A collaborative confrontation method based on TUU game

Abstract

The present application relates to a kind of multi-agent cooperative confrontation methods based on TUU game, belong to unmanned cluster system field.The present application method includes: determining detection alliance and / or confrontation alliance based on cooperative confrontation task;In a task iteration process: based on the link state of each alliance current judging each alliance foundation data;Based on the link state of each alliance current using the data interaction method of TUU game to stimulate the cooperation of Agent node in alliance to carry out data interaction;Based on cooperative detection efficiency adjustment detection alliance data interaction strategy and autonomous detection strategy;Based on cooperative confrontation efficiency adjustment confrontation alliance data interaction strategy, formation mode and autonomous confrontation strategy;Based on the network overall efficiency of the alliance adjustment alliance.The present application method can guarantee the reliable interaction of data in the process of multi-agent execution task, and can adjust the strategy of detection alliance and confrontation alliance in multi-agent to execute task based on detection efficiency and confrontation efficiency in time, improve task execution efficiency.

Description

Technical Field

[0001] This invention belongs to the field of unmanned swarm systems, and particularly relates to a cooperative adversarial method based on TUU game theory. Background Technology

[0002] Intelligent unmanned swarm systems are overall systems composed of several unmanned systems that collaborate to complete complex tasks within a certain time and space according to task assignments. In the military field, unmanned equipment has characteristics such as low cost, miniaturization, single function, and flexible networking. Autonomous collaborative combat by unmanned swarm systems with swarm intelligence as the core technology is an important pattern of future warfare.

[0003] In a network environment, a multi-dimensional heterogeneous unmanned swarm system composed of heterogeneous and intelligent systems (agents) that interact with each other through information exchange is called a multi-agent system, or agent system. During multi-agent task execution, regional agent disconnection due to electromagnetic interference, severe fragmentation of the communication network caused by sparse agent clusters, and data link broadcast storms caused by a large number of agents can severely affect the quality of information exchange in the communication network, making it difficult to guarantee reliable data exchange during multi-agent task execution. However, existing cooperative detection and adversarial methods mainly assume autonomous multi-agent interaction under stable data links and unrestricted communication bandwidth. When the number of agents suddenly expands, or when agents are destroyed on a large scale in an adversarial environment, the above-mentioned multi-agent information exchange methods cannot obtain ideal solutions. Under sparse / dense data links, messages sent by agents may not be successfully transmitted, or the data transmission delay may be too large, resulting in a high message loss rate. In particular, many critical messages may be dropped due to weak network communication capabilities, which greatly reduces the speed of task execution. On the other hand, existing collaborative detection and countermeasure methods lack accurate real-time assessments of detection and countermeasure effectiveness, as well as network effectiveness, which fails to provide strong data support for adjusting alliance strategies. Summary of the Invention

[0004] Based on the above analysis, this invention aims to provide a collaborative adversarial method based on TUU game theory, which ensures the integrity of the data link and the high message delivery rate during the execution of tasks by multiple agents, and provides real-time feedback on the detection effectiveness, adversarial effectiveness and network effectiveness evaluation results during the execution of tasks to provide strong support for adjusting strategies, thereby further improving task execution efficiency.

[0005] The method of the present invention specifically includes the following: determining a detection alliance and / or an adversarial alliance based on a cooperative adversarial task, wherein the alliance changes dynamically during the iteration of the task; wherein, during one task iteration, the method includes: determining the current link status of each alliance based on the basic data of each alliance;

[0006] Based on the current link status of each alliance, the TUU game data interaction method is used to incentivize the cooperation of Agent nodes in the alliance to conduct data interaction.

[0007] Adjust the data interaction strategy and autonomous detection strategy of the detection alliance based on the collaborative detection effectiveness; adjust the data interaction strategy, formation mode and autonomous confrontation strategy of the confrontation alliance based on the collaborative confrontation effectiveness;

[0008] Based on the overall network performance of the alliance, the alliance may be adjusted to proceed to the next task iteration or terminate the task.

[0009] Furthermore, the DS evidence theory is used to determine the link status of each alliance based on the basic data of each alliance, including:

[0010] The communication network state identification framework for each alliance is defined as Θ = (strong communication capability S, weak communication capability D);

[0011] The number of pieces of evidence for the DS theory was determined based on the basic data of each alliance.

[0012] Based on the number of evidences, the uncertainty calculation formula and decision criteria are used to determine the status of each alliance link as either strong communication capability (S) or weak communication capability (D).

[0013] Furthermore, the data interaction method based on the TUU game theory of each alliance link state to incentivize cooperation among agent nodes in the alliance includes:

[0014] Alliances with weak communication capability (D) in the link interruption state are classified as sparse networks, while alliances with strong communication capability (S) in the link interruption state are classified as dense networks.

[0015] For alliances that are determined to be sparse networks, in each round of TUU game, the agent nodes in the alliance are incentivized to cooperate based on the revenue of each agent node and the average connection rate.

[0016] For alliances that are determined to be dense networks, in each round of the TUU game, agent nodes are set to incentivize cooperation among agent nodes in the alliance based on the utility value of each agent node in the alliance.

[0017] Furthermore, the cooperative combat effectiveness includes formation effectiveness and combat effectiveness;

[0018] The formation effectiveness includes formation assembly effectiveness, formation maintenance effectiveness, formation evasion effectiveness, and overall dynamic formation effectiveness;

[0019] The adversarial effectiveness includes soft adversarial effectiveness, hard adversarial effectiveness, multi-agent survivability, and multi-agent collaborative adversarial dynamic effectiveness.

[0020] Furthermore, the calculation method for the multi-agent collaborative adversarial evaluation index is as follows:

[0021] P DJ =λ rss P A +λ gj P B +λ sc P p ;

[0022] Among them, P A For soft countermeasure effectiveness; P B For hard-counter effectiveness; P p For survival efficiency, λ rss , λ gj , λ sc The corresponding weights are assigned to each performance level.

[0023] Furthermore, the calculation method for the soft countermeasure effectiveness is as follows:

[0024]

[0025] Among them, A i P is an influencing factor on the effectiveness index of a certain soft countermeasure target function i; i The performance index of i is the function of the target itself under conditions where there is no soft countermeasure; P is Let m be the performance index of function i under a certain soft-kill countermeasure condition; m represents the number of effective agent nodes. This represents the summation over m valid Agent nodes.

[0026] Furthermore, the calculation method for the hard-counter performance is as follows:

[0027]

[0028] Where, N load N represents the maximum number of agents currently mounted in the alliance. need The number of agents required to conduct hard confrontation against the target. P represents the average probability of a single agent destroying a target. TP P represents the probability that the agent will break through the target. d This represents the probability that the agent discovers the target. This represents the average probability of damage to the target.

[0029] Furthermore, the multi-agent survivability calculation method is as follows:

[0030]

[0031] Among them, P CThis indicates the alliance's multi-agent generation capability; E represents the number of effective agents in the alliance; N represents the total number of agents in the alliance; P S P represents cooperative survivability. D Indicating sensitivity, P KSS Indicates fragility.

[0032] Furthermore, the overall network performance of the alliance includes network reachability and network trustworthiness. The platform adjusts the alliance's entry into the next task iteration or the termination of the task based on the overall network performance of the alliance, including:

[0033] Determine whether to terminate the task based on the network reachability:

[0034] If network reachability indicates a system failure and inability to function, then terminate the task;

[0035] Otherwise, the alliance is adjusted based on the network's credibility, and the next round of task iteration begins.

[0036] Furthermore, the network reachability is represented as:

[0037] A = [a0, a1, a2, a3];

[0038] Where a0 is the probability that the alliance is in normal working condition when it starts executing the mission; a1 is the probability that the alliance's protection range is less than the normal range but greater than 1 / 2 of the normal protection range; a2 is the probability that the alliance's protection range is greater than 1 / 4 of the normal protection range but less than 1 / 2 of the normal protection range; and a3 is the probability that the alliance is unable to work and is in a faulty state.

[0039] It should be noted that the network reachability of the alliance is calculated by the task planning platform based on the basic data of each agent within the alliance. When the network reachability is a3, it indicates that the alliance is faulty and cannot function.

[0040] The network trustworthiness is represented as: C1 = α1β1 + α2β2;

[0041] β1=γ1ε1+γ2ε2+γ3ε3+γ4ε4;

[0042] Where ε1 is the network data transmission accuracy of the alliance; γ1 is the weight of data transmission accuracy in availability; ε2 is the network data update rate of the alliance; γ2 is the weight of data update rate in availability; ε3 is the network data transmission latency of the alliance; γ3 is the weight of data transmission latency in availability; ε4 is the network data transmission precision of the alliance; γ4 is the weight of data transmission precision in availability;

[0043] β2=δ1θ1+δ2θ2+δ3θ3+δ4θ4;

[0044] Wherein, δ1 is the network data loss rate of the alliance; θ1 is the weight of the data loss rate in reachability; δ2 is the network action distance of the alliance; θ2 is the weight of the action distance in reachability; δ3 is the network internal connectivity of the alliance; θ3 is the weight of the internal connectivity of the alliance in reachability; δ4 is the cross-alliance connectivity of the alliance; θ4 is the weight of the cross-alliance connectivity in reachability.

[0045] Furthermore, the alliance is adjusted based on the network's trustworthiness to proceed to the next round of task iteration.

[0046] The present invention can achieve at least one of the following beneficial effects:

[0047] By using TUU game theory to incentivize cooperation among agent nodes in a multi-agent autonomous collaborative adversarial process, and considering both sparse and dense networks, the message delivery rate of data transmission during multi-agent task execution is improved. Furthermore, in each round of task iteration, the alliance strategy is adjusted based on the evaluation results of the alliance's collaborative detection efficiency, the collaborative adversarial efficiency of multiple agents in the alliance, and the network efficiency of the alliance, thereby improving task execution efficiency.

[0048] In sparse networks, by calculating the rewards and average connection rate incentives for each agent node, active cooperation among nodes is encouraged, increasing network connection opportunities, message transmission success rates, and data link connectivity. In dense networks, by setting up proxy nodes as intermediate forwarding nodes, frequent message interactions are controlled, reducing redundant packets, lowering network load, improving network connectivity and performance, and avoiding network storms. This invention is particularly effective in improving task completion efficiency for detection and adversarial tasks requiring the transmission of large amounts of images and videos.

[0049] By calculating multiple performance indicators, including detection effectiveness, countermeasure effectiveness, and network effectiveness, the detection and countermeasure effectiveness of the coordination alliance can be accurately assessed, providing a strong basis for the alliance to adjust its detection and countermeasure strategies in a timely manner.

[0050] Other features and advantages of the invention will be set forth in the following description, and some advantages may become apparent from the description or be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained from what is particularly pointed out in the description, claims, and drawings. Attached Figure Description

[0051] The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Throughout the drawings, the same reference numerals denote the same parts.

[0052] Figure 1 This is a flowchart of the multi-agent collaborative adversarial method of the present invention;

[0053] Figure 2 This is a schematic diagram of the TUU game process of the present invention. Detailed Implementation

[0054] Preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings, which form part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, but are not intended to limit the scope of the present invention.

[0055] A specific embodiment of the present invention takes homogeneous and isomorphic multi-agent as the research object and discloses a multi-agent cooperative adversarial method based on TUU game, including: determining the detection alliance and / or adversarial alliance based on the cooperative adversarial task, wherein the alliance changes dynamically during the iteration of the task; wherein, in one task iteration, the method includes: judging the current link status of each alliance based on the basic data of each alliance;

[0056] Based on the current link status of each alliance, the TUU game data interaction method is used to incentivize the cooperation of Agent nodes in the alliance to conduct data interaction.

[0057] Adjust the data interaction strategy and autonomous detection strategy of the detection alliance based on the collaborative detection effectiveness; adjust the data interaction strategy, formation mode and autonomous confrontation strategy of the confrontation alliance based on the collaborative confrontation effectiveness;

[0058] Based on the overall network performance of the alliance, the alliance may be adjusted to proceed to the next task iteration or terminate the task.

[0059] This embodiment of the method improves the message delivery rate of data transmission during multi-agent autonomous collaborative adversarial processes by using TUU game incentives to encourage cooperation among agent nodes in the alliance for data interaction. Furthermore, in each round of task iteration, the strategy is adjusted based on the evaluation results of detection effectiveness, adversarial effectiveness, and network effectiveness to improve task execution efficiency.

[0060] In one embodiment of the present invention, a multi-agent cooperative adversarial method based on TUU game specifically includes the following steps in each task iteration:

[0061] Step S01. Determine the current link status of each detection alliance based on the basic data of each detection alliance.

[0062] Specifically, a multi-Agent task planning platform is constructed. This platform manages each agent node and builds and manages detection alliances and / or adversarial alliances based on tasks. A detection alliance refers to a multi-Agent node alliance used to execute detection tasks. Agents within this alliance can autonomously build data transmission links for data forwarding and information exchange, enabling multi-Agent collaborative detection based on detection tasks. An adversarial alliance refers to a multi-Agent node alliance used to execute adversarial tasks. Agents within this alliance can autonomously build data transmission links for data forwarding and information exchange, enabling multi-Agent collaborative adversarial operations based on adversarial tasks.

[0063] It should be noted that there can be one detection alliance and multiple adversarial alliances. For cooperative adversarial missions, detection generally precedes adversarial action, and the detection alliance and adversarial alliance change dynamically during the iteration of cooperative adversarial missions.

[0064] Specifically, during the execution of tasks by multiple agents, the multi-agent task planning platform obtains information on the number of agent nodes in each alliance in real time, monitors the network status of single / batch nodes randomly joining / leaving each alliance, and obtains basic data such as different data link types (including self-organizing network, collaborative chain, telemetry and control chain, 5G, 6G, etc.), number of links, transmission bandwidth and effective distance between agents.

[0065] Specifically, during the execution of tasks by multiple agents, the multi-agent task planning platform acquires in real time basic data such as the detection methods (e.g., radar, visible light, infrared detection), adversarial methods (including soft and hard adversarial), detection capabilities (e.g., detection range, target status identification, and fusion capabilities) of agent nodes in each alliance, as well as the current collaborative relationships of agents (including detection collaboration and adversarial collaboration).

[0066] Furthermore, the basic data of each detection alliance includes the basic data described above.

[0067] Specifically, in step S01, the information fusion method based on DS evidence theory determines the current link status of each detection alliance based on the basic data of each detection alliance.

[0068] Furthermore, the DS evidence theory is used to determine the link status of each alliance based on the basic data of each alliance, including S11-S13:

[0069] S11. The communication network state identification framework for each alliance is defined as Θ = (strong communication capability S, weak communication capability D).

[0070] S12. Determine the number of pieces of evidence for the DS theory based on the basic data of each alliance.

[0071] Specifically, the number of pieces of evidence is determined based on fundamental alliance data such as the number of links and the effective distance between nodes of all agents in the alliance; specifically, there are four pieces of evidence, including: m1 = m 度分布 m2=m 网络链路数 m3 = m 聚类系数 m4 = m 平均路径长度 .

[0072] S13. Based on evidence, use uncertainty calculation formulas and decision criteria to determine the status of each alliance link as either strong communication capability (S) or weak communication capability (D).

[0073] Specifically, the formula for calculating the uncertainty of the forecast is as follows:

[0074] m j (Θ)=-k[pF*logpF+(1-pF)*log(1-pF)];

[0075] m j (S)=pF*(1-m j (Θ));

[0076] m j (D)=(1-pF)*(1-m j (Θ));

[0077] Where j is the evidence number index, pF represents the probability of network sparsity given by the prediction method, and 1-pF represents the probability of network density. k∈(0,1) is the normalization factor.

[0078] Specifically, a change in status from official to full-fledged status means:

[0079]

[0080] Here, Bel represents the trust function.

[0081] Specifically, the multi-agent task planning platform uses judgment criteria to determine whether the link status of the alliance is either strong communication capability (S) or weak communication capability (D).

[0082] Step S02. Based on the link state of the probe alliance, use the TUU game data interaction method to incentivize the cooperation of Agent nodes in the probe alliance to conduct data interaction.

[0083] Specifically, in each round of task iteration, the TUU game data interaction method is used to incentivize the cooperation of Agent nodes in the alliance in multiple stages. Generally, k=3 stages of incentives are performed.

[0084] Specifically, when the alliance link status is weak (D), the alliance is determined to be a sparse network. It should be noted that when an alliance is a sparse network, it means that the network node density is below a set threshold, requiring incentives for active cooperation among nodes in the network to increase network connection opportunities, improve message transmission success rate, and enhance data link connectivity.

[0085] Furthermore, for alliances determined to be sparse networks, in each round of the TUU game, the agent nodes in the alliance are incentivized to cooperate based on the revenue of each agent node and the average connection rate.

[0086] Furthermore, we assume that the neighbor relationships between the nodes in each alliance are symmetrical, the communication channels are bidirectional, and data exchange occurs from the source node to the destination node via forwarded data packets. We also assume that each node has a unique and valid ID. We define a transferable payoff alliance game (N, v), where alliance N represents all agents within a single alliance in a multi-agent pool, and v represents the total payoff for each non-empty alliance.

[0087] Furthermore, each Agent node i in the alliance calculates the average reward and makes a strategy choice in the strategy space [cooperate, betray], representing whether the intermediate node i chooses to forward or discard the data packet after receiving it.

[0088] Furthermore, the average revenue comprehensively considers the average connection rate, latency, distance to the destination node, and the node's reputation score. The calculation method is as follows:

[0089] u i(s) =∑(W i –cost)+∑(W i –2cost)-∑W i +T i ;

[0090]

[0091] Where α, β, γ, μ∈(0,1), are generally determined based on historical experience; see Table 1 for the meaning of each parameter in the superscript;

[0092]

[0093] Where i and j represent any member (i.e., Agent node) in the alliance. This is the benefit of i and j working together. The loss was caused by the cooperation between i and j, therefore, It is the net benefit of i and j cooperating; similarly, It is the net benefit of j cooperating with i; a iIt represents the degree of effort that i makes in contributing to the alliance; the total net benefit compensated to i or the total net benefit taken from i to compensate other members (i.e., T). i The value (which can be positive or negative) is the sum of the differences in net income between i and other members, redistributed according to the degree of effort i makes in contributing to the alliance.

[0094] Table 1. Meaning of each parameter in the revenue calculation.

[0095] <![CDATA[cp i ]]> Average connection rate of node i <![CDATA[W i ]]> Reward for successfully passing the message to the next hop α Weight of average connection rate β Delay weight γ Distance weight from the destination node μ Reputation Weight t The delay in forwarding a data packet d Distance from the destination node <![CDATA[m i ]]> Node reputation value

[0096] Furthermore, the average connectivity of node i to node j is calculated:

[0097] Specifically, cp i It is the average connectivity of node i; As the neighbor set of node i in stage k, node i maintains the dynamic data of each neighbor node j:

[0098]

[0099] Node i evaluates the average connectivity of node j:

[0100] in

[0101] if If the threshold is exceeded, node i forwards the data packet to node j, promoting cooperation and data interaction among agent nodes in the network when the alliance is sparse.

[0102] Specifically, when the alliance link status is "strong communication capability S", the alliance is determined to be a dense network. When an alliance is a dense network, it means that the network node density is higher than a set threshold.

[0103] Furthermore, for alliances identified as dense networks, in each round of the TUU game, proxy nodes are set to incentivize cooperation among agent nodes in the alliance based on the utility values ​​of each agent node. It should be noted that by setting proxy nodes as intermediate forwarding nodes, the frequent message exchanges in the dense network are controlled, reducing network redundancy, lowering network load, and improving network connectivity and performance.

[0104] Specifically, the utility function of proxy node i is:

[0105]

[0106]

[0107] Where Pr is the probability that node i is selected as the alliance agent node in this game phase, and W iis the reward that proxy node i receives for forwarding a message, and cost is the cost incurred in forwarding a message. U is the utility value of agent node i in the (k-1)th stage. t It is the utility threshold for each stage of the game; a Delivery b Cache , d Vel , Determined based on empirical values.

[0108] Furthermore, u t The initial threshold value is the utility value of the initial agent node. At the end of each stage of the game, the multi-agent task planning platform calculates the utility value u of each agent node. i With utility threshold u t Compare, if u i Greater than u t This indicates that the node can join the alliance as a proxy node, and the timer n is incremented by 1. When n≥n is detected... max =η*N indicates that there are too many proxy nodes in the dense network, resulting in a decrease in the success rate of network message delivery and redundant data interaction. In this case, the utility threshold u of the proxy nodes in the alliance should be adjusted immediately. t This stabilizes the alliance of proxy nodes throughout the network, thereby improving network performance (e.g.) Figure 2 (As shown in the figure). Where N represents the total number of network nodes, and η represents the maximum proportion of proxy nodes allowed in the dense network.

[0109] Step S03. Adjust the data interaction strategy and autonomous detection strategy of the detection alliance based on the collaborative detection effectiveness.

[0110] Specifically, the detection performance of each Agent node is determined based on the detection performance of the sensors connected to each Agent node in the detection alliance, and the detection performance of the detection alliance is determined based on the detection performance of each node.

[0111] Specifically, different sensor metrics are normalized, and the normalization formula is expressed as follows:

[0112]

[0113] Where, θ j This represents the normalized sensor metrics; i and j represent the two sensors respectively; h i and h j These represent sensor specifications; n represents the number of sensors used for collaborative detection.

[0114] Furthermore, the detection performance of each agent node is determined based on the normalized sensor metrics of the sensors mounted on each agent node. It should be noted that the types and numbers of sensors mounted on each agent node may differ, thus affecting their detection performance. The detection performance of each agent node is then used to determine the collaborative detection performance of n nodes in the detection alliance.

[0115] Furthermore, suppose the n nodes of the detection alliance X are represented as X = (X1, X2, ..., X...). n Its detection performance is expressed as x = (x1, x2, ..., x...). n If the detection efficiency of the alliance is F(x), then the detection efficiency of each node is calculated based on the detection efficiency of each node and the corresponding weight coefficient.

[0116] For example, when n=2, The term is a quadratic nonlinear term, representing the detection efficiency generated by the pairwise cooperation of nodes (where k and l represent two nodes respectively); where a kl The corresponding weights are determined based on empirical values; when n = 3, The term is a cubic nonlinear term, representing the detection performance generated by the three-three cooperation of nodes, where a kl The weights for the three-by-three collaboration of nodes are given; similarly, for n nodes, F(x) is calculated based on the detection efficiency of n nodes and the corresponding weight coefficients.

[0117] It should be noted that the above weighting coefficient 'a' kl and a kls The detection performance weight coefficients of the n nodes are obtained based on experience, which includes a comprehensive consideration of the capabilities weights of each sensor at each node. For example, Table 2 shows the weight coefficients of each sensor's capabilities in the system's detection performance.

[0118] Table 2. Ability Weighting Coefficient Table

[0119]

[0120] It should be noted that as the number of nodes involved in collaborative detection increases, the detection efficiency of the detection alliance improves accordingly.

[0121] Furthermore, the data interaction strategy and autonomous detection strategy of the detection alliance are adjusted based on the collaborative detection effectiveness. Adjusting the data interaction strategy includes selecting data interaction strategies such as self-organizing networks, telemetry and control chains, and collaborative chains; adjusting the detection strategy includes adjusting the types and number of sensors used in the collaborative detection process. For example, based on the current mission scenario, a data interaction strategy that improves data transmission efficiency and accuracy in the current scenario is selected; based on the characteristics of the current detection target, suitable sensor types are specifically selected for coordinated detection of the current target.

[0122] It should be noted that step S03 is carried out after each stage of the TUU game in step S02. That is, one stage of the TUU game corresponds to one step S03 to adjust the data interaction strategy and autonomous detection strategy of the detection alliance based on the collaborative detection effectiveness, and then enter the next stage of the TUU game of the detection alliance.

[0123] Step S04. Based on the link state of the adversarial alliance, use the TUU game data interaction method to incentivize the cooperation of Agent nodes in the adversarial alliance to conduct data interaction.

[0124] Specifically, step S04 uses the TUU game data interaction method to incentivize the cooperation of Agent nodes in the adversarial alliance, which is the same as the incentivization process of step SO2 to probe the alliance.

[0125] Step S05. Adjust the data interaction strategy, formation method and autonomous confrontation strategy of the confrontation alliance based on the collaborative confrontation effectiveness.

[0126] Specifically, collaborative combat effectiveness includes formation effectiveness and combat effectiveness;

[0127] The formation effectiveness includes formation assembly effectiveness, formation maintenance effectiveness, formation evasion effectiveness, and overall dynamic formation effectiveness;

[0128] The adversarial effectiveness includes soft adversarial effectiveness, hard adversarial effectiveness, multi-agent survivability, and multi-agent collaborative adversarial dynamic effectiveness.

[0129] Furthermore, the calculation method for formation assembly effectiveness is as follows:

[0130]

[0131] Among them, C nlxh C represents the energy consumed during mobilization. sx Indicating the timeliness of the assembly, C wx Indicates the magnitude of threats or obstacles encountered during the assembly process. These are the corresponding weighting coefficients.

[0132] Specifically, formation assembly refers to the process by which agents, starting from any initial state, adjust the maneuver parameters of each agent according to certain performance indicators and constraints, so that all agents in the formation can reach a certain assembly point within a finite time.

[0133] Furthermore, the calculation method for formation maintenance effectiveness is as follows:

[0134]

[0135] Where, N slIndicates the size of the formation, C fzd Indicates the formation complexity, P jg Indicates the interval between formations.

[0136] Specifically, formation maintenance effectiveness refers to the ability of agents to maintain formation or restore formation after encountering unforeseen circumstances that disrupt the formation during formation maneuvers.

[0137] Furthermore, the calculation method for formation evasion effectiveness includes the calculation of the cooperative evasion capabilities of a single formation and multiple formations, among which,

[0138] The calculation method for the evasion effectiveness of a single formation is as follows:

[0139]

[0140] Among them, g f To maximize the available overload of the task planning platform, g g For the maximum available overload of the threat target, v f For the platform's maximum speed, v g Let g(t) be the maximum velocity of the threatening target, and g(t) be a function of the effective destruction of the threatening target and time.

[0141] The calculation method for the cooperative evasion effectiveness of multiple formations is as follows: Where m represents the number of formations and j represents the formation number.

[0142] Furthermore, the calculation method for the overall effectiveness of dynamic formations is generally performed on a single formation, and the formula is expressed as:

[0143] E BD =λ jj E JJ +λ bc E BC +λ gb E GB

[0144] Among them, E JJ For formation assembly capability, E BC To maintain the formation's capabilities, E GB For formation evasion capabilities, λ jj , λ bc , λ gb The corresponding weights.

[0145] Furthermore, the calculation method for soft countermeasure effectiveness is as follows:

[0146]

[0147] Among them, A i P is an influencing factor on the effectiveness index of a certain soft countermeasure target function i;i The performance index of i is the function of the target itself under conditions where there is no soft countermeasure; P is Let m be the performance index of function i under a certain soft-kill countermeasure condition; m represents the number of effective agent nodes. This represents the summation over m valid Agent nodes.

[0148] It should be noted that the soft countermeasures include electromagnetic interference, satellite navigation deception, chaff reflection jamming, and long-range support jamming; the soft countermeasure target functions refer to the countermeasures against the target's electromagnetic emission, satellite navigation, chaff reflection, and other functions. The soft countermeasure effectiveness calculation is the countermeasure effectiveness formed by the combined use of two or more soft jamming methods.

[0149] Furthermore, the calculation method for hard confrontation effectiveness is as follows:

[0150]

[0151] Where, N load N represents the maximum number of agents currently mounted in the alliance. need The number of agents required to conduct hard confrontation against the target. P represents the average probability of a single agent destroying a target. TP P represents the probability that the agent will break through the target. d This represents the probability that the agent discovers the target. This represents the average probability of destroying the target. It should be noted that the effectiveness of hard-countermeasures is related to the agent's payload, type, navigation capabilities, penetration capabilities, and target characteristics. Among these, type, navigation capabilities, penetration capabilities, and target characteristics determine the probability of obtaining an image, the probability of detection, and the probability of destruction of the target.

[0152] Furthermore, the calculation method for multi-agent survivability is as follows:

[0153]

[0154] Among them, P C This indicates the alliance's multi-agent generation capability; E represents the number of effective agents in the alliance; N represents the total number of agents in the alliance; P S P represents cooperative survivability. D Indicating sensitivity, P KSSThis refers to vulnerability. Specifically, multi-agent survivability comprises two parts: sensitivity and vulnerability. Sensitivity refers to an agent's ability to evade threats; factors such as agent design, countermeasures, payload, and the threat environment all affect survivability. Vulnerability reflects an agent's inability to withstand damage and is a measure of the degree of damage upon being hit.

[0155] Furthermore, the calculation method for the dynamic effectiveness of multi-agent cooperative adversarial operations is as follows:

[0156] P DJ =λ rss P A +λ gj P B +λ sc P p ;

[0157] Among them, P A For soft countermeasure effectiveness; P B For hard-counter effectiveness; P p For survival efficiency, λ rss , λ gj , λ sc The corresponding weights are assigned to each performance level.

[0158] Furthermore, based on the effectiveness of collaborative confrontation, the data interaction strategy, formation method, and autonomous confrontation strategy of the confrontation alliance are adjusted:

[0159] The data interaction strategy is adjusted based on the overall efficiency of dynamic formation and the dynamic efficiency of multi-agent collaborative confrontation. For example, when the overall efficiency and / or the dynamic efficiency of multi-agent collaborative confrontation are low, a data interaction strategy that can enhance the data transmission rate in the current task scenario is selected according to the task situation.

[0160] The formation method is adjusted based on the formation assembly efficiency, formation maintenance efficiency, single formation evasion efficiency, multi-formation collaborative evasion efficiency, and overall dynamic formation efficiency. For example, for the corresponding efficiency value with a lower calculated result, the corresponding formation method is adjusted according to historical experience to improve the corresponding efficiency.

[0161] The adversarial strategy is adjusted based on the effectiveness of soft adversarial tactics, the effectiveness of hard adversarial tactics, the survivability of multiple agents, and the dynamic effectiveness of multi-agent collaborative adversarial tactics. The adversarial strategy includes soft adversarial tactics and hard adversarial tactics. For example, when the effectiveness of soft adversarial tactics is high, the adversarial strategy adopts a soft adversarial tactic as the main adversarial tactic; conversely, when the effectiveness of hard adversarial tactics is high, the adversarial strategy is adjusted to a hard adversarial tactic as the main adversarial tactic.

[0162] It should be noted that step S05 is carried out after each stage of the TUU game in step S04. That is, one stage of the TUU game corresponds to one step S05 that adjusts the data interaction strategy, formation method and autonomous confrontation strategy of the confrontation alliance based on the cooperative confrontation effectiveness, and then enters the next stage of the TUU game of the confrontation alliance.

[0163] Step S06. Adjust the alliance to enter the next task iteration or end the task based on the overall network performance of the alliance.

[0164] Specifically, the overall network performance of the alliance includes network reachability and network trustworthiness. The platform adjusts the alliance's entry into the next task iteration or the termination of the task based on the overall network performance of the alliance, including:

[0165] Determine whether to terminate the task based on the network reachability:

[0166] If network reachability indicates a system failure and inability to function, then terminate the task;

[0167] Otherwise, the alliance is adjusted based on the network's credibility, and the next round of task iteration begins.

[0168] Furthermore, the network reachability is represented as:

[0169] A = [a0, a1, a2, a3];

[0170] Where a0 is the probability that the alliance is in normal working condition when it starts executing the mission; a1 is the probability that the alliance's protection range is less than the normal range but greater than 1 / 2 of the normal protection range; a2 is the probability that the alliance's protection range is greater than 1 / 4 of the normal protection range but less than 1 / 2 of the normal protection range; and a3 is the probability that the alliance is unable to work and is in a faulty state.

[0171] The network trustworthiness is represented as: C1 = α1β1 + α2β2;

[0172]

[0173] Where ε1 is the network data transmission accuracy of the alliance; γ1 is the weight of data transmission accuracy in availability; ε2 is the network data update rate of the alliance; γ2 is the weight of data update rate in availability; ε3 is the network data transmission latency of the alliance; γ3 is the weight of data transmission latency in availability; ε4 is the network data transmission precision of the alliance; γ4 is the weight of data transmission precision in availability;

[0174] β2=δ1θ1+δ2θ2+δ3θ3+δ4θ4;

[0175] Wherein, δ1 is the network data loss rate of the alliance; θ1 is the weight of the data loss rate in reachability; δ2 is the network action distance of the alliance; θ2 is the weight of the action distance in reachability; δ3 is the network internal connectivity of the alliance; θ3 is the weight of the internal connectivity of the alliance in reachability; δ4 is the cross-alliance connectivity of the alliance; θ4 is the weight of the cross-alliance connectivity in reachability.

[0176] Furthermore, based on the network trustworthiness, the alliances are adjusted, including the number of agents in each alliance and data interaction strategies, and then the next round of task iteration is initiated, that is, step S01 is restarted for the next round of task iteration.

[0177] This embodiment discloses a cooperative adversarial method based on TUU game theory. In sparse networks, it calculates the payoff and average connection rate incentives for each agent node, encouraging active cooperation among nodes, increasing network connection opportunities, message transmission success rate, and data link connectivity. In dense networks, it sets up proxy nodes as intermediate forwarding nodes to control frequent message interactions, reduce redundant packets, lower network load, improve network connectivity and performance, and avoid network storms. This method is particularly effective in improving task completion efficiency for detection and adversarial tasks requiring large-scale image and video transmission. By calculating multiple performance indicators—detection effectiveness, adversarial effectiveness, and network effectiveness—it accurately evaluates the detection and adversarial effectiveness of the coordinated alliance, providing a strong basis for timely adjustments to the alliance's detection and adversarial strategies.

[0178] It should be noted that the above embodiments are based on the same inventive concept, and any parts not described repeatedly can be referenced from each other.

[0179] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A multi-agent cooperative adversarial method based on TUU game theory, characterized in that, The detection alliance and / or adversarial alliance are determined based on the collaborative adversarial task, and the alliance changes dynamically during the task iteration process; wherein, in one task iteration process, the following is included: determining the current link status of each alliance based on the basic data of each alliance; Based on the current link status of each alliance, the TUU game data interaction method is used to incentivize the cooperation of Agent nodes in the alliance to conduct data interaction. The detection alliance's data interaction strategy and autonomous detection strategy are adjusted based on collaborative detection effectiveness; the adversarial alliance's data interaction strategy, formation method, and autonomous adversarial strategy are adjusted based on collaborative adversarial effectiveness; the collaborative adversarial effectiveness includes formation effectiveness and adversarial effectiveness; the formation effectiveness includes formation assembly effectiveness, formation maintenance effectiveness, formation evasion effectiveness, and dynamic formation overall effectiveness; the adversarial effectiveness includes soft adversarial effectiveness, hard adversarial effectiveness, multi-agent survivability, and multi-agent collaborative adversarial dynamic effectiveness; wherein the calculation method for multi-agent collaborative adversarial dynamic effectiveness is as follows: ;in, For soft countermeasure effectiveness; For hard-counter effectiveness; For survival ability, , , The weights corresponding to each performance level are defined; the calculation method for the soft countermeasure performance is as follows: ;in, For a certain soft countermeasure target function i Factors influencing performance indicators; Functional under conditions where the target itself has no soft countermeasures i Performance indicators; The target function under certain soft-kill countermeasure conditions i The performance metric; m represents the number of effective Agent nodes; This represents the summation over m valid Agent nodes; Based on the overall network performance of the alliance, the alliance may be adjusted to proceed to the next task iteration or terminate the task.

2. The cooperative confrontation method according to claim 1, characterized in that, Using the DS evidence theory, the link status of each alliance is determined based on the basic data of each alliance, including: The communication network state identification framework for each alliance is determined as follows: = (Strong communication ability S, weak communication ability D); The number of pieces of evidence for the DS theory was determined based on the basic data of each alliance. Based on the number of evidences, the uncertainty calculation formula and decision criteria are used to determine the status of each alliance link as either strong communication capability (S) or weak communication capability (D).

3. The cooperative confrontation method according to claim 2, characterized in that, The data interaction method based on the link status of each alliance and using TUU game to incentivize cooperation among agent nodes in the alliance includes: Alliances with weak communication capability (D) in the link interruption state are classified as sparse networks, while alliances with strong communication capability (S) in the link interruption state are classified as dense networks. For alliances that are determined to be sparse networks, in each round of TUU game, the agent nodes in the alliance are incentivized to cooperate based on the revenue of each agent node and the average connection rate. For alliances that are determined to be dense networks, in each round of the TUU game, agent nodes are set to incentivize cooperation among agent nodes in the alliance based on the utility value of each agent node in the alliance.

4. The cooperative confrontation method according to claim 1, characterized in that, The calculation method for the hard-counter effectiveness is as follows: ； in, This represents the maximum number of agents currently mounted by the alliance. The number of agents required to conduct hard confrontation against the target. ; This represents the average probability of a single agent causing damage to a target. This represents the probability that the agent will break through the target. This represents the probability that the agent discovers the target. This represents the average probability of damage to the target.

5. The cooperative confrontation method according to claim 1, characterized in that, The method for calculating the survivability of multiple agents is as follows: ； in, This indicates the alliance's multi-agent generation capability; E represents the number of effective agents in the alliance; N represents the total number of agents in the alliance. Indicates cooperative survivability. Indicates sensitivity, Indicates fragility.

6. The cooperative confrontation method according to claim 1, characterized in that, The overall network performance of the alliance includes network reachability and network trustworthiness. Adjusting the alliance to enter the next task iteration or terminate the task based on the overall network performance includes: Determine whether to terminate the task based on the network reachability: If network reachability indicates a system failure and inability to function, then terminate the task; Otherwise, the alliance is adjusted based on the network's credibility, and the next round of task iteration begins.

7. The cooperative confrontation method according to claim 6, characterized in that, The network reachability is represented as follows: ； in, The probability that the alliance is in normal working order when it begins to carry out its mission; The probability that the alliance's coverage range is less than half the normal coverage range; The probability that the alliance's coverage is greater than 1 / 4 of the normal coverage and less than 1 / 2 of the normal coverage; The probability that the alliance is unable to function and is in a faulty state; The network trustworthiness is represented as follows: ; ； in, For the accuracy of network data transmission within the alliance; The weight of data transmission accuracy in availability; For the alliance's network data update rate; Weight of data update rate in availability; For network data transmission latency of the alliance; Weighting of data transmission latency in availability; To ensure the accuracy of network data transmission within the alliance; The weight of data transmission accuracy in availability; ； in, For the alliance's network data loss rate; Weighting of data loss rate in reachability; The network's effective range for the alliance; The weight of the effective distance in reachability; For the internal network connectivity of the alliance; Weights of connectivity within the alliance in terms of reachability; For the alliance's cross-alliance connectivity; Weights of cross-federal connectivity in reachability.

Images