A multi-agent cooperative decision-making method and device based on behavior characteristic quantitative modeling and weak connection mechanism

Abstract

The application discloses a kind of multi-agent cooperation decision-making method and device based on behavior characteristic quantification modeling and weak connection mechanism.The method quantitatively evaluates the impulsivity level of each agent, and establishes a weak connection interaction influence mechanism based on this to adjust the information interaction weight between agents, thereby effectively weakening the negative interference of impulsive individuals on group cooperation strategy. This method can be flexibly embedded in mainstream multi-agent reinforcement learning framework, improving the robustness and overall performance of multi-agent system containing behaviorally unstable individuals in cooperative tasks.

Description

Technical Field

[0001] This invention belongs to the field of artificial intelligence technology, specifically relating to a multi-agent collaborative decision-making method and device based on behavioral characteristic quantitative modeling and weak connection mechanism. Background Technology

[0002] Multi-agent systems have shown great application potential in many fields, such as robot swarms, distributed sensor networks, smart grids, traffic management, and virtual reality. The overall efficiency of the system highly depends on effective cooperation and information sharing among the agents. Current mainstream multi-agent reinforcement learning algorithms (such as MAPPO, QMIX, and VDN) typically assume that the agents are perfectly rational, with stable behavioral strategies aimed at maximizing long-term rewards. However, in real-world or complex simulation environments, some agents may exhibit irrational characteristics due to internal state changes, environmental disturbances, or specific design choices. These characteristics include short-sighted decision-making, over-response, and unstable behavioral patterns, similar to human "impulsive" behavior. The existence of such impulsive agents can interfere with the convergence of group decision-making, disrupt the consistency of cooperative strategies, and reduce the overall task completion efficiency and stability of the system. Existing research lacks explicit modeling of this specific irrational behavior and effective mechanisms to dynamically adjust the influence of impulsive individuals in group decision-making, thus limiting the adaptability and robustness of existing multi-agent systems in complex or adversarial environments. Therefore, a new method is urgently needed to assess and modulate the impact of impulsive agent behavior on group decision-making. Summary of the Invention

[0003] The purpose of this invention is to overcome the shortcomings of existing multi-agent systems, such as the lack of explicit modeling of individual irrational behavior and the inability to effectively mitigate the negative impact of impulsive agents during collaborative decision-making. This invention provides a multi-agent collaborative decision-making method and apparatus based on quantitative modeling of behavioral characteristics and a weak connection mechanism. This method quantitatively assesses the impulsive nature of agents and dynamically adjusts the information interaction weights based on the interaction graph topology and personality differences. This suppresses the influence of impulsive individual information, thereby improving the collaborative performance, robustness, and training stability of the multi-agent system in the presence of behaviorally unstable individuals.

[0004] To achieve the above objectives, the present invention employs the following technical solution:

[0005] This invention provides a multi-agent collaborative decision-making method based on behavioral characteristic quantitative modeling and weak connection mechanism, comprising:

[0006] The behavioral data of each agent in the multi-agent system is acquired, and the behavioral data is analyzed using a predefined impulsive personality scale to calculate a quantitative impulsive personality score for each agent.

[0007] Construct a multi-agent interaction graph at the current moment using agents as vertices, and identify potential interaction pairs between agents through topological relationships and communication ranges;

[0008] For the potential interaction pairs in the multi-agent interaction graph, calculate the topological connection strength;

[0009] Set a connection strength threshold, and determine a topology information filtering factor based on the comparison between the topology connection strength and the connection strength threshold;

[0010] Based on the differences in impulsive personality scores between the potential interaction pairs, the impulsive personality moderating weights are calculated.

[0011] The final effective information interaction weight is calculated by combining the topological information filtering factor with the impulsive personality adjustment weight.

[0012] The effective information interaction weights are integrated into the selected multi-agent reinforcement learning algorithm to train or drive the multi-agent system to make decisions in a collaborative task environment.

[0013] Preferably, acquiring the behavioral data of each agent in the multi-agent system includes:

[0014] Generate behavioral data of agents through large language models or extract behavioral data of agents from their interactions with the environment.

[0015] Preferably, the behavioral impulsivity scale adopts the Barrett Impulsivity Scale or an assessment standard adapted from the scale.

[0016] The quantified impulsive personality score refers to normalizing the score obtained using the impulsive personality scale to the [0,1] interval.

[0017] Preferably, the topology connection strength is calculated as follows:

[0018] , in, For potential interaction relationships of intelligent agents The topological connectivity strength, and respectively intelligent agents and The degree in the multi-agent interaction graph, where the value is the agent's degree. and The number of agent connections within the observable range. It refers to connecting intelligent agents and The number of edges in the shortest path between them.

[0019] Preferably, a topology information filtering factor is determined based on a comparison between the topology connection strength and the connection strength threshold, including:

[0020] Set a connection strength threshold ,when When this occurs, it is determined to be a weak connection, and the topology information filtering factor is adjusted. Otherwise, it is determined to be a strong connection, and the topology information filtering factor is set. .

[0021] Preferably, based on the differences in impulsive personality scores between the potential interaction pairs, the impulsive personality moderating weights are calculated, including:

[0022] Adjust the weight of impulsive personality The design is based on the difference in impulsive personality scores. A monotonically decreasing function. and respectively intelligent agents and intelligent agents A quantitative score for impulsive personality.

[0023] Preferably, the impulsive personality moderating weight Represented as:

[0024] ,

[0025] or,

[0026] ;

[0027] in, It is a positive adjustment coefficient.

[0028] Preferably, the topological information filtering factor is combined with the impulsive personality adjustment weight to calculate the final effective information interaction weight, including:

[0029] ,

[0030] in, Weights for effective information exchange.

[0031] Preferably, the multi-agent reinforcement learning algorithm is selected from value function decomposition-based algorithms, actor-critic-based algorithms, or attention mechanism-based algorithms.

[0032] This invention also provides a multi-agent collaborative decision-making device based on behavioral characteristic quantification modeling and weak connection mechanism, used to implement the above-mentioned multi-agent collaborative decision-making method based on behavioral characteristic quantification modeling and weak connection mechanism, the device comprising:

[0033] The agent scoring module is used to acquire behavioral data of each agent in the multi-agent system, analyze the behavioral data using a predefined impulsive personality scale, and calculate a quantitative impulsive personality score for each agent.

[0034] The topology construction module is used to construct a multi-agent interaction graph at the current moment with agents as vertices, and to identify potential interaction pairs between agents through topology and communication range.

[0035] The connection strength calculation module is used to calculate the topological connection strength for potential interaction pairs in the multi-agent interaction graph;

[0036] The filter factor calculation module is used to set a connection strength threshold and determine a topology information filter factor based on the comparison between the topology connection strength and the connection strength threshold.

[0037] The moderating weight calculation module is used to calculate the moderating weight of impulsive personality based on the difference between the impulsive personality scores of the potential interaction pairs.

[0038] The effective weight calculation module is used to combine the topological information filtering factor with the impulsive personality adjustment weight to calculate the final effective information interaction weight.

[0039] An integration module is used to integrate the effective information interaction weights into the selected multi-agent reinforcement learning algorithm to train or drive the multi-agent system to make decisions in a collaborative task environment.

[0040] The beneficial effects achieved by this invention are as follows:

[0041] (1) This invention introduces a behavioral characteristic modeling mechanism for the first time, clearly quantifies the irrational and short-sighted behavior of agents, and adjusts information contribution according to the difference in impulsiveness scores, making the interaction between high-impulsivity and low-impulsivity agents safer and more controllable, providing a new theoretical and engineering foundation for multi-agent heterogeneity modeling. By significantly reducing the interference of impulsive individuals through the weak connection mechanism, based on graph topology, only the information transmission in the weak connection path is retained, structurally limiting the scope of influence of impulsive agents.

[0042] (2) This invention is easy to integrate with the mainstream MARL framework without modifying the original policy network structure. It can be directly embedded as an information weighting module into algorithms such as MAPPO, QMIX, HATRPO, and G2AN.

[0043] (3) The present invention improves the robustness and performance of collaborative tasks. In environments containing highly impulsive individuals or behavioral noise, the method can significantly improve the success rate of tasks, training stability and team coordination. Attached Figure Description

[0044] Figure 1 This is a schematic diagram of the multi-agent collaborative decision-making method based on behavioral characteristic quantitative modeling and weak connection mechanism provided by the present invention;

[0045] Figure 2 This is a schematic diagram of the process of behavioral characteristic quantitative modeling based on LLM and AIMS-50 in an embodiment of the present invention;

[0046] Figure 3 This is a flowchart illustrating the construction and application of the interaction weight adjustment mechanism (weak connection) in an embodiment of the present invention.

[0047] Figure 4 This is a schematic diagram of the collaboration effect test in the SMAC environment in an embodiment of the present invention;

[0048] Figure 5 This is an example of the modified quantity representation used in the behavioral characteristic modeling process of this invention. Detailed Implementation

[0049] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the embodiments and accompanying drawings. Here, the illustrative embodiments and descriptions of this invention are used to explain the invention, but are not intended to limit the invention.

[0050] It should also be noted that, in order to avoid obscuring the invention with unnecessary details, only the structures and / or processing steps closely related to the solution according to the invention are shown in the accompanying drawings, while other details that are not closely related to the invention are omitted.

[0051] It should be emphasized that the term "including / comprises" as used herein refers to the presence of a feature, element, step, or component, but does not exclude the presence or addition of one or more other features, elements, steps, or components.

[0052] It should also be noted that, unless otherwise specified, the term "connection" in this article can refer not only to a direct connection, but also to an indirect connection involving an intermediary.

[0053] In the following description, embodiments of the invention will be illustrated with reference to the accompanying drawings. In the drawings, the same reference numerals represent the same or similar parts, or the same or similar steps.

[0054] It should be emphasized here that the step markers mentioned below are not a limitation on the order of the steps, but should be understood as meaning that the steps can be executed in the order mentioned in the embodiments, or in a different order than in the embodiments, or several steps can be executed simultaneously.

[0055] This invention provides a multi-agent collaborative decision-making method based on behavioral characteristic quantitative modeling and weak connection mechanism. See [link to relevant documentation]. Figure 1 This includes the following steps:

[0056] Step 1: Acquire behavioral data for each agent. Analyze the acquired behavioral data using a predefined behavioral impulsivity personality scale (or corresponding calculation model) for each agent. A quantitative impulsive personality score is calculated. Among them, the behavioral data of the intelligent agent refers to the observed state, location information and behavioral decision description of the intelligent agent in the collaborative task;

[0057] Step two: Construct the multi-agent interaction graph at the current moment, and identify the potential interaction relationships between the agents through topological relationships and communication range. ;

[0058] Step 3: For the potential interaction relationships in the multi-agent interaction graph Calculate its topological connectivity strength Topology connection strength Based on intelligent agents and Degree in the interaction diagram , and the shortest path length between them calculate;

[0059] Step 4: Set a connection strength threshold. According to topology connection strength With connection strength threshold Based on the comparison results, a topological information filtering factor is determined. :when It is then identified as a weak connection and ordered Otherwise, it is considered a strong connection and is set to... Weak connections are used to allow information to pass through, while strong connections are used to filter the propagation paths of unstable individuals.

[0060] Step 5, based on intelligent agents and intelligent agents Impulsive personality score and The differences between them were used to calculate the moderating weights for impulsive personality. Among them, the moderating weight of impulsive personality Designed to vary with rating differences Monotonically decreasing functions; for example, exponentially decreasing functions. or inverse proportional function ,in The positive adjustment coefficient ensures that agents with large score differences have less influence on each other;

[0061] Step 6: Filter the above topological information using the filter factors. Moderating weight of impulsive personality By combining these methods, the final effective information interaction weights are calculated. ;

[0062] Step 7: Integrate with the multi-agent collaborative decision-making framework, and apply the effective information interaction weights calculated in the above steps. This information is integrated into the selected multi-agent reinforcement learning (MARL) algorithm, and the effective information is used to interact with weights during the collaborative decision-making process of the MARL algorithm. To adjust the intelligent agent For intelligent agents The degree of influence, when When the value is zero or close to zero, the effect is significantly reduced or prevented.

[0063] Step 8: Based on the MARL algorithm with integrated weight adjustment mechanism, train or drive the multi-agent system to make decisions in a collaborative task environment, and finally output a multi-agent collaborative control strategy with high robustness to improve the overall return of the multi-agent system in an environment with impulsive individuals.

[0064] It should be noted that the collaborative tasks performed by the multi-agent system are adversarial game tasks or other collaborative tasks in the simulation environment.

[0065] In step one of this invention, behavioral data of the agent is generated through a large language model or extracted from the interaction between the agent and the environment, and an impulsive personality score for each agent is calculated using an impulsive personality scale. The scores are then normalized for subsequent interaction weight adjustments.

[0066] It should be noted that the behavioral impulsivity scale in this invention is the Barrett Impulsivity Inventory (BIS-11) or an assessment standard adapted from that scale. Figure 5 Example of adapted quantity representation.

[0067] In step two of this invention, constructing a multi-agent interaction graph specifically involves constructing the interaction graph based on the agent's observation range or communication relationship. ,in To represent a diagram, It is a picture The agent is considered as a set of vertices. It is a picture The set of edges in the middle. For each pair of agents. Calculate its degree value , and the number of edges of the shortest path ,in , They refer to intelligent agents and The number of agent connections within the observable range. It refers to connecting intelligent agents and The number of edges in the shortest path between them.

[0068] In step three of this invention, the topological connection strength is calculated as follows:

[0069] .

[0070] In step seven of this invention, the multi-agent reinforcement learning algorithm is selected from value function decomposition-based algorithms, actor-critic-based algorithms, or attention mechanism-based algorithms.

[0071] The following is a detailed description of the implementation process of the multi-agent collaborative decision-making method based on behavioral characteristic quantitative modeling and weak connection mechanism provided by the present invention, using a specific implementation case.

[0072] Part 1: Quantitative Modeling of Impulsive Personality Based on LLM and AIMS-50, see [link / reference] Figure 2 This includes the following steps:

[0073] S11 uses large language models, such as GPT-4, Claude, and Qwen, which have strong scene understanding and role-playing capabilities to generate agent behavior data.

[0074] S12, Design Prompt Templates: Design structured, tiered prompts for specific mission scenarios (such as StarCraft II reconnaissance and resource gathering missions). For example, in an 8v8 multi-agent team battle: "You are an agent participating in an 8v8 battle. Choose your behavior based on your impulsiveness value: Impulsiveness 1-30: The agent is relatively rational, focusing on long-term goals; Impulsiveness 31-70: The agent balances novelty and stability; Impulsiveness 71-100: The agent seeks immediate gratification and novelty, prioritizing short-term rewards."

[0075] S13, Generate Behavioral Descriptions: Input cues of different impulse levels into the large language model, and generate a series of behavioral decision description texts of the agent in a specific situation based on the scale content.

[0076] S14, Impulsive Personality Score:

[0077] Human rating: Psychology experts or trained raters are invited to rate the generated behavioral decision description texts according to the AIMS-50 scale standards to assess the effect of impulsive personality shaping.

[0078] Automatic rating: Train a natural language processing-based model (such as a fine-tuned large model) by taking the above behavioral decision description text as input and outputting a predicted impulsive personality rating. This model can be trained under supervision using human rating data.

[0079] S15, Rating Normalization and Application: The obtained raw ratings (total scale score) are normalized to the [0,1] interval, serving as the quantitative rating of the agent's impulsive personality. .

[0080] Part Two: Construction and Application of Interaction Weight Adjustment Mechanism (Weak Connection), see [link / reference] Figure 3 This includes the following steps:

[0081] S21, Construct a multi-agent interaction graph and calculate topological attributes, establishing the multi-agent interaction graph based on the observation range, communication links, or environmental reachability between agents. ,in To represent a diagram, It is a picture The agent is considered as a set of vertices. It is a picture The set of edges in the middle. For each pair of agents. Calculate its degree value , and the number of edges of the shortest path ,in , They refer to intelligent agents and The number of agent connections within the observable range. It refers to connection and The number of edges in the shortest path between them.

[0082] S22, Topology connectivity strength calculation and strong / weak connectivity determination: The topology connectivity strength is calculated using the following method:

[0083] ,

[0084] Set threshold ,when When defined as a weak connection, it is set as follows: Otherwise, it is considered a strong connection and is set to... Weak connections are used to allow information to pass through, while strong connections are used to suppress the spread of unstable information.

[0085] S23, Constructing moderating weights based on impulsivity differences, defining moderating weights for impulsive personality such that the greater the impulsivity difference, the smaller the weight:

[0086] ,

[0087] in This is an adjustment coefficient used to enhance sensitivity to impulsive differences.

[0088] S24, forming the final effective weights, combines the weak connection filtering factor and the impulsive personality moderating weights to obtain the effective information interaction weights:

[0089] ,

[0090] The weight is determined by the agent. Information for intelligent agents The extent of its effective influence.

[0091] Part Three: Collaborative Performance Verification in the SMAC Environment. The following describes the test results of the method of this invention in a real multi-agent reinforcement learning environment. See [link to relevant documentation]. Figure 4 This includes the following steps:

[0092] S31. Select the test environment and algorithm framework. Use typical SMAC environment maps such as 8m, 3s5z, 1c3s5z, etc.; and select mainstream multi-agent reinforcement learning algorithms such as MAPPO and QMIX.

[0093] S32, construct a heterogeneous impulsive team, setting different proportions of impulsive personalities, such as 10%, 30%, and 60%, and designating some agents as highly impulsive. The rest are low impulsivity This creates a heterogeneous team.

[0094] S33 introduces a weak connection information exchange weight mechanism to dynamically calculate topological attributes during training. , Strong and weak connection filter factors Impulsiveness difference moderating weight Ultimate effective interaction weight And use it in the following modules:

[0095] Information fusion:

[0096] ,

[0097] in Represents intelligent agents The receiving agent The message conveyed Represents intelligent agents Observational information.

[0098] Attention mechanism:

[0099] ,

[0100] in These are the weighted attention parameters. These are the original attention parameters.

[0101] Centralized commentators are used for stable value function estimation.

[0102] S34, Experimental comparison and analysis, setting up a control group: using the standard MARL algorithm (MAPPO), the agents do not contain impulsive personality traits, or contain impulsive personality traits but no targeted adjustment mechanism is enabled, the communication weights or attention weights between agents are automatically learned by the network, and no topological constraints are imposed.

[0103] Experimental group: The method described in this invention was used. Specifically, in a heterogeneous team, 30% of the agents were configured to have high impulsivity scores, multi-agent interaction graph construction was enabled, and the weak-connection-based decision-making method proposed in this invention was applied. ) and impulsive personality differences ( The weight adjustment mechanism of ).

[0104] Test win rate During the evaluation phase, execution A complete match (Episode) is recorded, showing the number of wins. .

[0105] .

[0106] Strategy stability testing: After the statistical model converges, measure the standard deviation of the win rate or return over 100 training epochs. .

[0107] Experimental results show that the method of the present invention can still maintain a higher win rate and better policy stability in scenarios with a high proportion of impulsive agents, and its robustness is significantly improved.

[0108] Based on the above-mentioned inventive concept, the present invention also provides a multi-agent collaborative decision-making device based on behavioral characteristic quantitative modeling and weak connection mechanism, for implementing the above-mentioned multi-agent collaborative decision-making method based on behavioral characteristic quantitative modeling and weak connection mechanism, the device comprising:

[0109] The agent scoring module is used to acquire behavioral data of each agent in the multi-agent system, analyze the behavioral data using a predefined impulsive personality scale, and calculate a quantitative impulsive personality score for each agent.

[0110] The topology construction module is used to construct a multi-agent interaction graph at the current moment with agents as vertices, and to identify potential interaction pairs between agents through topology and communication range.

[0111] The connection strength calculation module is used to calculate the topological connection strength for potential interaction pairs in the multi-agent interaction graph;

[0112] The filter factor calculation module is used to set a connection strength threshold and determine a topology information filter factor based on the comparison between the topology connection strength and the connection strength threshold.

[0113] The moderating weight calculation module is used to calculate the moderating weight of impulsive personality based on the difference between the impulsive personality scores of the potential interaction pairs.

[0114] The effective weight calculation module is used to combine the topological information filtering factor with the impulsive personality adjustment weight to calculate the final effective information interaction weight.

[0115] An integration module is used to integrate the effective information interaction weights into the selected multi-agent reinforcement learning algorithm to train or drive the multi-agent system to make decisions in a collaborative task environment.

[0116] It is worth noting that this device embodiment corresponds to the above method embodiment. The implementation methods of the above method embodiments are all applicable to this device embodiment and can achieve the same or similar technical effects, so they will not be described in detail here.

[0117] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0118] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0119] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0120] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0121] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A multi-agent collaborative decision-making method based on behavioral characteristic quantitative modeling and weak connection mechanism, characterized in that, include: The behavioral data of each agent in the multi-agent system is acquired, and the behavioral data is analyzed using a predefined impulsive personality scale to calculate a quantitative impulsive personality score for each agent. Construct a multi-agent interaction graph at the current moment using agents as vertices, and identify potential interaction pairs between agents through topological relationships and communication ranges; For the potential interaction pairs in the multi-agent interaction graph, calculate the topological connection strength; Set a connection strength threshold, and determine a topology information filtering factor based on the comparison between the topology connection strength and the connection strength threshold; Based on the differences in impulsive personality scores between the potential interaction pairs, the impulsive personality moderating weights are calculated. The final effective information interaction weight is calculated by combining the topological information filtering factor with the impulsive personality adjustment weight. The effective information interaction weights are integrated into the selected multi-agent reinforcement learning algorithm to train or drive the multi-agent system to make decisions in a collaborative task environment.

2. The multi-agent collaborative decision-making method based on behavioral characteristic quantification modeling and weak connection mechanism according to claim 1, characterized in that, The acquisition of behavioral data for each agent in the multi-agent system includes: Generate behavioral data of agents through large language models or extract behavioral data of agents from their interactions with the environment.

3. The multi-agent collaborative decision-making method based on behavioral characteristic quantification modeling and weak connection mechanism according to claim 1, characterized in that, The behavioral impulsiveness scale uses the Barrett Impulsivity Scale or an assessment standard adapted from that scale. The quantitative impulsive personality score refers to normalizing the score obtained using the impulsive personality scale to the [0,1] interval.

4. The multi-agent collaborative decision-making method based on behavioral characteristic quantification modeling and weak connection mechanism according to claim 1, characterized in that, The topology connection strength is calculated as follows: ， in, For potential interaction relationships of intelligent agents The topological connectivity strength, and respectively intelligent agents and The degree in the multi-agent interaction graph, where the value is the agent's degree. and The number of agent connections within the observable range. It refers to connecting intelligent agents and The number of edges in the shortest path between them.

5. The multi-agent collaborative decision-making method based on behavioral characteristic quantification modeling and weak connection mechanism according to claim 4, characterized in that, A topology information filtering factor is determined based on a comparison between the topology connection strength and the connection strength threshold, including: Set a connection strength threshold ,when When this occurs, it is determined to be a weak connection, and the topology information filtering factor is adjusted. Otherwise, it is determined to be a strong connection, and the topology information filtering factor is set. .

6. The multi-agent collaborative decision-making method based on behavioral characteristic quantification modeling and weak connection mechanism according to claim 5, characterized in that, Based on the differences in impulsive personality scores between the potential interaction pairs, impulsive personality moderating weights are calculated, including: Adjust the weight of impulsive personality The design is based on the difference in impulsive personality scores. A monotonically decreasing function. and respectively intelligent agents and intelligent agents A quantitative score for impulsive personality.

7. The multi-agent collaborative decision-making method based on behavioral characteristic quantification modeling and weak connection mechanism according to claim 6, characterized in that, Impulsive personality regulation weight Represented as: ， or, ； in, It is a positive adjustment coefficient.

8. The multi-agent collaborative decision-making method based on behavioral characteristic quantification modeling and weak connection mechanism according to claim 6, characterized in that, The final effective information interaction weight is calculated by combining the topological information filtering factor with the impulsive personality adjustment weight, including: ， in, Weights for effective information exchange.

9. The multi-agent collaborative decision-making method based on behavioral characteristic quantification modeling and weak connection mechanism according to claim 1, characterized in that, The multi-agent reinforcement learning algorithm is selected from value function decomposition-based algorithms, actor-critic-based algorithms, or attention mechanism-based algorithms.

10. A multi-agent collaborative decision-making device based on behavioral characteristic quantitative modeling and weak connection mechanism, characterized in that, The apparatus for implementing the multi-agent collaborative decision-making method based on behavioral characteristic quantification modeling and weak connection mechanism as described in claim 1 includes: The agent scoring module is used to acquire behavioral data of each agent in the multi-agent system, analyze the behavioral data using a predefined impulsive personality scale, and calculate a quantitative impulsive personality score for each agent. The topology construction module is used to construct a multi-agent interaction graph at the current moment with agents as vertices, and to identify potential interaction pairs between agents through topology and communication range. The connection strength calculation module is used to calculate the topological connection strength for potential interaction pairs in the multi-agent interaction graph; The filter factor calculation module is used to set a connection strength threshold and determine a topology information filter factor based on the comparison between the topology connection strength and the connection strength threshold. The moderating weight calculation module is used to calculate the moderating weight of impulsive personality based on the difference between the impulsive personality scores of the potential interaction pairs. The effective weight calculation module is used to combine the topological information filtering factor with the impulsive personality adjustment weight to calculate the final effective information interaction weight. An integration module is used to integrate the effective information interaction weights into the selected multi-agent reinforcement learning algorithm to train or drive the multi-agent system to make decisions in a collaborative task environment.

Images