A capability center driving-based multi-agent overlapping alliance optimization method

Abstract

The present application relates to the technical field of multi-agent system cooperation, and particularly relates to a multi-agent overlapping alliance optimization method based on a capability center drive, which comprises the following steps: collecting a plurality of agents around a given task agent to obtain a total available capability pool in a region; if it is determined that the total available capability pool in the region meets the demand, constructing an alliance for the given task agent and the plurality of agents around the given task agent; screening an alliance that meets the task demand and has the optimal system efficiency; implementing dynamic resource allocation on the optimal alliance, and updating the resource state of the agent in real time; when a new task arrives, the remaining resources of the agents in the constructed alliance are deducted, and a new round of alliance construction is performed. The present application focuses on dynamic task allocation and cooperation optimization of multi-agents, and is particularly suitable for collaborative work and resource management of heterogeneous agents in a complex and topologically time-varying environment.

Description

A Capability Center-Driven Multi-Agent Overlapping Coalition Optimization Method Technical Field

[0001] This invention relates to the field of multi-agent system collaboration technology, specifically to a multi-agent overlapping alliance optimization method based on capability center-driven approach. Background Technology

[0002] Multi-agent collaboration has demonstrated significant value in complex scenarios such as disaster relief, intelligent transportation, and drone swarm operations. These fields rely on heterogeneous agents to achieve efficient cooperation through complementary capabilities to address the challenges of limited resources and diverse tasks. Alliance formation, as a key link in multi-agent collaboration, determines the rationality of task allocation and system efficiency. However, current research largely focuses on the formation of disjoint (or isolated) alliances, creating stable structures by assigning agents to non-overlapping alliances. This approach is limited by agents participating in only a single alliance, leading to low resource utilization, especially in multi-task parallel scenarios where it's difficult to fully leverage the potential of heterogeneous agents. Overlapping alliance formation thus becomes an important alternative, allowing agents to participate in multiple alliances simultaneously, significantly improving resource utilization efficiency and the system's flexibility and robustness in complex environments.

[0003] In recent years, preliminary progress has been made in the field of multi-agent overlapping alliance formation. Xue Shuxin, Ma Yajie, Jiang Bin, et al. ("A Distributed Task Allocation Algorithm for Heterogeneous UAV Swarms Based on Alliance Formation Game Theory," Science in China: Information Science, 2024, 54(11): 2657-2673) focused on UAV swarms, explored the matching of heterogeneous capabilities and task requirements based on alliance formation game theory, proposed a distributed task allocation algorithm, emphasized real-time performance and local adjustments to cope with fault scenarios, and demonstrated the application potential of overlapping alliances in dynamic environments. However, existing overlapping alliance methods suffer from low resource allocation efficiency and insufficient dynamic adaptability, especially in complex environments with topology time-varying and communication constraints, making it difficult to achieve accurate matching of agent capabilities and task requirements. Therefore, it is urgent to develop an overlapping alliance formation method driven by capabilities, which can realize the aggregation of agent capabilities and optimization of task matching under topology time-varying and communication constraints, thereby improving the overall system efficiency. Summary of the Invention

[0004] To address the shortcomings of traditional disjoint alliance formation methods in multi-agent cooperative task allocation, this invention provides a capability-centric driven multi-agent overlapping alliance optimization method. This method establishes a collaborative framework centered on agent capabilities, achieving efficient alliance generation and optimization under topology-varying and communication constraints. Through capability aggregation and real-time matching of task requirements, this method significantly improves system resource utilization, alliance stability, and adaptability to complex environments. It is suitable for highly dynamic task scenarios such as disaster relief and UAV swarms, providing a reliable cooperative solution for multi-agent systems.

[0005] The first objective of this invention is to provide a capability-centric driven multi-agent overlapping alliance optimization method for multi-agent alliances based on tasks, comprising:

[0006] The agent for a given task aggregates information with multiple surrounding agents and collects the communication link status between each agent, as well as the original physical capability vector of each agent. The environmental attenuation factor is obtained based on the communication link status between each agent. Based on the environmental attenuation factor and the original physical capability vector of each agent, the total available capability pool of the region is obtained.

[0007] If the available total capacity pool in the region is determined to meet the demand, an alliance is formed between the agent for the given task and multiple surrounding agents.

[0008] When building an alliance, iterate through all combinations formed between the agent with the given task and multiple surrounding agents, and select the alliance that meets the task requirements and has the best system performance.

[0009] Dynamic resource allocation is implemented for the optimal alliance, and the resource status of the agents is updated in real time. The updated resource status of the agents is then fed back to provide the latest agent resource situation for the next round of task allocation.

[0010] When a new task arrives, the remaining resources after the alliance agents have deducted their usage are used to build a new alliance.

[0011] In one embodiment, the total available capacity pool for the region is obtained according to the following steps:

[0012] A dynamic adjacency matrix is ​​constructed based on the communication link status between each agent;

[0013] The environmental attenuation factor is obtained based on the dynamic adjacency matrix;

[0014] Based on the environmental degradation factor and the original physical capability vector of each agent, obtain the effective capability vector of each agent;

[0015] The effective capability vectors of each agent are pre-summarized to generate a pool of total available capabilities for the region.

[0016] In one embodiment, selecting the alliance that meets the task requirements and has the best system performance involves screening all alliance combinations that meet the resource constraints, calculating the score of each alliance using a system performance function, and finally selecting the alliance with the highest performance.

[0017] In one embodiment, all alliance combinations that satisfy resource constraints are filtered out, denoted as... :

[0018]

[0019] In the formula, The task requirement vector is obtained by parsing the task; The effective capability vector corresponding to each agent;

[0020] For a given task, all combinations formed between the agent and multiple surrounding agents.

[0021] To gather potential alliance members.

[0022] In one embodiment, the score of each league is calculated using a system performance function, and the league with the highest performance is selected, denoted as [missing information]. :

[0023]

[0024] In the formula, Alliance combinations to meet resource constraints; The value of a task is obtained by analyzing the task. , All are weighting coefficients; For the alliance The size of the alliance is used to punish excessively large alliances in order to reduce communication overhead. This is a term representing the decay of task value over time. The difference between the estimated execution time of the task and the current time; A configurable parameter that shows how the value of a task changes over time.

[0025] In one embodiment, dynamic resource allocation for the optimal alliance is achieved by calculating the resource contribution of each agent within the optimal alliance to the task, as follows:

[0026]

[0027] In the formula, This represents the resource contribution of each agent in the optimal alliance to the task. The effective capability vector corresponding to each agent; The most effective alliance; The robust buffer coefficient; This is the task requirement vector; For intelligent agents The corresponding effective capability vector.

[0028] In one embodiment, it further includes:

[0029] If the environmental degradation factor changes or the given task changes, causing the current alliance's effectiveness to fall below a preset threshold or the capability constraints to fail, a dynamic adjustment mechanism is triggered to rebuild the alliance structure based on the remaining effective nodes.

[0030] The second objective of this invention is to provide a system for a capability-centric driven multi-agent overlapping alliance optimization method, comprising an environment unit, a capability unit, and a task unit;

[0031] The environment unit is used to collect the communication link status between each agent; and to obtain the environment attenuation factor based on the communication link status between each agent.

[0032] Capability unit is used to collect the original physical capability vector of each agent and obtain the total available capability pool of the region based on the environmental attenuation factor and the original physical capability vector of each agent.

[0033] The task unit is used to aggregate the intelligence agent for a given task with multiple surrounding intelligence agents. If the total capacity pool of the region can meet the requirements, an alliance is built between the intelligence agent for the given task and multiple surrounding intelligence agents.

[0034] When building an alliance, iterate through all combinations formed between the agent with the given task and multiple surrounding agents, and select the alliance that meets the task requirements and has the best system performance.

[0035] Dynamic resource allocation is implemented for the optimal alliance, and the resource status of the agents is updated in real time;

[0036] When a new task arrives, the remaining resources after the alliance agents have deducted their usage are used to build a new alliance.

[0037] In one embodiment, the environmental unit includes an environmental monitoring module, a topology analysis module, and a dynamic adjustment module;

[0038] The environmental monitoring module is used to collect the communication link status between intelligent agents in real time;

[0039] The topology analysis module is used to construct a dynamic adjacency matrix based on the link status;

[0040] The dynamic adjustment module is used to calculate the environmental attenuation factor based on the adjacency matrix;

[0041] The capability unit includes a capability integration module, a capability assessment module, and a capability aggregation module;

[0042] The capability integration module is used to collect the original physical capability vectors of each intelligent agent;

[0043] The capability assessment module is used to combine the environmental degradation factor from the environmental unit to correct the original physical capability into an effective capability vector.

[0044] The capability aggregation module is used to scan all communicable agent nodes around the task to be executed, pre-aggregate the effective capability vectors of these agents, and generate a total available capability pool for the region.

[0045] In one embodiment, the task unit includes a task requirement management module, an alliance building module, and a resource allocation module;

[0046] The task requirement management module is used to parse the task description and generate a requirement vector and task value;

[0047] The alliance building module is used to receive the total available capacity pool in the region from the capacity aggregation module. If it is determined that the total available capacity pool meets the requirements, the alliance building process based on the contract network is started. The process iterates through all combinations formed between the agent of the given task and multiple surrounding agents, and selects the alliance that meets the task requirements and has the best system efficiency.

[0048] The resource allocation module is used to dynamically allocate resources to the optimal alliance and update the resource status of the agent in real time.

[0049] The resource allocation module also includes feeding back the updated agent resource status to the capability unit, providing the latest agent resource status for the next round of task allocation, thereby achieving overlapping allocation under multiple tasks.

[0050] The present invention has at least the following beneficial effects:

[0051] This invention provides a capability-centric driven multi-agent overlapping alliance optimization method. This method fully utilizes the dynamic integration of agent capabilities and the precise matching of task requirements to improve the resource utilization and task completion efficiency of the alliance. Compared with traditional methods, this invention can improve resource allocation efficiency in multi-task parallel scenarios and effectively adapt to topology-time-varying environments through a dynamic environmental response mechanism, thereby enhancing system robustness. Attached Figure Description

[0052] Figure 1 is a framework diagram of overlapping alliances driven by capability centers.

[0053] Figure 2 is a flowchart of the alliance building process based on the contract network.

[0054] Figure 3 shows an example diagram of alliance structure generation for 6 agents targeting 2 objectives. Figure (a) shows an example diagram of alliance structure generation targeting objective T2, and Figure (b) shows an example diagram of alliance structure generation targeting objective T1. Detailed Implementation

[0055] In order to illustrate the technical means and effects adopted by the present invention to achieve the intended purpose, the following detailed description is provided in conjunction with the embodiments.

[0056] The purpose of this invention is to provide a capability-centric driven multi-agent overlapping alliance optimization method. This method addresses the shortcomings of existing technologies, which limit agents to participating in only a single alliance, leading to low resource utilization efficiency, especially in multi-task parallelism and topology-varying environments where they struggle to adapt to dynamic demands. Furthermore, while existing overlapping alliance formation methods improve flexibility, they fail to effectively integrate the dynamic matching of agent capabilities with task requirements, resulting in insufficient resource allocation optimization and poor environmental adaptability. This new method, through real-time matching of capability aggregation and task requirements, significantly improves system resource utilization, alliance stability, and adaptability to complex environments. It is suitable for highly dynamic task scenarios such as disaster relief and drone swarms, providing a reliable collaborative solution for multi-agent systems.

[0057] To achieve the above objectives, this invention provides a capability-centric driven multi-agent overlapping alliance optimization method, used for multiple agents to form alliances based on tasks, including:

[0058] S1. The agent for the given task aggregates information with multiple surrounding agents and collects the communication link status between each agent, as well as the original physical capability vector of each agent; the environmental attenuation factor is obtained based on the communication link status between each agent; the total available capability pool of the region is obtained based on the environmental attenuation factor and the original physical capability vector of each agent.

[0059] The total available capacity pool for the region is obtained according to the following steps:

[0060] A dynamic adjacency matrix is ​​constructed based on the communication link status between each agent;

[0061] The environmental attenuation factor is obtained based on the dynamic adjacency matrix;

[0062] Based on the environmental degradation factor and the original physical capability vector of each agent, obtain the effective capability vector of each agent;

[0063] The effective capability vectors of each agent are pre-summarized to generate a pool of total available capabilities for the region.

[0064] S2. If the total capacity pool available in the region meets the requirements, an alliance is formed between the agent for the given task and multiple surrounding agents.

[0065] When building an alliance, iterate through all combinations formed between the agent with the given task and multiple surrounding agents, and select the alliance that meets the task requirements and has the best system performance.

[0066] The process of selecting the alliance that meets the task requirements and has the best system performance involves filtering all alliance combinations that meet the resource constraints, calculating the score of each alliance using the system performance function, and finally selecting the alliance with the highest performance.

[0067] Filter all alliance combinations that satisfy the resource constraints, denoted as . :

[0068]

[0069] In the formula, The task requirement vector is obtained by parsing the task; The effective capability vector corresponding to each agent;

[0070] For a given task, there are all possible combinations between the agent and multiple surrounding agents.

[0071] To gather potential alliance members.

[0072] The score for each league is calculated using the system performance function, and the league with the highest performance is selected, denoted as [League Name]. :

[0073]

[0074] In the formula, Alliance combinations to meet resource constraints; The value of a task is obtained by analyzing the task. , All are weighting coefficients; For the alliance The size of the alliance is used to punish excessively large alliances in order to reduce communication overhead. This is a term representing the decay of task value over time. The difference between the estimated execution time of the task and the current time; A configurable parameter that shows how the value of a task changes over time.

[0075] S3. Implement dynamic resource allocation for the optimal alliance, update the agent resource status in real time, and provide feedback on the updated agent resource status to provide the latest agent resource situation for the next round of task allocation;

[0076] Dynamic resource allocation for the optimal alliance is achieved by calculating the resource contribution of each agent within the alliance to the task, as follows:

[0077]

[0078] In the formula, This represents the resource contribution of each agent in the optimal alliance to the task. The effective capability vector corresponding to each agent; The most effective alliance; The robust buffer coefficient; This is the task requirement vector; For intelligent agents The corresponding effective capability vector.

[0079] S4. When a new task arrives, the remaining resources after the alliance agents have deducted their usage are used to start a new round of alliance building.

[0080] The present invention also includes: if the environmental decay factor changes or the given task changes, causing the current alliance's effectiveness to fall below a preset threshold or the capability constraint to fail, a dynamic adjustment mechanism is triggered to rebuild the alliance structure based on the remaining effective nodes.

[0081] To achieve efficient overlapping alliance formation of heterogeneous multi-agent systems in a topologically time-varying environment, the following explanation is provided in conjunction with the accompanying drawings.

[0082] Referring to Figure 1, a multi-agent overlapping alliance optimization method based on capability center-driven approach is described below:

[0083] (1) Construct a capability center-driven overlapping alliance formation framework, which consists of three core units: environment unit, capability unit and task unit.

[0084] a) The environmental unit, serving as the system's perception foundation, comprises an environmental monitoring module, a topology analysis module, and a dynamic adjustment module. The environmental monitoring module collects real-time communication link states between agents; the topology analysis module constructs a dynamic adjacency matrix based on the link states; and the dynamic adjustment module calculates the environmental attenuation factor based on the adjacency matrix. This factor quantifies the obstruction effect of the physical environment on capability transmission and injects it into the capability unit in real time.

[0085] b) The capability unit is responsible for capability standardization and prediction, and includes a capability integration module, a capability assessment module, and a capability aggregation module. Capability integration module: collects data from various intelligent agents. The original physical capability vector, such as sensor detection radius, computing power, and maximum load, is denoted as... Capability assessment module: combining environmental degradation factors from environmental units ( (This will correct the original physical capabilities into an effective capability vector.) The calculation formula is: This reflects the effective output that the intelligent agent can actually contribute to the task under the current environment; Capability aggregation module: This module is the task area, and it scans the tasks to be executed. All surrounding communicable agent nodes, and the effective capability vectors of these agents. A preliminary aggregation is performed to generate a pool of total available capacity for the region. This data flows to the task unit to help determine whether the current region has sufficient resource potential to initiate the alliance building process.

[0086] c) Task Unit: Responsible for the alliance's decision-making and execution, including the task requirement management module, the alliance construction module, and the resource allocation module.

[0087] Task Requirements Management Module: Parses task descriptions and generates requirement vectors. and task value ;

[0088] Alliance Construction Module: This is the core decision-making module. It receives the "Regional Available Total Capacity Pool" information from the Capacity Aggregation Module. If it determines that the regional available total capacity meets the demand, it initiates the contract-based alliance construction process. This module filters all feasible alliance sets that meet resource constraints ( The system uses a system performance function to calculate the score for each feasible coalition, and finally selects the coalition with the highest performance. ;

[0089] It should be noted that if the total available capacity of the region is greater than the target requirement, the alliance construction will be initiated. This judgment process takes place in the alliance construction module because it requires task information from the region's total capacity pool and the task requirement management module.

[0090] Resource allocation module: for winning bid alliances The system calculates the specific resource allocation ratio for each member agent. After allocation, it updates the remaining resource status of each agent and feeds this status back to the capability integration module. This provides the latest agent resource information for the next round of task allocation, thereby achieving overlapping allocation in multi-task scenarios. .

[0091] (2) When building an alliance based on the contract network, firstly, the online construction of the initial alliance is performed, given the task. The alliance building module generates the optimal alliance by combining the available intelligent agents in the surrounding area with the following steps:

[0092] a) Bidding and Feasibility Screening: The agent with the given task broadcasts its requirements as the bidding party. The surrounding intelligent agents, as bidders, submit their effective capabilities. The alliance building module calculates all those that meet the reported capability data. The intelligent agents are combined to form a feasible set of alliances. This involves iterating through all possible combinations and selecting the set of all candidate alliances that satisfy the capability constraints, i.e., the set of candidate alliances whose sum of capabilities exceeds the requirements. This set is denoted as... :

[0093]

[0094] b) Performance Evaluation and Optimization: To balance task completion with system overhead, such as communication and collaboration costs and personnel size, the following system performance function is adopted. Each viable alliance is scored.

[0095]

[0096] The system selects the alliance with the highest score. As the ultimate executor:

[0097]

[0098] In the formula, Indicates task The initial value; This represents the difference between the estimated execution time of the task and the current time. This represents the decay of task value over time. Indicates alliance The size of the alliance is used to punish excessively large alliances in order to reduce communication overhead. These are the weighting coefficients.

[0099] Referring to Figure 2, the contract network-based alliance building process is executed as follows: When a new task arrives or the task queue is updated, the alliance building phase begins. The capability aggregation module first scans the communicable agent nodes around the task. If the total capability of the area meets the activation threshold, the alliance building module in the task unit initiates the contract network protocol process.

[0100] a) Bidding and Tendering: Task Release Requirement Vector The surrounding intelligent agents report their effective capability vectors. .

[0101] b) Feasibility Set Calculation: Based on the reported capability data, the alliance building module calculates all feasible sets that meet the requirements. Combining intelligent agents to form a feasible alliance set .

[0102] c) Performance Optimization: To minimize system overhead while satisfying the task requirements, the system performance function is used to optimize each feasible coalition. An assessment will be conducted.

[0103] (3) Realize dynamic resource allocation and status feedback in determining the optimal alliance Then, the resource allocation module calculates the capabilities of each agent within the alliance based on the proportion of capabilities of the alliance members. For the task resource contribution The allocation principle is to distribute resources proportionally to capacity while retaining minimum redundancy.

[0104]

[0105] In the formula, This is the robust buffer coefficient.

[0106] After allocation is completed, a state feedback operation is performed, and the agent receives feedback from its total capabilities. deducting from Update remaining available capacity This data is then transmitted back to the capability integration module in real time, and the intelligent agent... Update its remaining available capabilities: This updated state is immediately fed back to the capability integration module of the capability unit. This enables the intelligent agent... Participating in the alliance Meanwhile, if there are still remaining resources, they can continue to participate as candidate nodes in the "capability aggregation" and "alliance building" of other parallel tasks, thereby physically realizing the formation of overlapping alliances.

[0107] It should be noted that state feedback is key to achieving overlapping alliances: when the next parallel task arrives, the capability integration module uses the remaining capabilities of the agent after deducting the occupied capabilities. A new round of evaluation and bidding will be conducted. As long as... The agent can then be selected again to join a new alliance, thus enabling it to belong to multiple alliances within the same time window.

[0108] (4) Dynamic adjustment under topological time-varying conditions, the specific process is as follows:

[0109] During task execution, the system continuously and cyclically performs environmental monitoring as described in step one. If the environmental degradation factor... Significant changes occur (e.g., communication interruption leads to) (or changes in mission requirements, leading to reduced effectiveness of the current alliance) If the number of nodes falls below a preset threshold or the capability constraints fail, a dynamic adjustment mechanism is triggered. The system will then use the remaining valid nodes as a basis to re-trigger the bidding and selection process in step two, quickly reconstructing the alliance structure to adapt to the time-varying topology environment.

[0110] To further illustrate the capability center-driven multi-agent overlapping alliance optimization method provided by this invention, specific examples are used for explanation.

[0111] As shown in Figure 3, this example uses 6 agents ( The alliance structure formed for the two objectives (T1 and T2) is as follows:

[0112] Figure 3(a) and (b) This represents the six agents in the task area, with their initial capabilities as follows: The mission area has two objectives, with resource requirements as follows: Initially, the agents do not know the location of the target or the required resources, and each agent performs a cooperative search task; see Figure 3(a), when At that time, the intelligent agents around target T2 that can communicate are Its total available capabilities meet the requirements of target T2, thus constructing an alliance. See Figure 3(b), when At that time, the intelligent agents around target T1 that can communicate are It built an alliance .

[0113] This invention provides a system for a capability-centric driven multi-agent overlapping alliance optimization method, comprising an environment unit, a capability unit, and a task unit;

[0114] The environment unit is used to collect the communication link status between each agent; and to obtain the environment attenuation factor based on the communication link status between each agent.

[0115] The environmental unit includes an environmental monitoring module, a topology analysis module, and a dynamic adjustment module;

[0116] The environmental monitoring module is used to collect the communication link status between intelligent agents in real time;

[0117] The topology analysis module is used to construct a dynamic adjacency matrix based on the link status;

[0118] The dynamic adjustment module is used to calculate the environmental attenuation factor based on the adjacency matrix.

[0119] Capability unit is used to collect the original physical capability vector of each agent and obtain the total available capability pool of the region based on the environmental attenuation factor and the original physical capability vector of each agent.

[0120] The capability unit includes a capability integration module, a capability assessment module, and a capability aggregation module;

[0121] The capability integration module is used to collect the original physical capability vectors of each intelligent agent;

[0122] The capability assessment module is used to combine the environmental degradation factor from the environmental unit to correct the original physical capability into an effective capability vector.

[0123] The capability aggregation module is used to scan all communicable agent nodes around the task to be executed, pre-aggregate the effective capability vectors of these agents, and generate a total available capability pool for the region.

[0124] The task unit is used to aggregate the intelligence agent for a given task with multiple surrounding intelligence agents. If the total capacity pool of the region can meet the requirements, an alliance is built between the intelligence agent for the given task and multiple surrounding intelligence agents.

[0125] When building an alliance, iterate through all combinations formed between the agent with the given task and multiple surrounding agents, and select the alliance that meets the task requirements and has the best system performance.

[0126] Dynamic resource allocation is implemented for the optimal alliance, and the resource status of the agents is updated in real time;

[0127] When a new task arrives, the remaining resources after the alliance agents have deducted their usage are used to build a new alliance.

[0128] The task unit includes a task requirement management module, an alliance building module, and a resource allocation module;

[0129] The task requirement management module is used to parse the task description and generate a requirement vector and task value;

[0130] The alliance building module is used to receive the total available capacity pool in the region from the capacity aggregation module. If it is determined that the total available capacity pool meets the requirements, the alliance building process based on the contract network is started. The process iterates through all combinations formed between the agent of the given task and multiple surrounding agents, and selects the alliance that meets the task requirements and has the best system efficiency.

[0131] The resource allocation module is used to dynamically allocate resources to the optimal alliance and update the resource status of the agent in real time.

[0132] The resource allocation module also includes feeding back the updated agent resource status to the capability unit, providing the latest agent resource status for the next round of task allocation, thereby achieving overlapping allocation under multiple tasks.

[0133] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A multi-agent overlapping alliance optimization method based on capability center-driven approach, characterized in that, This system is used for multi-agent alliances based on tasks, including: aggregating the information of the agent for a given task with multiple surrounding agents, collecting the communication link status between agents, and the original physical capability vector of each agent; obtaining an environmental attenuation factor based on the communication link status between agents; obtaining the total available capability pool for the region based on the environmental attenuation factor and the original physical capability vector of each agent; if the total available capability pool for the region meets the requirements, constructing an alliance between the agent for the given task and multiple surrounding agents; during alliance construction, iterating through all combinations formed between the agent for the given task and multiple surrounding agents, and selecting the alliance that meets the task requirements and has the best system efficiency; implementing dynamic resource allocation for the optimal alliance, updating the agent resource status in real time; and feeding back the updated agent resource status to provide the latest intelligent capabilities for the next round of task allocation. The system considers the resource status of the alliance. When a new task arrives, the remaining resources after deducting the occupied resources of the alliance agents are used to construct a new alliance. The total available capacity pool for the region is obtained by following these steps: constructing a dynamic adjacency matrix based on the communication link status between each agent; obtaining an environmental attenuation factor based on the dynamic adjacency matrix; obtaining the effective capacity vector of each agent based on the environmental attenuation factor and the original physical capacity vector of each agent; and pre-summarizing the effective capacity vectors of each agent to generate the total available capacity pool for the region. The original physical capacity vector of each agent includes: sensor detection radius, computing power, and maximum load. If the environmental attenuation factor changes or the given task changes, causing the current alliance's efficiency to fall below a preset threshold or the capacity constraints to fail, a dynamic adjustment mechanism is triggered to rebuild the alliance structure based on the remaining effective nodes.

2. The multi-agent overlapping alliance optimization method based on capability center-driven approach according to claim 1, characterized in that, The process of selecting the alliance that meets the task requirements and has the best system performance involves screening all alliance combinations that meet the resource constraints, calculating the score of each alliance using the system performance function, and finally selecting the alliance with the highest performance.

3. The multi-agent overlapping alliance optimization method based on capability center-driven approach according to claim 2, characterized in that, Filter all alliance combinations that satisfy the resource constraints, denoted as . ： In the formula, The task requirement vector is obtained by parsing the task; For intelligent agents The corresponding effective capability vector; For a given task, there are all possible combinations between the agent and multiple surrounding agents. To gather potential alliance members.

4. The multi-agent overlapping alliance optimization method based on capability center-driven approach according to claim 2, characterized in that, The score for each league is calculated using the system performance function, and the league with the highest performance is selected, denoted as [League Name]. ： In the formula, Alliance combinations to meet resource constraints; The value of a task is obtained by analyzing the task. 、 All are weighting coefficients; For the alliance The size of the alliance is used to punish excessively large alliances in order to reduce communication overhead. This is a term representing the decay of task value over time. The difference between the estimated execution time of the task and the current time; A configurable parameter that shows how the value of a task changes over time.

5. The multi-agent overlapping alliance optimization method based on capability center-driven approach according to claim 2, characterized in that, Dynamic resource allocation for the optimal alliance is achieved by calculating the resource contribution of each agent within the alliance to the task, as follows: In the formula, This represents the resource contribution of each agent in the optimal alliance to the task. For intelligent agents The corresponding effective capability vector; The most effective alliance; The robust buffer coefficient; This is the task requirement vector; For intelligent agents The corresponding effective capability vector.

6. A system for the capability center-driven multi-agent overlapping alliance optimization method as described in claim 1, characterized in that, It includes an environment unit, a capability unit, and a task unit; the environment unit is used to collect the communication link status between each agent; and to obtain the environment attenuation factor based on the communication link status between each agent; the capability unit is used to collect the original physical capability vector of each agent; and to obtain the total available capability pool of the region based on the environment attenuation factor and the original physical capability vector of each agent. The task unit is used to aggregate the intelligence agent for a given task with multiple surrounding intelligence agents. If the total available capacity pool in the region meets the requirements, an alliance is built between the intelligence agent for the given task and multiple surrounding intelligence agents. When building an alliance, all combinations formed between the intelligence agent for the given task and multiple surrounding intelligence agents are traversed, and the alliance that meets the task requirements and has the best system efficiency is selected. Dynamic resource allocation is implemented for the optimal alliance, and the resource status of the intelligence agents is updated in real time. When a new task arrives, the remaining resources of the intelligence agents already in the alliance are deducted after deducting their occupied resources, and a new round of alliance building is carried out.

7. The system according to claim 6, characterized in that, The environment unit includes an environment monitoring module, a topology analysis module, and a dynamic adjustment module; the environment monitoring module is used to collect the communication link status between agents in real time; the topology analysis module is used to construct a dynamic adjacency matrix based on the link status; the dynamic adjustment module is used to calculate the environment attenuation factor based on the adjacency matrix; the capability unit includes a capability integration module, a capability assessment module, and a capability aggregation module. The capability integration module is used to collect the original physical capability vectors of each intelligent agent; The capability assessment module is used to combine the environmental degradation factor from the environmental unit to correct the original physical capability into an effective capability vector. The capability aggregation module is used to scan all communicable agent nodes around the task to be executed, pre-aggregate the effective capability vectors of these agents, and generate a total available capability pool for the region.

8. The system according to claim 6, characterized in that, The task unit includes a task requirement management module, an alliance building module, and a resource allocation module. The task requirement management module parses the task description and generates a requirement vector and task value. The alliance building module receives the total available capability pool from the capability aggregation module. If the available capability pool meets the requirements, it initiates an alliance building process based on a contract network, traversing all combinations formed between the agent for the given task and multiple surrounding agents, and selecting the alliance that meets the task requirements and has the best system efficiency. The resource allocation module dynamically allocates resources to the optimal alliance, updating the agent resource status in real time. The resource allocation module also feeds back the updated agent resource status to the capability unit, providing the latest agent resource information for the next round of task allocation, thereby achieving overlapping allocation under multiple tasks.

Images