the strategy; 2) Expected defense utility, representing the expected utility of the strategy on the node; 3) Availability conditions, evaluating the availability of the strategy on the node based on node status and the probability of strategy success, as shown in Table 2.

[0084] Table 2. Set of feasible strategies for defenders

[0085]

[0086] S2) Based on the participants' understanding of the offensive and defensive game, a set of possible subgames is constructed, and the participants' beliefs about the occurrence of possible subgames and their uncertainty about their current beliefs are initialized, as shown in Table 4. In this embodiment, as shown in Table 3, the set of possible subgames includes four possible subgames, namely:

[0087] Subgame 1: The defender will not use a deceptive strategy set;

[0088] Subgame 2: The defender uses a deceptive strategy – a honeypot;

[0089] Subgame 3: The defender uses a deceptive strategy – the honey information;

[0090] Subgame 4: The defender uses a deception strategy – honeypot, honey information.

[0091] In this embodiment, the deception strategy used by the defender is a deceptive defense strategy that includes honeypot network and honey information. For example... Figure 2 As shown, when a honeypot strategy is selected, honeypot nodes are activated, which alters the network topology and lures attackers to target them. To further protect legitimate nodes, only regular nodes are allowed to reach honeypot nodes, and all honeypot nodes are fully connected, but attackers cannot reach regular nodes from honeypot nodes. The honey information strategy can lure attackers by spreading false information, such as honey tags and fake patches. The data in these files and databases is designed to be attractive to attackers, but in reality, it is of no practical use to help attackers understand the true state of the computing network.

[0092] Table 3 shows the four possible subgames.

[0093]

[0094] Table 4 Examples of participant beliefs and uncertainty initialization in hypergame paradigms

[0095]

[0096]

[0097] S3) Based on the specific computing power network state, calculate the payoff matrix for all attack and defense strategies in the current possible subgame, and construct the hypergame paradigm matrix by combining the perception parameters; the hypergame paradigm constructed in this embodiment is shown in Table 13:

[0098] Table 13 Game Theory Paradigm Matrix

[0099]

[0100] The attack and defense payoff matrix under specific attack and defense strategies is calculated as follows:

[0101] S31) Define the system's loss function:

[0102]

[0103] In the formula, AV represents the minimum computing power required for the defined computing network, and power... i The initial computing power of node i, ay i val represents the available state of a node. i Here, p1 is the node importance parameter, p2 is the acceptable proportion of network damage, and max is defined as a maximum value as the penalty for network unavailability caused by system damage; nd i The degree of damage to node i, with a value between 0 and 1; nd i The larger the value, the higher the degree of node damage, as shown in the formula below:

[0104]

[0105] S32) Profit function: The model uses the overall damage level of the network to assess the impact of the strategy on the computing network, and represents the profit brought by the player to exercise each strategy;

[0106] impact(D i S j A q ) = SF * -SF;

[0107] In the formula, D i Conventional defense strategies i, S adopted by the defender j Deception strategy j, A adopted by the defender q The attack strategy q adopted by the attacker, SF * SF indicates the degree of damage to the system after the policy is executed, while SF indicates the degree of damage to the system before the policy is executed.

[0108] S33) Define the defender's utility function, which aims to minimize network operating costs;

[0109] U D (D i S j A q )=-costD(D i S j )+impact(D i S j A q )*μ+aff(t)*θ;

[0110] In the formula, D i Conventional defense strategies i, S adopted by the defenderj Deception strategy j, A adopted by the defender q The attack strategy adopted by the attacker, q, impact(D) i S j A q ) represents the impact of the strategy combination on the structural integrity of the system, μ is the impact weight, aff(t) is the long-term payoff of the deception strategy, and θ is the long-term payoff weight;

[0111] S34) Define the attacker's utility function, which aims to damage computing network nodes as much as possible while considering the cost of the attack.

[0112]

[0113] Where D i Conventional defense strategies i, S adopted by the defender j Deception strategy j, A adopted by the defender q The attack strategy adopted by the attacker, q, impact(D) i S j A q This indicates the impact of strategy combinations on the structural integrity of the system. Here, aff(t) represents the long-term payoff of the deception strategy, and v represents the long-term payoff weight.

[0114] S35) Define the cost function for implementing the defense strategy. For the defender, this embodiment defines the defender's basic strategy D. i Deception Strategy S j The strategy is then applied to the set of nodes, Node. Therefore, the cost function is:

[0115]

[0116] In the formula, when S is taken j When S1 is reached, the defender uses the deceptive defense strategy S1 for the first time. The cost of the deceptive strategy is defined based on the designed number of honeypot nodes. The HoneyNode set represents the set of honeypot nodes, and stage represents the game stage, i.e., the cost of strategy S1 decreases as the game progresses; for S... j =S2, the honey information only applies to the current stage of the game, defined as follows: Where c Di c s1 c s2 All of these are derived from prior public knowledge;

[0117] S36) For attackers, this embodiment defines the attacker's chosen attack strategy A. q , strategy A qThe policy is applied to the set of nodes Anode, c Aq Based on prior public knowledge, the cost function for executing the attack strategy is defined as:

[0118]

[0119] S4) Based on the hypergame paradigm matrix, the hypergame expected utility of each strategy is calculated by comprehensively considering the participants' expected strategic payoffs and the worst possible outcome.

[0120] In this embodiment, the expression for the expected utility of the hypergame is as follows:

[0121] HEU(R x )=(1-g D )×EU(R x A all )+g D ×EU worst (R x A all );

[0122] HEU(A q )=(1-g A )×EU(R all A q )+g A ×EU worst (R all A q );

[0123] In the formula, R x =(D i S j To treat the classical defense strategy and the defensive deception strategy as a single strategy combination, the strategy space of both the attacker and defender is confined to a two-dimensional plane, g. D EU(R) represents the uncertainty perceived by participants in the current game. x A all ) represents the strategy combination R x The expected utility, EU(R) all A q ) represents attack strategy A q Expected utility, g A EU represents the attacker's uncertainty about the game. worst (R x A all ) indicates that the participant adopts the worst-case scenario from the set of strategies.

[0124] The hypergame expected utility is used to evaluate the expected utility of a row player when choosing strategy i, given that the column player chooses one strategy from all available strategies:

[0125]

[0126] EU worst (R x A all )=n*min{U D (R x A q ), A q ∈A set}*w min ;

[0127]

[0128] EU worst (R all A q )=m*min{U A (R x A q ), R x ∈R set}*w min ;

[0129] In the formula, w q U represents the probability that attack strategy q is selected. D (R x A q ) represents the strategy combination U D (R x A q The defensive utility of ); w min w represents the probability that the worst-case strategy will be chosen. x For the defensive strategy combination R x The probability of being selected, U A (R x A q ) represents the strategy combination U D (R x A q The attacker's utility; n and m are the number of defense strategies and the number of attack strategies, respectively.

[0130] Based on the hypergame paradigm, the probability expression for a row participant choosing strategy j for a column participant is as follows:

[0131]

[0132] In the formula, P K Let col be the probability that the participants believe subgame k will proceed. kj Let w be the probability that a row participant believes a column participant will adopt strategy j in subgame k. j In a hypergame, let be the probability that a row participant believes a column participant will adopt a certain strategy.

[0133] Based on the hypergame definition of participants' perception of possible subgames, for participants and column participants in the strategy payoff matrix, the probability that a column participant will use strategy i in subgame k is defined in the perception of a row participant as follows:

[0134]

[0135] In the formula, CMS k In a subgame k, the row participant's preference vector for the column participant's strategy selection;

[0136] Define the preference vector of the row participants in subgame k as:

[0137]

[0138] In the formula, RMS k In subgame k, row represents the preference vector for the participants choosing their strategies. ki Let be the probability that a participant chooses strategy i in subgame k;

[0139] Define the row participant's belief about the hypergame, that is, the probability that the column participant chooses to play subgame j, as follows:

[0140]

[0141] In the formula, P k Let $k$ be the probability that the participants believe subgame k will proceed.

[0142] S5) Participants choose strategies that maximize expected utility, and update the network node states and participant uncertainties based on the strategy choices of both participants.

[0143] In this embodiment, the defender's strategy and the attacker's strategy will directly lead to changes in the damaged state of the nodes, and the changes in the state of the computing network nodes are as follows:

[0144]

[0145] Among them, nd i Indicates the degree of damage to the current node i, e(D) i ,S j A q ) represents the defensive strategy combination (D i ,S j and attack strategy A q The impact on the degree of node damage This represents the degree of damage to node i after the update.

[0146] S6) In each stage of the game, participants choose strategies that maximize the expected utility function and act accordingly, affecting the state of the computing network until the computing network system collapses.

[0147] This embodiment uses Nash equilibrium to realize strategy decision-making in a game of complete information. Based on the strategy payoff matrix, it finds the optimal countermeasure of each player against the strategy combination of the other players. When the strategy adopted by any party is the best countermeasure against the strategy combination of all other parties, it is the Nash equilibrium solution, as shown in Tables 5, 6, and 7. Considering the diversity of Nash solutions in computing power network attack and defense games, it combines pure Nash equilibrium and mixed Nash equilibrium to find the Nash equilibrium solution of the game.

[0148] Table 5. Finding Nash Equilibrium Solutions through Payoff Matrix in the Initial Stage of a Complete Information Game

[0149]

[0150]

[0151] Table 6. Finding Nash Equilibrium Solutions via Payoff Matrix in the Mid-Stage of Complete Information Games

[0152]

[0153]

[0154] Table 7. Finding Nash Equilibrium Solutions via Payoff Matrix in the Later Stages of Complete Information Games

[0155]

[0156]

[0157]

[0158] This embodiment considers an attacker-defender game that is a repeated game involving both complete and incomplete information, where each game is a multi-stage game. It is assumed that both players have their own understanding of the ongoing subgame. If we define the subgame perceived by both players as the original complete information game, then the current game is a complete information game. The model allows players to repeatedly play within a given computing power network structure. In each game, the attacker and defender take multiple actions, impacting the network structure in real time.

[0159] As shown in Tables 5, 6, and 7, gray squares represent strictly dominated strategies, and blue squares represent target Nash equilibrium strategies. Analysis of the payoff matrix and Nash equilibrium strategies reveals that, due to the inability to use botnet-based strategies in the early stages of an attack under the current computing power network conditions, some attack strategies degenerate into inaction, leading to multiple Nash equilibrium solutions. At different stages of the game, as the degree of network damage increases, the choice of attack and defense strategies changes. Attackers evolve from monitoring the network to conducting distributed attacks or botnet attacks based on already captured nodes. Long-term gains encourage defenders to use honeypots in deceptive defense strategies in the early stages. Simultaneously, defenders select honey information in the later stages. Observation of the payoff matrix shows that deceptive strategies can effectively reduce the attacker's expected gains and bring higher gains to the defender.

[0160] Furthermore, this embodiment also verifies the effectiveness of the hypergame decision-making method and deception model through simulation experiments. The solution analysis of the hypergame paradigm under different subgame beliefs of the attacker is shown in Tables 8 and 9. If the participant has some awareness of the other ongoing subgames, modifying the uncertainty will affect the participant's expectations of the game. As the participant's uncertainty increases, the participant's expectations of the game decrease, which may change the strategy selection. As shown in Table 2, when the attacker is completely certain that the defender chooses to engage in all four subgames with equal probability, the attacker will choose strategy 5; as the attacker's uncertainty about the ongoing game increases, the attacker may choose the more conservative strategy 2.

[0161] Table 8 shows the initial hypergame utility calculated based on the hypergame paradigm, including four subgames.

[0162]

[0163]

[0164]

[0165] Table 9 shows the initial stage of a hypergame containing two subgames, calculated based on the hypergame paradigm to determine the expected utility of the hypergame.

[0166]

[0167]

[0168] like Figure 3-5As shown in Table 10, after repeating the experiment 30 times and comparing the complete game without deception with the complete game including deception, it was found that introducing deception strategies into the defender's strategy set effectively prolongs the network crash time and reduces the attacker's payoff. At the end of the sixth round, the attacker's cumulative average payoff without deception was greater than 600, while the cumulative average attacker's payoff with deception was less than 200. Introducing deception strategies extended the network duration by 27%. Although honeypots can confuse attackers to some extent, their deception effect on attackers is relatively limited due to the limited number of nodes added. Experimental results show that even a limited number of deception strategies are sufficient to improve the model's round duration.

[0169] Table 10 System collapse times for no-deception and deception strategies in complete and hypergame scenarios.

[0170]

[0171] By comparing complete information games and hypergames that incorporate deception strategies, attackers can, to some extent, perceive the defender's deception strategies by comprehensively considering multiple subgames. The results show that even though attackers can predict and clearly recognize that the defender might employ a specific deception strategy, this strategy can still effectively prolong the time the network structure remains intact.

[0172] like Figure 6 As shown in Table 11, using the same decision-making algorithm, the network exhibits higher integrity when the attacker is unaware of the deception strategy, indicating that the deception strategy effectively protects the integrity of the computing network structure. Comparing parameters such as game duration, cumulative utility of participants, and network integrity, it is shown that the introduction of the deception strategy successfully interferes with the attacker's attack strategy and successfully extends the network's uptime for the defender. Even if the defender may be aware of or fully believe in the existence of the deception strategy, the deception strategy is still effective. As the attacker's belief in the occurrence of the deception strategy increases, the network crash time decreases, and the attacker's attack efficiency increases. Experimental simulations demonstrate that in the hypergame process, by reasonably modeling the possible existence of subgames, participants can achieve higher payoffs, helping them achieve their goals faster in games with incomplete information.

[0173] Table 11 Duration of Systemic Game under Different Initial Beliefs (Attacks)

[0174]

[0175] like Figure 7As shown in Table 12, the two deception strategies are compared under a complete game scenario. Under the conditions of a complete information game, the deception strategy can effectively prolong network crash time and maintain the security of the computing power network. This simulation verifies the effectiveness of the deception strategy under a complete game scenario. If the use of deception information and the deployment of a small number of honeypots are considered in the network defense strategy, the defender is likely to choose to execute honeypot operations in the early stages of the game. As the game progresses, when the attacker begins to adjust its strategy to cope with the initial honeypot inducement, the defender can use the deception information to have a more direct and profound impact on the attacker.

[0176] Table 12 shows the game duration for different deception strategies in a game of complete information.

[0177]

[0178] like Figure 8 , 9 As shown, participants' initial understanding and cognition of the game largely determine the overall outcome and development trend of the game, emphasizing the importance of correctly assessing the opponent's strategy. a When g = 0, the attacker fully trusts the ongoing game and makes decisions based on their understanding of the defender and the subgames. a When the uncertainty factor is 1, the attacker completely distrusts their own understanding and makes decisions solely based on fear of the worst-case scenario. In this simulation experiment, the attacker exhibits greater aggression under worse expectations and is more likely to choose more destructive strategies, but this comes at a higher cost. Therefore, when using hypergame decision-making, the level of uncertainty should be reasonably controlled to achieve a balance between gains and costs.

[0179] The embodiments and descriptions above are merely illustrative of the principles and preferred embodiments of the present invention. Various changes and modifications may be made to the present invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed.

Claims

1. A hypergame-based computing power network attack-defense adversarial game model, characterized in that, The construction of the model includes the following steps: S1) Design a computing power network architecture, simplifying the computing power network into a distributed network with three levels of nodes; S2) Based on the participants' understanding of the offensive and defensive game, construct a set of possible subgames, and initialize the participants' beliefs about the occurrence of possible subgames and the uncertainty of the participants' current beliefs. S3) Based on a specific computing power network state, calculate the payoff matrix for all attack and defense strategies in the current possible subgame, and construct the hypergame paradigm matrix by combining the perception parameters; S4) Based on the hypergame paradigm matrix, the hypergame expected utility of each strategy is calculated by comprehensively considering the participants' expected strategic payoffs and the worst possible outcome. S5) Participants choose strategies that maximize expected utility, and update the network node states and participant uncertainties based on the strategy choices of both participants. S6) In each stage of the game, participants choose strategies that maximize the expected utility function and act accordingly, affecting the state of the computing network until the computing network system collapses.

2. The computing power network attack and defense adversarial game model based on hypergame as described in claim 1, characterized in that: In step S1), the distributed network of the three-level nodes is a distributed network with three-level node hierarchical interconnection from edge to end to cloud, and the defense strategy is uniformly deployed in the cloud, and fake nodes are introduced to realize the honeypot network.

3. The computing power network attack and defense adversarial game model based on hypergame as described in claim 1, characterized in that: In step S2), the subgame set includes four possible subgames, namely: Subgame 1: The defender will not use a deceptive strategy set; Subgame 2: The defender uses a deceptive strategy – a honeypot; Subgame 3: The defender uses a deceptive strategy – the honey information; Subgame 4: The defender uses a deceptive strategy – honeypot, honey information; The deception strategy used by the defender is a deceptive defense strategy that includes honeypot network and honey information.

4. The computing power network attack and defense adversarial game model based on hypergame as described in claim 1, characterized in that: In step S3), the attack and defense benefit matrix under specific attack and defense strategies is calculated as follows: S31) Define the system's loss function: ; In the formula, AV represents the minimum computing power required for the defined computing network. Initialize the computing power for node i. This indicates the available status of the node. Let i be the degree of damage to node i. p2 is the node importance parameter, p2 is the acceptable network damage ratio, and max is defined as a maximum value as the penalty for network unavailability caused by system damage. S32) Define the profit function: The model uses the overall damage level of the network to evaluate the impact of the strategy on the computing network, and represents the profit brought by the player to exercise each strategy; ; In the formula, The conventional defense strategy adopted by the defender, i Deception strategy j adopted by the defender The attack strategy q adopted by the attacker. This indicates the degree of damage to the system after the policy is implemented. Indicates the extent of system damage before the strategy is executed; S33) Define the defender's utility function: ; In the formula, The conventional defense strategy adopted by the defender, i Deception strategy j adopted by the defender The attack strategy q adopted by the attacker. This indicates the impact of strategy combinations on the structural integrity of the system. To influence the weight, For the long-term benefits of deception strategies, Weighted for long-term returns; S34) Define the attacker's utility function ; in The conventional defense strategy adopted by the defender, i Deception strategy j adopted by the defender The attack strategy q adopted by the attacker. This indicates the impact of strategy combinations on the structural integrity of the system. For the corresponding weights, For the long-term benefits of deception strategies, Weighted for long-term returns; S35) Define the cost function for implementing the defense strategy: ; In the formula, when taking At that time, the defender used a deceptive defense strategy for the first time. The cost of the deception strategy is defined based on the number of honeypot nodes designed. The HoneyNode set represents the set of honeypot nodes, and the stage represents the game stage, i.e., the strategy. The cost decreases as the game progresses; for The honey information only applies to the current stage of the game, and is defined as follows: ;in , , All of these are derived from prior public knowledge; S36) Define the cost function: For the attacker, define the attacker's chosen attack strategy. , strategy It is applied to the set of nodes Anode corresponding to the strategy. Based on prior public knowledge, the cost function for executing an attack strategy is defined as: 。 5. The computing power network attack and defense adversarial game model based on hypergame as described in claim 4, characterized in that: In step S4), the expression for the expected utility of the hypergame is as follows: ; ; In the formula, To treat classical defense strategies and defensive deception strategies as a single strategy combination, the strategy space for both attackers and defenders is confined to a two-dimensional plane. This indicates the defender's uncertainty about the current game. Indicates strategy combination Expected utility Indicate attack strategy Expected utility This indicates the attacker's uncertainty about the current game. This indicates that the participant adopts the worst-case scenario from the set of strategies.

6. The computing power network attack and defense adversarial game model based on hypergame as described in claim 5, characterized in that: In step S4), the hypergame expected utility is used to evaluate the expected utility of the row player when choosing strategy i, given that the column player chooses one strategy from all available strategies: ; ; ; ; In the formula, Let q be the probability that attack strategy q is selected. For defensive strategy combination attack strategy The defensive utility of a combination; The probability that the worst strategy is chosen; For defensive strategy combination The probability of being selected For defensive strategy combination attack strategy Attacker utility under combination; , These represent the number of defense strategies and the number of attack strategies, respectively.

7. The computing power network attack and defense adversarial game model based on hypergame as described in claim 6, characterized in that: In step S4), based on the hypergame paradigm, the probability expression for the row participant choosing strategy j for the column participant is as follows: ; In the formula, Let be the probability that the participants believe subgame k will proceed. Let $\mathbf{k}$ be the probability that a row participant believes a column participant will adopt strategy $j$ in subgame $k$. Let be the probability that a row participant believes a column participant will adopt strategy j in a hypergame.

8. The computing power network attack and defense adversarial game model based on hypergame as described in claim 7, characterized in that: Based on the hypergame definition of participants' perception of possible subgames, for participants and column participants in the strategy payoff matrix, the probability that a column participant will use strategy i in subgame k is defined in the perception of a row participant as follows: ; In the formula, Let be the preference vector of the row participants in subgame k regarding the strategy choices of the column participants; Define the preference vector of the row participants in subgame k as: ; In the formula, Let be the preference vector of the row participants in subgame k when choosing their strategies. Let be the probability that a participant chooses strategy i in subgame k.

9. A computing power network attack and defense adversarial game model based on hypergame as described in claim 8, characterized in that: Define the row participant's belief about the hypergame, that is, the probability that the column participant chooses to play subgame j, as follows: ; In the formula, Let be the probability that the participants believe subgame k will proceed.

10. The computing power network attack and defense adversarial game model based on hypergame as described in claim 1, characterized in that: In step S5), the defender's strategy and the attacker's strategy will directly lead to changes in the damaged state of the nodes, and the changes in the state of the computing network nodes are as follows: ; in, This indicates the degree of damage to the current node i. Indicates a combination of defensive strategies and attack strategies The impact on the degree of node damage This represents the degree of damage to node i after the update.

Citation Information

Patent Citations

  • Power distribution network defense strategy design method based on dynamic game under network attack

    CN115348064A

  • Network spoofing defense strategy optimization method and system based on intelligent real-time game

    CN117220995A