A Multi-Agent Collaborative System for Weapon Simulation Based on a Large Model

By constructing a multi-agent collaborative system, the shortcomings of existing weapon simulation technology in simulating complex battlefield environments and multi-service coordination have been addressed, achieving efficient and low-cost simulation results and reducing trial-and-error costs and safety risks.

CN123116807APending Publication Date: 2026-11-17LANZHOU UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611283458.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-24
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing weapons simulation technologies are ill-suited to the systematic, intelligent, multi-dimensional, and highly dynamic characteristics of modern warfare, and are ill-suited to simulating complex battlefield environments and multi-service coordination. Furthermore, they are subject to high trial-and-error costs and significant security risks.

Method used

Construct a multi-agent collaborative system based on a large model, including strategic, execution, environmental perception, and group-level agents. Conduct collaborative decision-making through information production, dissemination, and feedback stages. Combine the multi-dimensional attributes of the public sub-agent and the information filtering of the media sub-agent to simulate the real battlefield public opinion environment and group cognition.

Benefits of technology

It improves the realism and rationality of simulation results, reduces trial and error costs, adapts to the complex scenarios of modern warfare, reduces manpower, material and time costs, and lowers security risks.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123116807A_ABST
    Figure CN123116807A_ABST
Patent Text Reader

Abstract

This invention relates to the field of intelligent analysis technology, specifically to a multi-agent collaborative system for weapons simulation based on a large model. The system includes an agent module for establishing several agents at different levels. Agents at different levels execute corresponding information flow strategies according to the information processing stage. The levels include a strategic layer, an execution layer, an environmental perception layer, and a group layer. Agents at the strategic layer are used for macro-level situation assessment and overall strategy formulation; agents at the execution layer are used for the implementation of specific tasks; agents at the environmental perception layer are used for monitoring and feedback on the external environment; and agents at the group layer are used to represent the collective behavior of the group. This invention relies on multi-level, multi-agent collaborative interaction and the intelligent game-playing capabilities of a large model to accurately simulate complex combat scenarios such as dynamic and sudden battlefield factors, equipment condition fluctuations, and multi-weapon, multi-force coordinated confrontation, effectively improving the realism and battlefield relevance of weapons simulation results.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of intelligent analysis technology, specifically to a multi-agent collaborative system for weapon simulation based on a large model. Background Technology

[0002] Current mainstream weapon simulation technologies are mainly divided into two categories: traditional rule-driven simulation and single-agent data-driven simulation. These rely on fixed mathematical models, preset combat rules, and static equipment parameters to complete simulations, enabling basic adversarial simulations in conventional and simple scenarios. However, as modern warfare exhibits a trend towards systematization, intelligence, multi-dimensionality, and strong dynamism, scenarios involving complex battlefield environments, multi-weapon coordination, dynamic tactical games, and sudden battlefield situations are constantly increasing, highlighting the shortcomings and limitations of existing weapon simulation technologies.

[0003] Existing simulation systems mostly employ fixed rule models and static parameter systems, resulting in weak model generalization capabilities. They struggle to adapt to unforeseen dynamic changes on the battlefield and fail to simulate complex, random scenarios such as equipment malfunctions, environmental interference, on-the-spot tactical adjustments, and autonomous coordination among multiple branches of service and weapons in real combat. This leads to significant discrepancies between the simulation models and actual battlefield conditions, resulting in insufficient realism and consistency in the simulation results. Current technologies often employ a single-agent independent simulation mode, with each combat unit and weapon simulation module operating independently. This lack of multi-agent collaborative interaction, dynamic game theory, and coordinated decision-making capabilities makes it difficult to achieve systemic confrontation simulations involving multiple weapons, forces, and dimensions, and to simulate the characteristics of modern integrated joint operations. Existing simulation systems lack flexibility in trial and error, hindering rapid iteration and optimization of simulation models and operational plans. Relying on real battlefields for weapon testing and tactical trial and error would incur extremely high costs in terms of manpower, resources, and time, and pose significant security risks, failing to meet the application requirements of modern intelligent weapon simulations for high efficiency, realism, low cost, and high adaptability.

[0004] Therefore, this invention proposes a multi-agent collaborative system for weapon simulation based on a large model, in order to improve the realism of simulation results and reduce the trial-and-error costs in the real world. Summary of the Invention

[0005] To address the aforementioned issues, this invention provides a multi-agent collaborative system for weapon simulation based on a large model, which improves the realism of simulation results and reduces the trial-and-error costs in the real world.

[0006] To achieve the above objectives, the technical solution of the present invention is as follows: A multi-agent cooperative system for weapon simulation based on a large model, comprising: The intelligent agent module is used to establish several intelligent agents at different levels based on different job functions; intelligent agents at different levels execute corresponding information flow strategies according to the information processing stage. The hierarchy includes the strategic layer, the execution layer, the environmental awareness layer, and the group layer; the information processing stage includes the information production stage, the information dissemination stage, and the information feedback stage. Strategic layer agents are used for macro-level situation assessment and overall strategy formulation; execution layer agents are used for the implementation of specific tasks; environmental perception layer agents are used for monitoring and feedback on the external environment; and swarm agents are used for representing the collective behavior of a group of people.

[0007] Furthermore, within the swarm intelligence agent, several sub-intelligent agents of the public with different backgrounds are established; each sub-intelligent agent of the public includes fixed elements such as regional attributes, political orientation, information sources, education level, and core concerns.

[0008] Furthermore, information flow strategies include: Information production stage: Based on its own position and goals, the strategic layer intelligent agent produces raw information; during the raw information production process, the strategic layer intelligent agent executes the narrative generation reasoning chain and outputs multi-dimensional metadata; Information dissemination stage: In the environmental perception layer, several media sub-agents representing different positions are established; each media sub-agent filters and reconstructs information based on its own editing preferences; the filtered and reconstructed information enters the group dissemination network composed of public sub-agents for cognitive processing; and the filtering and reconstruction follows the behavioral logic of the corresponding media sub-agent. The public's sub-agents combine their own background to judge credibility, compare it with existing cognitive frameworks, and decide whether to accept or spread it; Information feedback phase: The propagation results of the public sub-agents have a reverse impact on the strategic layer agents.

[0009] Furthermore, changes in the attitudes of each citizen sub-agent influence other agents at different levels connected to the citizen sub-agent through social networks, creating a ripple effect.

[0010] Furthermore, the narrative generation reasoning chain, from top to bottom, includes: Realistic interest analysis: The strategic layer agent first identifies the connection between the current issue and its own core interests, conducts a comprehensive analysis of multi-dimensional information, and assesses the advantages and disadvantages of the current issue for itself. Audience entry point identification: The strategic layer agent analyzes the concerns, information receiving habits and existing cognitive frameworks of different audience groups to find the optimal entry point; Narrative Framework Construction: Based on the results obtained from the analysis of real-world interests and the identification of audience entry points, the strategic layer agent constructs the narrative structure, including the core argument, supporting evidence, keyword system, and expected rebuttal plan; In the process of establishing the narrative structure, the strategic layer agent anticipates potential attack points and prepares defensive arguments in advance for those potential attack points. Self-vulnerability scanning: While generating the narrative structure, the strategic layer agent simultaneously conducts self-criticism to identify hypothetical vulnerabilities in its own discourse that can be exploited by the adversary.

[0011] Furthermore, when the strategic layer agent generates logic as the counterattacker, it breaks down the opponent's arguments based on the opponent's specific points while generating its own. The specific process is as follows: Adversary discourse analysis: The strategic layer intelligent agent deeply analyzes the adversary's narrative structure and core arguments, and conducts multi-dimensional evaluation and analysis; Counterattack angle selection: Based on the evaluation and analysis results, the strategic layer agent generates several counterattack strategies, evaluates the effect of each counterattack strategy, and selects the optimal counterattack strategy after evaluation. Hedging Narrative Construction: After selecting a counterattack strategy, the strategic layer agent constructs a targeted narrative structure.

[0012] Furthermore, the media sub-agent processes the original narrative as follows: Selective attention: Based on its own positioning, the media sub-agent selectively emphasizes certain information elements; supporting arguments are placed at the top, while unfavorable details are relegated to the end or omitted entirely; and the selection logic follows the media sub-agent's own narrative logic and audience expectations. Emotional tone adjustment: The media sub-agent adjusts the emotional tone of the words used to package the importance of objective facts.

[0013] Framework Restructuring: When the media sub-agent performs the highest level of media filtering, it restructures the framework, shifting the topic from one dimension to another.

[0014] Furthermore, each citizen's sub-agent possesses three key elements: persistence and memory, differentiated interpretation, and social influence network. Persistence and Memory: The public intelligent agent is a "fixed individual" that maintains continuity throughout the entire deduction process, recording the attitude, information received, and judgments formed in the previous round of deduction, and updating it based on the memory of the previous round in the new round of deduction; Differentiated interpretation: For the same piece of information, sub-agents from different backgrounds output different interpretations; Social Influence Network: Implicit social influence relationships exist between citizen sub-agents; when a citizen sub-agent decides to "forward" a message, it will affect the message reception of other citizen sub-agents within that citizen sub-agent's social circle.

[0015] Furthermore, it also includes an emergency module; the emergency module is used to inject several unexpected situations during the simulation process.

[0016] Furthermore, the environmental perception layer intelligent agent introduces battlefield perception and information fog mechanisms: Reconnaissance mechanism: Participating parties need to "illuminate" the fog of war through reconnaissance operations; Information Update: The information obtained through reconnaissance changes accordingly over time; Deception and counter-deception: During the simulation, participating parties are allowed to use interference tactics to mislead the opponent's reconnaissance and judgment.

[0017] The above approach has the following beneficial effects: 1. This solution constructs a four-tiered differentiated intelligent agent collaborative architecture comprising a strategic layer, an execution layer, an environmental perception layer, and a group layer. This optimizes traditional weapon simulation technologies, which suffer from isolated calculations by single intelligent agents, fragmented modules, and a lack of collaborative interaction. Based on the functional division of labor among different levels of intelligent agents and a full-process information flow strategy, this system achieves comprehensive, interconnected simulation of battlefield situation assessment, tactical task execution, dynamic environmental perception, and group behavior simulation. Simultaneously, through multi-dimensional attribute configuration of the civilian sub-agent and information filtering and reconstruction mechanism of the media sub-agent, it accurately recreates the impact of public opinion, group cognition, and information dissemination on the operational situation in real battlefield conditions. Coupled with a large-scale model-driven narrative generation and reasoning chain, it can autonomously complete battlefield interest analysis, audience adaptation, narrative construction, and vulnerability self-checking. It can also precisely dismantle and counterattack enemy arguments, effectively solving the problems of weak generalization ability, lack of dynamic game theory ability, and limited scenario simulation in traditional simulation models. This improves the fit between weapon simulation and real battlefield conditions, making the simulation results more realistic, reasonable, and valuable for reference. 2. This solution adds an emergency module and battlefield information fog and reconnaissance deception mechanisms. Combining the characteristics of civilian intelligent agent memory iteration, differentiated interpretation, and social ripple propagation, it constructs highly dynamic, highly random, and highly interactive intelligent simulation scenarios. These scenarios are adapted to the systematic, multi-dimensional, highly dynamic, and unpredictable characteristics of modern warfare, efficiently simulating complex combat scenarios such as battlefield environmental interference, equipment status fluctuations, on-the-spot tactical adjustments, evolving public opinion, and dynamic multi-agent games. Furthermore, relying on the collaborative decision-making and autonomous game-playing capabilities of multi-level intelligent agents, it eliminates the need for weapons testing and tactical trial-and-error on the actual battlefield, thereby reducing the manpower, material resources, and time costs required for traditional combat simulations and mitigating the security risks associated with actual combat testing.

[0018] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the logical structure of an embodiment of the weapon simulation multi-agent collaborative system based on a large model according to the present invention. Detailed Implementation

[0020] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] The following detailed description illustrates the specific implementation method: Example 1:

[0022] As attached Figure 1 As shown, a multi-agent collaborative system for weapon simulation based on a large model includes an agent module for establishing several agents at different levels based on their different job functions. Agents at different levels execute corresponding information flow strategies according to the information processing stage, thereby interpreting the situational awareness, forming the intentions of both sides, and determining the main attack direction in a confrontational scenario. The levels include a strategic layer, an execution layer, an environmental perception layer, and a group layer; the information processing stage includes an information production stage, an information dissemination stage, and an information feedback stage.

[0023] Specifically, the functions performed by intelligent agents at different levels are as follows: The strategic layer intelligent agent is used for macro-situation assessment and overall strategy formulation: in tactical confrontation scenarios, it analyzes the impact of current issues on its own interests, identifies the most effective audience entry point, and constructs a persuasive narrative framework.

[0024] The execution layer agents are used to implement specific tasks, and when given instructions from the strategic layer agents, they autonomously adjust themselves according to the situation on the ground. For example, they determine whether there are any abnormal movements at the current location, whether resources are sufficient, and whether coordination with neighboring units is smooth. Based on these judgments, they can autonomously decide whether to strictly execute HQ instructions or take flexible actions according to the actual situation.

[0025] The environmental perception layer agents are used to monitor and provide feedback on the external environment. While these agents do not directly participate in decision-making, they continuously collect and disseminate information, creating information flow among the agents and influencing the cognition of the others.

[0026] Swarm intelligence is used to represent the collective behavior of a group of people. Within the swarm intelligence, several sub-intelligent agents with different backgrounds are established; each sub-intelligent agent includes fixed elements such as regional attributes, political leanings, information sources, education level, and core concerns. Thus, each sub-intelligent agent represents a specific social group profile, and each sub-intelligent agent possesses three key elements: persistence and memory, differentiated interpretation, and social influence network. Persistence and Memory: The public intelligent agent is a "fixed individual" that maintains continuity throughout the entire deduction process, recording the attitude, information received, and judgments formed in the previous round of deduction, and updating it based on the memory of the previous round in the new round of deduction; Differentiated Interpretation: For the same piece of information, sub-agents from different backgrounds will output different interpretations. For example, workers with a politically conservative orientation may pay more attention to the stability of policies, while scholars with a politically liberal orientation may pay more attention to the innovativeness of policies; those who obtain information from official media may be more trusting of policies, while those who obtain information from social media may be more skeptical.

[0027] Social Influence Networks: Implicit social influence relationships exist among sub-agents; when a sub-agent decides to "forward" a message, it affects the information reception of other sub-agents within that sub-agent's social circle. This social network effect makes the dissemination of information and the evolution of attitudes exhibit complex non-linear relationships.

[0028] The information flow mechanism between intelligent agents (i.e., information within the system) is not an instantaneous broadcast across the entire network, but rather flows along specific paths, constantly being processed, interpreted, and distorted during the flow, forming the three information processing stages mentioned above. The specific information flow strategy is as follows: Information Production Stage: Based on its own stance and goals, the strategic layer agent produces raw information. During the raw information production process, the strategic layer agent executes a narrative generation reasoning chain, outputting multi-dimensional metadata. The produced information is not isolated text, but contains metadata with a clear target audience, expected effects, and potential vulnerabilities.

[0029] Information dissemination stage: Within the environmental perception layer, several media sub-agents representing different stances are established. Each media sub-agent filters and reconstructs information based on its own editing preferences (e.g., selectively processing raw information, amplifying some, downplaying others, or distorting some; this processed information is then transmitted to downstream agents). The filtered and reconstructed information enters the collective dissemination network composed of public sub-agents for cognitive processing. Furthermore, the filtering and reconstruction follows the behavioral logic of the corresponding media sub-agent. The citizen sub-agents assess credibility based on their own background, compare it with existing cognitive frameworks, and decide whether to accept or spread the message. Simultaneously, changes in the attitudes of each citizen sub-agent influence other agents at different levels connected to them through social networks, creating a ripple effect.

[0030] Information feedback phase: The propagation results of the public sub-agents influence the strategic layer agents in turn. For example, when they see that their own narrative has gained widespread support, the strategic layer agents may reinforce this direction; when they find that certain arguments have triggered strong backlashes, they may adjust their strategies. This feedback is not immediate or complete, but rather has a time lag and attenuation, making it closer to the real situation.

[0031] Furthermore, the complexity of the system is also reflected in the cooperative evolution among agents. The behavior of a single agent may be relatively simple, but when dozens of agents operate simultaneously and influence each other, unpredictable macroscopic phenomena emerge. Stance polarization: Initially, the attitudes of the public agents may follow a normal distribution; however, as the game progresses, agents with neutral attitudes may gradually be drawn to the arguments on both sides, eventually leading to a polarized pattern. This polarization is not a result of system design, but rather an emergent product of the mutual influence between agents.

[0032] Information cocoon: When individuals continuously receive information from specific sources, they reinforce their existing stances and become less receptive to external information. Over time, this creates a cognitive gap between different groups, with the same information eliciting drastically different reactions in different groups, which are reflected in evaluations.

[0033] In cognitive adversarial deduction, the initiation of the narrative generation reasoning chain is not a simple template filling, but a complex process involving multiple steps of reasoning. The specific process, from top to bottom, includes: Realistic Stake Analysis: The strategic agent first identifies the connection between the current issue and its own core interests, conducts a comprehensive analysis of multi-dimensional information (such as the issue's background, known facts, and the positions of participating parties), and assesses the advantages and disadvantages of the current issue for itself. For example, the strategic agent might evaluate: "What does this issue mean for us? What resources can we utilize? Where are our most vulnerable points?"

[0034] Audience Entry Point Identification: The strategic layer agent analyzes the concerns, information receiving habits, and existing cognitive frameworks of different audience groups to find the optimal entry point. In this process, the strategic layer agent weighs: "Is appealing to rational logic more effective, or is emotional mobilization more effective? Is emphasizing shared interests more persuasive, or is creating conflict more effective in spreading the message?"

[0035] Narrative Framework Construction: Based on the results of realistic interest analysis and audience segmentation, the strategic layer agent constructs the narrative structure, including core arguments, supporting evidence, a keyword system, and anticipated rebuttal plans. Furthermore, during the construction of the narrative structure, the strategic layer agent anticipates hypothetical attack points and prepares defensive arguments in advance to address these points.

[0036] Self-vulnerability scanning: While generating the narrative structure, the strategic layer agent simultaneously conducts self-criticism to identify hypothetical vulnerabilities in its own discourse that can be exploited by the adversary.

[0037] When the strategic layer agent generates logic as the counterattacker, its task becomes more complex: while generating its own arguments, it deconstructs them based on the opponent's specific points. The specific process is as follows: Adversary Argument Analysis: The strategic-level intelligent agent deeply analyzes the opponent's narrative structure and core arguments, conducting multi-dimensional evaluation and analysis (e.g., what are the supporting arguments? Where are the weak points in the logical chain? Which arguments are facts, and which are speculations? Which conclusions are over-derived?). Counterattack Angle Selection: Based on the evaluation and analysis results, the strategic layer agent generates several counterattack strategies and evaluates the effectiveness of each strategy. After evaluation, the optimal counterattack strategy is selected (is a direct confrontation more powerful, or a flanking maneuver more effective? Is exposing the opponent's logical flaws more likely to shake their foundation, or is shifting the topic more likely to gain the initiative? Different choices will lead to completely different game outcomes). Hedging Narrative Construction: After selecting a counterattack strategy, the strategic-level agent constructs a targeted narrative structure. Effective counterattacks are not simple denials, but rather the provision of alternative explanatory frameworks—different interpretations of the same facts; different solutions to the same concerns. This "framework competition" is the core of cognitive confrontation.

[0038] Although the system is configured with several media sub-agents, each with a clear stance and editing style, their processing of the same original narrative may vary greatly. Therefore, the processing of the original narrative by the media sub-agents is as follows: Selective attention: Based on its own positioning, the media sub-agent selectively emphasizes certain information elements; supporting arguments are placed at the top, while unfavorable details are relegated to the end or omitted entirely; and the selection logic follows the media sub-agent's own narrative logic and audience expectations.

[0039] Emotional Tone Adjustment: The media sub-agent adjusts the emotional tone of its wording to package objective facts in terms of importance. The same objective fact can be packaged as a "major breakthrough" or a "crisis signal." The media sub-agent adjusts the emotional tone of its wording—whether it's "decisive decision" or "reckless action," "rational voice" or "indifferent statement"—these subtle differences in wording significantly affect audience perception.

[0040] Frame Restructuring: When performing high-level media filtering, the media sub-agent restructures the frame, shifting the issue from one dimension to another. For example, economic issues are restructured into security issues, and technological issues into value issues. This frame shift can fundamentally change the way the audience thinks.

[0041] Example 2:

[0042] The difference from Example 1 lies in the fact that, in joint operations and border stability simulations, a layered command structure is constructed: HQ Layer (Command Layer): Responsible for overall situational awareness and operational intent formulation. The HQ intelligent system comprehensively analyzes multiple factors, including the balance of power between enemy and friendly forces, terrain, resource availability, and unforeseen circumstances introduced by the command center, to formulate the overall strategy for this round. This strategy is not a specific action order, but rather an intentional guideline—"What is the main direction of attack? What are the priority targets? What risks need to be avoided?"

[0043] Unit Layer (Execution Layer): Responsible for executing specific tasks. Each unit agent has its own attributes and characteristics—"Is it a reconnaissance unit or a strike unit? What is its current location? What is its resource status? What is its mission?" They receive intent guidance from the HQ layer and make autonomous decisions based on the actual situation on the ground. This "intent-based command" rather than "command-based command" design is closer to the actual situation of modern military command.

[0044] Example 3:

[0045] Unlike Embodiment 2, this system also includes a contingency module to inject several unexpected situations into the simulation process, further increasing the realism of the simulation. Specifically, the unexpected situations mainly include two categories: environmental disturbances and frictional effects. Environmental disturbances: Severe weather affects reconnaissance effectiveness, communication disruptions cause command delays, and civilian traffic congestion slows down resupply. These environmental factors are not pre-set before the simulation but are randomly injected during the simulation, requiring participants to make decisions under conditions of incomplete information and unclear situations.

[0046] Friction Effect: Even the most meticulously planned operations will encounter various "friction" during execution. The contingency module simulates these uncertainties at the execution level—mobile forces may be delayed due to complex terrain, strike operations may fail due to intelligence errors, and coordinated operations may break down due to poor communication. This friction effect makes the simulation results unpredictable and closer to the chaotic characteristics of the real world.

[0047] Example 4:

[0048] The difference from Example 3 is that in hard adversarial simulations, information is not completely transparent. Therefore, a battlefield awareness and information fog mechanism is introduced into the environmental awareness layer agent (each participant has its own perception range, and enemy situations outside the range are "fog"): Reconnaissance Mechanism: Participating parties need to "illuminate" the fog of war through reconnaissance operations. The scope, accuracy, and timeliness of reconnaissance are all constrained by resources, which requires trade-offs. This trade-off is one of the core challenges in tactical decision-making (more reconnaissance resources mean less strike power; an expanded reconnaissance range may decrease accuracy).

[0049] Information Updates: Information obtained through reconnaissance changes over time. Reconnaissance information is not permanently valid; the enemy situation may change over time, and outdated information can lead to misjudgments. Participating parties need to continuously invest reconnaissance resources to maintain real-time awareness of the battlefield situation.

[0050] Deception and Counter-Deception: During the simulation, participants are allowed to employ interference tactics (such as information deception—feints, false targets, electronic jamming, etc.) to mislead the opponent's reconnaissance and judgment. This also trains participants' ability to identify opponent deception and avoid being misled by false information.

[0051] Obviously, the above embodiments are merely illustrative examples for clear explanation and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations here. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.

Claims

1. A multi-agent collaborative system for weapon simulation based on a large model, characterized in that, include: The agent module is used to establish several agents at different levels based on different job functions; Different levels of intelligent agents execute corresponding information flow strategies based on the information processing stage; The hierarchy includes the strategic layer, the execution layer, the environmental awareness layer, and the group layer; the information processing stage includes the information production stage, the information dissemination stage, and the information feedback stage. Strategic layer agents are used for macro-level situation assessment and overall strategy formulation; execution layer agents are used for the implementation of specific tasks; environmental perception layer agents are used for monitoring and feedback on the external environment; and swarm agents are used for representing the collective behavior of a group of people.

2. The weapon simulation multi-agent cooperative system based on a large model according to claim 1, characterized in that, Within the swarm intelligence agent, several sub-intelligent agents of the public with different backgrounds are established; each sub-intelligent agent of the public includes fixed elements such as regional attributes, political orientation, information sources, education level, and core concerns.

3. The weapon simulation multi-agent cooperative system based on a large model according to claim 2, characterized in that, Information flow strategies include: Information production stage: Based on its own position and goals, the strategic layer intelligent agent produces raw information; during the raw information production process, the strategic layer intelligent agent executes the narrative generation reasoning chain and outputs multi-dimensional metadata; Information dissemination stage: In the environmental perception layer, several media sub-agents representing different positions are established; each media sub-agent filters and reconstructs information based on its own editing preferences; the filtered and reconstructed information enters the group dissemination network composed of public sub-agents for cognitive processing; and the filtering and reconstruction follows the behavioral logic of the corresponding media sub-agent. The public's sub-agents combine their own background to judge credibility, compare it with existing cognitive frameworks, and decide whether to accept or spread it; Information feedback phase: The propagation results of the public sub-agents have a reverse impact on the strategic layer agents.

4. The weapon simulation multi-agent cooperative system based on a large model according to claim 3, characterized in that, The attitude changes of each citizen sub-agent affect other agents at other levels associated with the citizen sub-agent through social networks, creating a ripple effect.

5. The weapon simulation multi-agent cooperative system based on a large model according to claim 4, characterized in that, The narrative generation reasoning chain, from top to bottom, includes: Realistic interest analysis: The strategic layer agent first identifies the connection between the current issue and its own core interests, conducts a comprehensive analysis of multi-dimensional information, and assesses the advantages and disadvantages of the current issue for itself. Audience entry point identification: The strategic layer agent analyzes the concerns, information receiving habits and existing cognitive frameworks of different audience groups to find the optimal entry point; Narrative Framework Construction: Based on the results obtained from the analysis of real-world interests and the identification of audience entry points, the strategic layer agent constructs the narrative structure, including the core argument, supporting evidence, keyword system, and expected rebuttal plan; In the process of establishing the narrative structure, the strategic layer agent anticipates potential attack points and prepares defensive arguments in advance for those potential attack points. Self-vulnerability scanning: While generating the narrative structure, the strategic layer agent simultaneously conducts self-criticism to identify hypothetical vulnerabilities in its own discourse that could be exploited by the adversary.

6. The multi-agent cooperative system for weapon simulation based on a large model according to claim 1, characterized in that, When the strategic layer agent generates logic as the counterattacker, it breaks down its own arguments based on the opponent's specific points while generating its own. The specific process is as follows: Adversary discourse analysis: The strategic layer intelligent agent deeply analyzes the adversary's narrative structure and core arguments, and conducts multi-dimensional evaluation and analysis; Counterattack angle selection: Based on the evaluation and analysis results, the strategic layer agent generates several counterattack strategies, evaluates the effect of each counterattack strategy, and selects the optimal counterattack strategy after evaluation. Hedging Narrative Construction: After selecting a counterattack strategy, the strategic layer agent constructs a targeted narrative structure.

7. The weapon simulation multi-agent cooperative system based on a large model according to claim 5, characterized in that, The media sub-agent processes the original narrative as follows: Selective attention: Based on its own positioning, the media sub-agent selectively emphasizes certain information elements; supporting arguments are placed at the top, while unfavorable details are relegated to the end or omitted entirely; and the selection logic follows the media sub-agent's own narrative logic and audience expectations. Emotional tone adjustment: The media sub-agent adjusts the emotional tone of the words used to package the importance of objective facts; Framework Restructuring: When the media sub-agent performs the highest level of media filtering, it restructures the framework, shifting the topic from one dimension to another.

8. The multi-agent cooperative system for weapon simulation based on a large model according to claim 7, characterized in that, Each citizen sub-agent possesses three key elements: persistence and memory, differentiated interpretation, and social influence network. Persistence and Memory: The public intelligent agent is a "fixed individual" that maintains continuity throughout the entire deduction process, recording the attitude, information received, and judgments formed in the previous round of deduction, and updating it based on the memory of the previous round in the new round of deduction; Differentiated interpretation: For the same piece of information, sub-agents from different backgrounds output different interpretations; Social Influence Network: Implicit social influence relationships exist between citizen sub-agents; when a citizen sub-agent decides to "forward" a message, it will affect the message reception of other citizen sub-agents within that citizen sub-agent's social circle.

9. The multi-agent cooperative system for weapon simulation based on a large model according to claim 1, characterized in that, It also includes a contingency module; the contingency module is used to inject several contingency situations during the simulation process.

10. The multi-agent cooperative system for weapon simulation based on a large model according to claim 1, characterized in that, The environmental perception layer intelligent agent introduces battlefield perception and information fog mechanisms: Reconnaissance mechanism: Participating parties need to "illuminate" the fog of war through reconnaissance operations; Information Update: The information obtained through reconnaissance changes accordingly over time; Deception and counter-deception: During the simulation, participating parties are allowed to use interference tactics to mislead the opponent's reconnaissance and judgment.