Simulation system for dynamically building combat resources based on multi-agent system

Abstract

A simulation system for dynamically constructing combat resources based on a multi-agent system includes a combat center server and a multi-agent system. The combat center server includes a command unit, a support unit, an analysis unit, and multiple combat units. The command unit issues simulated combat missions; the support unit generates combat support information, which is referred to as simulated combat information. The analysis unit generates simulated combat formation information based on the available combat resources of each combat unit. The simulated combat formation information includes combat agents and a combat command map. The multi-agent system includes agents corresponding to each combat unit. The multi-agent system accesses the corresponding agents based on the simulated combat formation information and establishes command relationships. This invention achieves efficient resource scheduling and management, enhancing the flexibility and reliability of the system.

Description

Technical Field

[0001] The present invention relates to the technical field of simulation resource management, and more specifically, to a simulation system for dynamically constructing combat resources based on a multi-agent system. Background Art

[0002] Existing simulation combat resource management systems often use static configurations, making them difficult to adapt to the rapidly changing simulated battlefield environment. This results in inflexible resource scheduling and slow response times. Furthermore, combat resource management systems in traditional simulation systems have numerous shortcomings in handling the dynamic access and withdrawal of resources and the dynamic allocation of tasks.

[0003] In existing technologies, centralized management architectures are commonly used in combat resource scheduling systems. This architecture utilizes centralized servers to centrally manage and schedule various resources, but its limitations are obvious. Centralized management systems face significant challenges in scalability and flexibility in the face of rapidly changing battlefield environments. Furthermore, if a centralized system fails, the reliability of the entire system is significantly compromised, creating a high risk of single points of failure.

[0004] For example, existing technology 1 adopts a centralized architecture, which uses a centralized server architecture. All combat resource data and scheduling instructions are processed and managed by one or a few central servers. The servers communicate with various virtual combat units and equipment through the network to obtain real-time data and send scheduling instructions. This implementation method has certain advantages in centralized management and unified scheduling, but it also suffers from problems such as poor system scalability, low resource scheduling efficiency, low system reliability, high data processing pressure, and slow response speed. Existing technology 2 adopts a distributed server architecture. Each distributed node independently manages and schedules local combat resources, while communicating and coordinating with other nodes through the network. Each node has a certain amount of computing and storage capabilities and can independently process and schedule local resources. This implementation method has certain advantages in distributed management and resource scheduling, but it also suffers from problems such as increased system complexity, uneven resource utilization, high data processing pressure, system reliability challenges, and fluctuating response speeds.

[0005] Existing combat resource management systems, whether centralized or distributed, suffer from poor scalability when dealing with rapidly changing battlefield environments. Centralized systems rely on central servers, making them difficult to scale and respond to increasing resource demands or environmental changes. While distributed systems distribute the load, they present significant challenges in communication and coordination between nodes. Adding new nodes or adjusting existing ones is also complex.

[0006] Therefore, the problems existing in the prior art need to be further improved and developed. Summary of the Invention

[0007] (1) Purpose of the invention: To solve the problems existing in the above-mentioned prior art, the purpose of the present invention is to provide a simulation system for dynamically constructing combat resources based on a multi-agent system, which can achieve efficient scheduling and management of resources and enhance the flexibility and reliability of the system.

[0008] (II) Technical Solution: To solve the above technical problems, this technical solution provides a simulation system for dynamically building combat resources based on a multi-agent system, including a combat center server and a multi-agent system.

[0009] The combat center server includes a command unit, a support unit, an analysis unit and multiple combat units. The command unit is used to issue simulated combat missions; the support unit generates combat support information based on the simulated combat missions. The simulated combat missions and combat support information are called simulated combat information; the analysis unit confirms the target combat area based on the simulated combat missions and sends the simulated combat information to the combat units corresponding to the target combat area. The combat units that receive the simulated combat information obtain the available combat resources of the current combat units based on the simulated combat information. The combat resources corresponding to each combat unit include the intelligent agents corresponding to each combat unit; the analysis unit generates simulated combat grouping information based on the available combat resources of each combat unit; the simulated combat grouping information includes combat intelligent agents and combat command maps, and the combat command maps include the command relationship between each combat intelligent agent and the combat unit;

[0010] The multi-agent system includes an agent corresponding to each combat unit; the multi-agent system accesses the corresponding agent according to the simulated combat formation information and establishes a command relationship.

[0011] The simulation system for dynamically constructing combat resources based on a multi-agent system, wherein the combat center server also includes a map construction unit, the map construction unit is used to generate the combat command map, and the map construction unit stores the original combat command map.

[0012] The simulation system for dynamically constructing combat resources based on a multi-agent system, wherein the map construction unit generates a combat command map by implementing the following steps:

[0013] Step 1: Collect and integrate existing data to obtain collected information;

[0014] Step 2: Using natural language processing to perform entity recognition and relationship extraction on the collected information to obtain an updated graph;

[0015] Step three: Match and fuse the updated graph with the original combat command graph through entity alignment and connection to obtain an updated combat command graph.

[0016] The simulation system for dynamically constructing combat resources based on a multi-agent system, wherein existing data collection and integration includes collecting information from structured data and unstructured data, and performing data cleaning and integration;

[0017] Existing data includes manpower, equipment, logistical support and intelligence information. Existing data collection channels include historical combat records, equipment performance data, command preferences and decision-making behavior data.

[0018] The simulation system for dynamically constructing combat resources based on a multi-agent system, wherein entities are specific objects or concepts, including commanders, simulated combat tasks, combat equipment, and support units; relationships are semantic associations between entities, including scheduling, support, and command relationships between entities.

[0019] The simulation system for dynamically constructing combat resources based on a multi-agent system, wherein, when the analysis unit generates simulated combat formation information according to the available combat resources of each combat unit, the entities and relationships in the target combat command map are selected in the updated combat command map, that is, the combat command map generated by the map construction unit, through a hybrid recommendation method of the resource recommendation model. Specifically, the analysis unit maps the entities and relationships in the updated combat command map to a low-dimensional vector space, and then the analysis unit uses a machine learning algorithm to calculate and recommend similarity between the available combat resources of each combat unit and the vectors corresponding to the entities and relationships in the updated combat command map mapped to the low-dimensional vector space to obtain the best resource plan, that is, the target combat command map; the combat command map in the simulated combat formation information is the target combat command map.

[0020] The simulation system for dynamically constructing combat resources based on a multi-agent system, wherein the analysis unit includes a resource recommendation model, and the resource recommendation model includes a task judgment layer, a multi-branch resource feature extraction module and a prediction layer. The task judgment layer is used to judge the task requirements of the current simulated combat task, the multi-branch resource feature extraction module is used to extract resource features related to the simulated combat task from the combat command map and / or the updated combat command map, and the prediction layer recommends resources based on the extracted resource features.

[0021] The simulation system for dynamically constructing combat resources based on a multi-agent system, wherein the analysis unit also includes constructing a target resource recommendation model based on a resource recommendation model and performing resource recommendation, the specific process includes:

[0022] In step A, the analysis unit obtains historical combat resource interaction data and updates the combat command map, trains the resource recommendation model, optimizes the model parameters, and obtains a first recommendation model;

[0023] In step B, the analysis unit verifies the resource recommendation effect of the first recommendation model based on historical combat resource interaction data and updated combat command maps; when the resource recommendation effect reaches the first effect, the first recommendation model is the target recommendation model; when the resource recommendation effect does not reach the first effect, continue to execute step A to train the resource recommendation model and optimize the model parameters.

[0024] In step C, the analysis unit recommends resources through a target recommendation model to obtain simulated combat formation information.

[0025] The simulation system for dynamically constructing combat resources based on a multi-agent system, wherein in step A, the analysis unit uses historical combat resource interaction data to train a resource recommendation model, adjusts the position of the entity, and obtains a first recommendation model.

[0026] The simulation system for dynamically constructing combat resources based on a multi-agent system, wherein the analysis unit analyzes the multi-hop relationships between different entities based on the entities and relationships in the updated combat command map, calculates the path scores between the commander and the recommended resources, and selects resource candidate entities with high correlation.

[0027] (3) Beneficial effects: The present invention provides a simulation system for dynamically constructing combat resources based on a multi-agent system, which solves the problems of poor scalability, low resource scheduling efficiency, and low reliability existing in the prior art, realizes efficient scheduling and management of resources, and enhances the flexibility and reliability of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 This is a schematic diagram of the structure of a simulation system for dynamically constructing combat resources based on a multi-agent system according to the present invention;

[0029] Figure 2 This is a schematic diagram of the steps of generating a combat command map by a map construction unit of a simulation system for dynamically constructing combat resources based on a multi-agent system according to the present invention;

[0030] Figure 3 It is a schematic diagram of the steps of constructing a target resource recommendation model based on a resource recommendation model and performing resource recommendation by an analysis unit of a simulation system for dynamically constructing combat resources based on a multi-agent system of the present invention. DETAILED DESCRIPTION

[0031] The present invention is further described in detail below in conjunction with preferred embodiments. More details are set forth in the following description to facilitate a full understanding of the present invention. However, the present invention can obviously be implemented in a variety of other ways different from the description. Those skilled in the art can make similar generalizations and deductions based on actual application situations without violating the connotation of the present invention. Therefore, the scope of protection of the present invention should not be limited by the content of this specific embodiment.

[0032] The accompanying drawings are schematic diagrams of embodiments of the present invention. It should be noted that the drawings are merely examples and are not drawn to scale, and should not be used to limit the actual scope of protection claimed in the present invention.

[0033] A simulation system based on a multi-agent system to dynamically build combat resources, such as Figure 1 The method includes a combat center server and a multi-agent system. The combat center server generates simulated combat grouping information according to the simulated combat mission. The multi-agent system accesses the corresponding agent according to the simulated combat grouping information and establishes a command relationship.

[0034] The combat center server includes a command unit, a support unit, an analysis unit, and multiple combat units. The command unit is used to issue simulated combat missions, and the support unit generates combat support information based on the simulated combat missions. The simulated combat missions and combat support information are referred to as simulated combat information. The analysis unit confirms the target combat area based on the simulated combat mission and sends the simulated combat information to the combat unit corresponding to the target combat area. The combat unit that receives the simulated combat information obtains the available combat resources of the current combat unit based on the simulated combat information. The combat resources corresponding to each combat unit include the intelligent agent corresponding to the corresponding combat unit. The analysis unit generates simulated combat formation information based on the available combat resources of each combat unit. The simulated combat formation information includes combat intelligent agents and a combat command map. The combat command map includes the command relationship between each combat intelligent agent and the combat unit.

[0035] The combat center server further includes a map construction unit, which is used to generate the combat command map. The map construction unit may store an original combat command map.

[0036] When the map construction unit generates the combat command map, Figure 2 As shown, it can be achieved by the following steps:

[0037] Step 1: Collect and integrate existing data to obtain collected information. This involves collecting information from both structured and unstructured data, and then cleaning and integrating the data. Existing data includes information on manpower, equipment, logistics support, and intelligence. Existing data collection channels include historical combat records, equipment performance data, command preferences, and decision-making behavior data.

[0038] Step 2: Use natural language processing to perform entity recognition and relationship extraction on the collected information to obtain an updated graph. Entities are specific objects or concepts, such as commanders, simulated combat missions, combat equipment, and support units. Relationships are semantic associations between entities, such as scheduling, support, and command relationships between entities.

[0039] Step three: Match and merge the updated graph with the original battle command graph through entity alignment and connection, resulting in an updated battle command graph. At this point, the updated battle command graph is the battle command graph generated by the graph construction unit. Specifically, the graph construction unit uses the original battle command graph as a basis and, based on the connection relationships between entities in the updated graph, places the entities in the updated graph into the original battle command graph. Furthermore, based on the relationships between entities in the original battle command graph and the updated graph, it updates the relationships between entities.

[0040] When the analysis unit generates the simulated combat formation information based on the available combat resources of each combat unit, it generates the simulated combat formation information based on the available combat resources of each combat unit in the updated combat command map. In this case, the combat command map in the simulated combat formation information is the target combat command map.

[0041] When the analysis unit generates simulated combat formation information based on the available combat resources of each combat unit, it can select entities and relationships in the target combat command map in the updated combat command map (that is, the combat command map generated by the map construction unit) through a hybrid recommendation method of the resource recommendation model. Specifically, the analysis unit maps the entities and relationships in the updated combat command map to a low-dimensional vector space, and then uses a machine learning algorithm to perform similarity calculation and recommendation on the available combat resources of each combat unit and the vectors corresponding to the entities and relationships in the updated combat command map mapped to the low-dimensional vector space to obtain the optimal resource plan, that is, the target combat command map.

[0042] The analysis unit selects the combat unit's available combat resources as recommended resources based on the similarity calculation results. Similarity calculation can be performed using cosine similarity, Euclidean distance, Manhattan distance, Jaccard similarity coefficient, etc. Here, cosine similarity calculation is used as an example for illustration. The analysis unit sorts the similarity calculation results from high to low based on the available combat resources of each combat unit and the similarity calculation results of the corresponding vectors of entities and relationships mapped to the low-dimensional vector space in the updated combat command map. The analysis unit selects the combat resources of the combat unit with the highest similarity in the similarity calculation results as the recommended resources. For example, the analysis unit selects the combat resources of the combat unit with the smallest difference between 1 and the calculation result as the recommended resources.

[0043] Then, the analysis unit associates the recommended resources according to the entities and relationships in the updated combat command map to obtain the optimal resource solution.

[0044] The analysis unit includes a resource recommendation model, which includes a task judgment layer, a multi-branch resource feature extraction module and a prediction layer. The task judgment layer is used to judge the task requirements of the current simulated combat task. The multi-branch resource feature extraction module is used to extract resource features related to the simulated combat task from the combat command map and / or the updated combat command map. The prediction layer recommends resources based on the extracted resource features.

[0045] The analysis unit also includes building a target resource recommendation model based on the resource recommendation model and performing resource recommendation, such as Figure 3 As shown, the specific process includes:

[0046] In step A, the analysis unit obtains historical combat resource interaction data and updates the combat command map, trains the resource recommendation model, optimizes the model parameters, and obtains the first recommendation model.

[0047] The combat resource interaction data includes the specific resources dispatched and used by commanders in different mission scenarios. The analysis unit obtains recommendation preference information based on historical combat resource interaction data; updates the combat command map, including the relationship between combat resources, and describes in detail the resource information at the support, reconnaissance, firepower, protection and other levels in the form of entity-relationship-entity triples.

[0048] In step B, the analysis unit verifies the resource recommendation effectiveness of the first recommendation model based on historical combat resource interaction data and the updated combat command map. If the resource recommendation effectiveness reaches the first effect, the first recommendation model becomes the target recommendation model. If the resource recommendation effectiveness does not reach the first effect, step A is continued to train the resource recommendation model and optimize the model parameters. The first effect is a preset effect, which can be a specified expected value or a desired effect.

[0049] In step C, the analysis unit recommends resources through a target recommendation model to obtain simulated combat formation information.

[0050] When the analysis unit obtains an updated combat command map, it maps the entities and relationships in the updated map into a low-dimensional vector space. Using machine learning algorithms, the analysis unit captures the deep connections between entities, providing rich semantic information for recommending the analysis unit. In step A, the analysis unit uses historical combat resource interaction data to train a resource recommendation model, adjusting the positions of entities—that is, optimizing model parameters—to bring preferred entities of the same commander (combat unit) closer together in vector space, thereby improving the accuracy of recommendations.

[0051] The analysis unit uses a machine learning algorithm to capture the deep associations between entities and provide rich semantic information for the analysis unit. Specifically, it can be performed through a graph neural network algorithm for association analysis and inference, and supplemented by association rules and support vector machines to provide additional rules or judgment basis; it can also be processed through graph neural networks and processing rules to process complex multi-hop relationships and entity associations to achieve recommendations, that is, using graph neural networks for relationship inference and supplemented by association rules or other traditional methods. For simpler resource recommendation tasks, that is, the principle is simple and the application scenarios of resource recommendation are generally not complex, a support vector machine can be used.

[0052] Based on the entities and relationships in the updated combat command map, the analysis unit analyzes the multi-hop relationships between different entities, calculates the path score between commanders and recommended resources, and selects candidate resource entities with high correlation. The path score between commanders and recommended resources reflects the complex relationships between entities and supports personalized recommendations.

[0053] The analysis unit combines the low-dimensional vectors and path scores of the entities and relationships in the updated combat command map to achieve hybrid resource recommendation. Specifically, the analysis unit selects candidate resources by updating the low-dimensional vectors of the entities and relationships in the combat command map, and then sorts and filters them according to the path scores between the commander and the recommended resources to select the best resource solution, thereby achieving more comprehensive and accurate resource recommendations.

[0054] The path score between the commander and the recommended resource can be that the recommended resource is marked with a score according to the distance between the commander and the recommended resource. The higher the score, the farther the distance between the recommended resource and the commander. When sorting and filtering according to the path score between the commander and the recommended resource, the lower the path score, the better the corresponding resource plan.

[0055] The analysis unit uses historical combat resource interaction data and updated combat command maps to train the resource recommendation model, uses verification data to evaluate the recommendation effect of the resource recommendation model, performs performance evaluation through indicators such as accuracy, recall rate and F1 score, optimizes model parameters, and improves the effectiveness and reliability of the recommendation system.

[0056] For the combat resource interaction data including specific resources dispatched and used by commanders in different mission scenarios, the analysis unit obtains recommendation preference information based on historical combat resource interaction data. The following example is given here.

[0057] a) Commander Type 1-99 Main Battle Tank (Tank)

[0058] b) Commander 2-XXX combat satellites (satellites)

[0059] c) Commander 3-XX 122mm self-propelled howitzer (artillery)

[0060] d) Commander 4 - Man-Portable Air Defense Missile (AAM)

[0061] The updated combat command map includes the relationships between combat resources. The resource information at the support, reconnaissance, firepower, and protection levels is described in detail in the form of entity-relationship-entity triples. The following examples are given here.

[0062] a) Satellite-monitoring-technology position

[0063] b) Artillery-Support-Infantry Fighting Vehicles

[0064] c) Air Defense Missiles - Protecting Bases

[0065] d) Logistics Support Battalion - Maintenance - Tanks

[0066] The details of the interactive data of combat resources and auxiliary information of the combat command map are shown in Table 1, where the ratio of training, verification and testing is 6:2:2.

[0067]

[0068] Table 1 Statistics of the dataset

[0069] The following describes the resource recommendations for combat agents in different tasks.

[0070] During reconnaissance missions, commanders need to select appropriate reconnaissance resources, such as drones, ground radar, and satellites. A knowledge graph-based resource recommendation model, based on the specific requirements of the reconnaissance mission and the commander's historical decision-making behavior, recommends the optimal combination of reconnaissance resources, improving the efficiency and accuracy of reconnaissance missions.

[0071] During a fire strike mission, commanders need to coordinate various firepower resources, such as artillery, missiles, and aircraft. A knowledge graph-based resource recommendation model can combine the firepower resource information in the knowledge graph to recommend the most appropriate combination of firepower resources based on the specific requirements of the fire strike mission, ensuring the effectiveness and accuracy of the fire strike.

[0072] During logistics support missions, commanders must dispatch various logistical resources, such as supply vehicles, maintenance teams, and medical teams. The timeliness and accuracy of logistics support are crucial to the success of simulated combat missions. The knowledge graph-based resource recommendation model, based on the specific requirements of the logistics support mission and the logistics resource information in the knowledge graph, recommends the optimal combination of logistics resources, thereby improving the efficiency and effectiveness of logistics support.

[0073] Combat resource interaction data can be multimodal data, which can be used to build a more comprehensive and accurate combat command map by integrating multiple data types such as images, videos, and text.

[0074] The multi-agent system includes an agent corresponding to each combat unit. That is, all agents within each combat unit constitute the multi-agent system. Each agent includes a control device, and each agent is labeled with combat capabilities. Each agent's control device enables access to and from the multi-agent system, information transmission, and combat operations.

[0075] The multi-agent system broadcasts access protocol data based on simulated combat formation information, and requests the corresponding agent that needs access to match the access rules after receiving the data, confirms the legitimacy and necessity of the request, and completes the establishment of a direct connection channel for the agent after successful matching, ensuring that the accessed agent obtains task information and command instructions in a timely manner, and generates an intelligent agent, which is used to control the task performance of the corresponding agent and the data feedback of the agent.

[0076] Access matching rules include the required data protocol (ICD), security protocol, and access condition rules. Physical matching methods include data decryption, and establishing a communication channel uses the handshake protocol.

[0077] The access condition rule may include access under the condition of being discovered by the enemy, access time, etc.

[0078] When all combat agents in the simulated combat formation information complete the establishment of direct connection channels, all combat agents in the simulated combat formation information form a multi-agent network structure corresponding to the current simulated combat mission.

[0079] When the current simulated combat mission is completed or the combat agent is damaged, the multi-agent system manages the exit of the combat agent and updates the agent network structure. Specifically, when the current simulated combat mission is completed, the multi-agent system manages the exit of all combat agents in the simulated combat formation information; when the current simulated combat mission is not completed and the connected combat agent is damaged, the multi-agent system manages the exit of the combat agent, connects a replaceable combat agent, and updates the multi-agent network structure corresponding to the current simulated combat mission.

[0080] When generating simulated combat formation information, the analysis unit of the operations center server simultaneously creates a backup list for each agent. The multi-agent system calculates the matching degree of the backup agents, i.e., their task completion capabilities, in real time. When an agent withdraws, the highest-priority agent in the corresponding backup list is selected to send a connection request. If the response received is unsatisfactory, the task request is reissued. Unsatisfactory responses may include, for example, that the backup unit page is damaged or has been assigned to other tasks.

[0081] After the analysis unit identifies the target combat zone based on the simulated combat mission, it sends the simulated combat information to the combat units corresponding to all agents in the target combat zone. The combat units that receive the simulated combat information then determine the available combat resources for their respective units, specifically all agents in the target combat zone. All agents in the target combat zone are then connected to a support resource pool. The multi-agent system then connects to the corresponding agents based on the simulated combat mission and combat zone, establishing a command relationship.

[0082] When generating simulated combat formation information, the analysis unit divides the work among multiple intelligences based on rules and laws derived from domain knowledge and experience, as well as target resource recommendation models generated through historical data learning and training. For example, three drones each performed an aerial survey mission along an assigned route, collectively mapping a 40km x 40km area.

[0083] After confirming the target combat area based on the simulated combat mission, the analysis unit sends simulated combat information to combat units with fixed combat resources within the target combat area. The receiving combat units then use this information to determine their available combat resources, including all agents under the receiving combat unit. The analysis unit generates simulated combat formation information based on each combat unit's available combat resources. Fixed combat resources include launch sites, concealed standby positions, fuel depots, and military depots.

[0084] The analysis unit can also match support combat resources among all combat units based on the combat unit's request. Based on the required support combat resources, the analysis unit selects the combat resource corresponding to the support combat resource closest to the target combat area from all combat units as the target support combat resource. The analysis unit adds the target support combat resource to the simulated combat formation information.

[0085] The multi-agent system monitors the status of corresponding agents in the simulated combat formation information in real time, including their availability, location, and performance. If the availability, location, and performance of an agent fail to meet the requirements of the wall-retaining combat, the corresponding combat unit requests matching support combat resources from the analysis unit. At this point, the combat unit requests matching support combat resources from the analysis unit.

[0086] Each agent connected to the multi-agent network structure corresponding to the current simulated combat mission feeds back mission data, status information, etc. to the multi-agent system in real time, realizing information sharing between agents, as follows:

[0087] Information demand analysis: The multi-agent system analyzes the type and content of information that needs to be shared between agents based on the simulated combat mission, and obtains sharing demand information;

[0088] Information sharing plan generation: The multi-agent system generates an information sharing plan based on the sharing demand information, including the information sharing mechanism and sharing channel;

[0089] Access request broadcast: each agent sends an access request to the multi-agent system by broadcasting access protocol data according to the information sharing scheme;

[0090] Access request matching: After receiving the access request, the multi-agent system performs rule matching to confirm the legitimacy and necessity of the request;

[0091] Establishment of information sharing channel: After the multi-agent system and the agent rules are successfully matched, the multi-agent system establishes an information sharing channel to ensure that information can be shared in a timely manner;

[0092] Intelligent agent generation: The multi-agent system generates a corresponding intelligent agent based on the access request, and the intelligent agent is responsible for the execution of information sharing tasks and data feedback.

[0093] Access request matching includes:

[0094] Agent verification request: After receiving the access request, the agent performs verification, including identity verification and permission verification;

[0095] Matching access rules: The agent matches the pre-set access rules based on the information in the access request to confirm whether access is allowed.

[0096] After the intelligent agent of the multi-agent network structure corresponding to the current simulated combat mission completes the access, the intelligent agent in the multi-agent system controls the intelligent agent to complete the combat mission, and according to the completion status of the combat mission, the intelligent agent that completes the simulated combat mission will exit accordingly to realize the update of the multi-agent network structure.

[0097] After receiving the data, the multi-agent system requests the corresponding agent to access, and the agent performs self-evaluation to determine whether the access conditions are met. When the access conditions are met, the agent accesses the multi-agent system and becomes a part of the system. The specific steps are as follows:

[0098] Access Request Transmission: The multi-agent system sends an access request containing information such as the current mission requirements, the roles and responsibilities of the agents, etc. The access request is sent via a wireless or wired network to ensure that all agents receive the access request. The current mission requirements include information such as the functions, capabilities, and location of the required agents. For example, if the simulated combat mission is a reconnaissance mission, the multi-agent system requires a drone with a high-resolution camera and long flight time. In this case, the access request includes these specific requirements.

[0099] Self-assessment: After receiving the request, the agent to be connected conducts a self-assessment to determine whether it meets the access conditions. The self-assessment includes the agent's capabilities, current status, and location. The agent's capabilities can be sensor type and performance, its current status can be battery charge and consumption, and its location can be whether it is within the mission area.

[0100] Establishing a connection: When the self-assessment result of the agent meets the conditions, the agent establishes a connection with the multi-agent system, thereby ensuring that the multi-agent system and the agent can receive instructions and / or feedback data in real time; after the multi-agent system and the agent establish a connection, it is necessary to ensure the stability and security of the connection between the multi-agent system and the agent to prevent data leakage or connection interruption;

[0101] Generate intelligent agents: The multi-agent system generates intelligent agents for the connected agents. The intelligent agent corresponding to each agent is responsible for its task execution and data management. The intelligent agent coordinates the actions of the agents and monitors their status based on their functions and simulated combat mission requirements. Specifically, it optimizes the work paths of the agents, assigns specific tasks, and collects feedback data. For example, in reconnaissance missions, the intelligent agent commands the drone's flight path, shooting angle, and image transmission frequency.

[0102] When the current simulated combat mission is completed or the combat agent is damaged, the multi-agent system manages the combat agent to exit and updates the agent network structure. The agent exit includes monitoring the agent status, issuing an exit request, and updating the system network structure, specifically including the following steps:

[0103] Status monitoring: The multi-agent system monitors the status of connected agents in real time. When it detects that the simulation task is completed or the agent fails, it triggers the agent to exit. Monitoring the status of connected agents includes detecting the battery level, sensor status, communication quality, etc. of the agent. When the multi-agent system detects an abnormal state of the agent, such as low battery level or sensor failure, it triggers the exit procedure.

[0104] Sending an exit request: The multi-agent system sends an exit request to the corresponding agent, requesting the agent to exit the current task. The exit request includes exit steps and instructions to ensure that the agent exits the agent network structure. The agent's exit from the current task may be required to return to the base, turn off sensors, disconnect communication connections, etc. The exit request includes basic information about the agent and the reason for exiting.

[0105] Disconnection: After receiving the exit request, the agent disconnects from the multi-agent system. When the multi-agent system and the agent are disconnected, the security of the disconnection between the multi-agent system and the agent needs to be ensured to prevent accidental disconnection or data loss. Specifically, before disconnection, the agent can upload all task data to the multi-agent system and confirm that the data is complete and correct.

[0106] Updating the Agent Network Structure: The multi-agent system updates the agent network structure and reallocates tasks to ensure that other agents can continue to execute the simulated combat mission. Updating the agent network structure includes removing exiting agents, reestablishing the chain of command, and adjusting resource allocation. For example, when an agent exits, the multi-agent system assigns its tasks to other agents to ensure mission continuity and integrity.

[0107] When the agents establish a connection, a link16 data link may be used to ensure the stability and security of the connection between the multi-agent system and the agents to prevent data leakage or connection interruption.

[0108] When the agent is disconnected, the multi-agent system performs exit confirmation, including task completion confirmation, resource damage confirmation, etc.

[0109] When the agent completes the exit, the multi-agent system releases the agent's resources and the agent exits the combat system.

[0110] The automatic entry and exit of agents enables the multi-agent system to quickly adapt to changes, improving flexibility and scalability. This not only addresses the scalability and single-point failure limitations of traditional multi-agent systems, but also ensures the system's efficient adaptation to changing battlefield environments through dynamic agent adjustments. Dynamic agent entry allows the system to flexibly increase or decrease the number of agents based on simulated combat missions, while the automatic exit of agents ensures the system promptly removes inactive or damaged agents, maintaining efficient operation.

[0111] The following is an explanation of different simulated combat missions.

[0112] When the simulated combat mission is a reconnaissance mission, high-resolution imagery and real-time video transmission are often required. Through the dynamic addition and removal of agents, the multi-agent system flexibly deploys drones equipped with these capabilities based on mission requirements. For example, at the start of a mission, the multi-agent system sends an access request, allowing a drone with a high-resolution camera and long endurance to connect to the system and perform the reconnaissance mission. Upon mission completion or when a drone's battery runs low, the multi-agent system sends an exit request, ensuring the drone's safe return and disconnecting. Simultaneously, the multi-agent system can connect new drones, ensuring the continuity and effectiveness of the reconnaissance mission.

[0113] When the simulated combat mission involves a natural disaster or battlefield rescue mission, the multi-agent system needs to quickly mobilize different types of rescue equipment and personnel. By dynamically switching agents on and off, the multi-agent system flexibly deploys rescue resources based on on-site needs. For example, during earthquake rescue, the multi-agent system integrates intelligent robots equipped with life detectors and demolition tools to conduct rubble search and rescue operations. When a robot completes its mission or requires maintenance, the multi-agent system sends a request to exit and connects a new robot, ensuring the continuity and efficiency of the rescue effort.

[0114] When the simulated combat mission involves intelligence analysis, the multi-agent system needs to process large amounts of data and extract key information. By dynamically switching agents in and out, the multi-agent system flexibly deploys different types of analysis tools and algorithms based on data processing needs. For example, when analyzing enemy communications data, the multi-agent system connects agents with speech recognition and natural language processing capabilities to perform data analysis. When the analysis task is completed or an algorithm update is required, the multi-agent system sends a request to exit and connects to a new analysis tool, ensuring the accuracy and timeliness of intelligence analysis.

[0115] The above practical application scenarios demonstrate the significant advantages of automated agent entry and exit in improving system flexibility and scalability. This approach not only addresses the scalability and single-point-of-failure limitations of traditional systems, but also ensures that multi-agent systems can efficiently adapt to diverse complex environments and task requirements through dynamic agent adjustments.

[0116] The multi-agent system and each agent are respectively provided with a control selection device, and the control selection device includes an input device and a display device. When an agent enters or exits the multi-agent system, the user confirms or selects the entry and exit operations through the control selection device of the multi-agent system and each agent respectively. For example, during the agent access process, after the multi-agent system sends a broadcast access request, the user performs identity authentication and authority verification on the agent, and only after confirmation can the agent access be performed; during the agent exit process, after the multi-agent system broadcasts an exit request, the user performs task completion confirmation and resource damage confirmation on the agent, and only after confirmation can the agent enter or exit. Thus, it adapts to combat scenarios with higher security and accuracy.

[0117] When agents join or leave the multi-agent system, they can also adopt a centralized joining and distributed leaving scheme. Specifically, during the agent joining process, the combat center server performs unified management and scheduling. During the agent leaving process, the corresponding combat unit of the agent is independently managed and scheduled. This improves the efficiency of joining to a certain extent while ensuring the flexibility of leaving. The specific process is as follows:

[0118] Centralized access: During the access process of the agent, the multi-agent system sends an access request to the combat center server. The combat center server verifies and matches the access rules, and after confirmation, connects the agent to the multi-agent system.

[0119] Distributed exit: During the exit process, the agent sends an exit request to the corresponding combat unit. After the combat unit confirms the exit request, it updates the agent network structure and completes the agent exit.

[0120] A simulation system for dynamically constructing combat resources based on a multi-agent system realizes real-time dynamic management of combat resources through the automatic entry and exit of agents, thereby improving combat response speed; the flexibility of the system is enhanced by the ability of agents to autonomously join or exit the combat system according to combat needs; the use of a multi-agent system avoids the bottleneck problem of centralized control and improves the overall processing capability of the system; the communication between agents adopts an efficient protocol, which reduces the communication overhead of the system; the collaborative work between agents is carried out according to pre-set rules, which simplifies the complexity of collaborative work; and the automation of the entry and exit process of agents reduces the maintenance cost of the system.

[0121] The above content is an explanation of the preferred embodiments of the present invention, which can help those skilled in the art to more fully understand the technical solutions of the present invention. However, these embodiments are merely illustrative, and it cannot be determined that the specific implementation methods of the present invention are limited to the description of these embodiments. For those skilled in the art of the present invention, without departing from the concept of the present invention, several simple deductions and transformations can be made, which should be deemed to fall within the scope of protection of the present invention.

Claims

1. A simulation system for dynamically building combat resources based on a multi-agent system, characterized by: Including combat center server and multi-agent system, The combat center server includes a command unit, a support unit, an analysis unit and multiple combat units. The command unit is used to issue simulated combat missions; the support unit generates combat support information based on the simulated combat missions. The simulated combat missions and combat support information are called simulated combat information; the analysis unit confirms the target combat area based on the simulated combat missions and sends the simulated combat information to the combat units corresponding to the target combat area. The combat units that receive the simulated combat information obtain the available combat resources of the current combat units based on the simulated combat information. The combat resources corresponding to each combat unit include the intelligent agents corresponding to each combat unit; the analysis unit generates simulated combat grouping information based on the available combat resources of each combat unit; the simulated combat grouping information includes combat intelligent agents and combat command maps, and the combat command maps include the command relationship between each combat intelligent agent and the combat unit; The multi-agent system includes an agent corresponding to each combat unit; The multi-agent system accesses the corresponding agent according to the simulated combat formation information and establishes a command relationship; When the analysis unit generates simulated combat formation information based on the available combat resources of each combat unit, the entities and relationships in the target combat command map are selected in the updated combat command map through the hybrid recommendation method of the resource recommendation model. Specifically, the analysis unit maps the entities and relationships in the updated combat command map to a low-dimensional vector space, and then uses a machine learning algorithm to perform similarity calculation and recommendation on the available combat resources of each combat unit and the vectors corresponding to the entities and relationships mapped to the low-dimensional vector space in the updated combat command map to obtain the best resource plan, that is, the target combat command map; the combat command map in the simulated combat formation information is the target combat command map; the updated combat command map is the combat command map generated by the map construction unit; The analysis unit includes a resource recommendation model. The analysis unit constructs a target resource recommendation model based on the resource recommendation model and performs resource recommendation. The specific process includes: In step A, the analysis unit obtains historical combat resource interaction data and an updated combat command map, trains a resource recommendation model, optimizes model parameters, and obtains a first recommendation model. When the analysis unit obtains the updated combat command map, it maps entities and relationships in the updated combat command map into a low-dimensional vector space. The analysis unit uses a machine learning algorithm to capture deep associations between entities and provide semantic information for the analysis unit. In step A, the analysis unit uses the historical combat resource interaction data to train the resource recommendation model and adjust the positions of entities so that preferred entities of the same combat unit are closer in the vector space. In step B, the analysis unit verifies the resource recommendation effect of the first recommendation model based on the historical combat resource interaction data and the updated combat command map; when the resource recommendation effect reaches the first effect, the first recommendation model becomes the target recommendation model; when the resource recommendation effect does not reach the first effect, step A is continued to train the resource recommendation model and optimize the model parameters; Step C, the analysis unit recommends resources through a target recommendation model to obtain simulated combat formation information; The analysis unit uses a machine learning algorithm to capture the deep associations between entities and provide the analysis unit with rich semantic information, specifically including performing association analysis and inference through a graph neural network algorithm, and supplementing it with association rules and support vector machines to provide additional rules or judgment basis; or processing complex multi-hop relationships and entity associations through graph neural networks and processing rules to achieve recommendations, that is, using graph neural networks to perform relationship inference, and supplementing it with association rules or support vector machines to provide additional rules or judgment basis.

2. The simulation system for dynamically constructing combat resources based on a multi-agent system according to claim 1, characterized in that: The combat center server further includes a map construction unit, which is used to generate the combat command map. The map construction unit stores an original combat command map.

3. The simulation system for dynamically constructing combat resources based on a multi-agent system according to claim 2, characterized in that: When the map construction unit generates the combat command map, it is implemented by the following steps: Step 1: Collect and integrate existing data to obtain collected information; Step 2: Using natural language processing to perform entity recognition and relationship extraction on the collected information to obtain an updated graph; Step three, matching and fusing the updated graph with the original combat command graph through entity alignment and connection to obtain an updated combat command graph; the updated combat command graph is the combat command graph generated by the graph construction unit, including that the graph construction unit takes the original combat command graph as the basis, places the entities in the updated graph in the original combat command graph according to the connection relationship of the entities in the updated graph, and updates the relationship between entities according to the relationship between entities in the original combat command graph and the updated graph.

4. The simulation system for dynamically constructing combat resources based on a multi-agent system according to claim 3, characterized in that: Existing data collection and integration includes collecting information from structured and unstructured data, and performing data cleaning and integration; Existing data includes manpower, equipment, logistical support and intelligence information. Existing data collection channels include historical combat records, equipment performance data, command preferences and decision-making behavior data.

5. The simulation system for dynamically constructing combat resources based on a multi-agent system according to claim 3, characterized in that: Entities are specific objects or concepts, including commanders, simulated combat missions, combat equipment, and support units; relationships are semantic associations between entities, including scheduling, support, and command relationships between entities.

6. The simulation system for dynamically constructing combat resources based on a multi-agent system according to claim 1, characterized in that: The analysis unit uses a machine learning algorithm to perform similarity calculation and recommendation. Specifically, the similarity calculation uses cosine similarity, Euclidean distance, Manhattan distance, or Jaccard similarity coefficient. The analysis unit sorts the similarity calculation results from high to low according to the available combat resources of each combat unit and the similarity calculation results of the corresponding vectors of entities and relationships mapped to the low-dimensional vector space in the updated combat command map. The analysis unit selects the combat resources of the combat unit with the highest similarity in the similarity calculation results as the recommended resources; Then, the analysis unit associates the recommended resources according to the entities and relationships in the updated combat command map to obtain the optimal resource solution.

7. The simulation system for dynamically constructing combat resources based on a multi-agent system according to claim 6, characterized in that: The resource recommendation model includes a task judgment layer, a multi-branch resource feature extraction module and a prediction layer. The task judgment layer is used to judge the task requirements of the current simulated combat task. The multi-branch resource feature extraction module is used to extract resource features related to the simulated combat task from the combat command map and / or the updated combat command map. The prediction layer recommends resources based on the extracted resource features.

8. The simulation system for dynamically constructing combat resources based on a multi-agent system according to claim 6, characterized in that: The analysis unit uses historical combat resource interaction data and updated combat command maps to train the resource recommendation model, uses verification data to evaluate the recommendation effect of the resource recommendation model, evaluates performance through accuracy, recall rate and F1 score indicators, and optimizes model parameters.

9. The simulation system for dynamically constructing combat resources based on a multi-agent system according to claim 8, characterized in that: In step A, the analysis unit trains the resource recommendation model using historical combat resource interaction data and adjusts the position of the entity to obtain a first recommendation model.

10. The simulation system for dynamically constructing combat resources based on a multi-agent system according to claim 8, characterized in that: The analysis unit analyzes the multi-hop relationships between different entities based on the entities and relationships in the updated combat command map, calculates the path scores between the commander and the recommended resources, and selects resource candidate entities with high correlation; The analysis unit combines the low-dimensional vectors and path scores of the entities and relationships in the updated combat command map to implement hybrid resource recommendation. Specifically, the analysis unit selects candidate resources by updating the low-dimensional vectors of the entities and relationships in the combat command map. Then, the candidate resources are sorted and screened based on the path scores between the commander and the recommended resources to select the best resource solution. Recommended resources are scored according to the distance between them and the commander. The higher the score, the farther the distance between the recommended resource and the commander. When sorting and filtering based on the path score between the commander and the recommended resources, the lower the path score, the better the corresponding resource plan.

Images