Multi-agent coordination method based on dynamic Nash equilibrium and related equipment thereof

Abstract

The invention discloses a multi-agent coordination method based on dynamic Nash equilibrium and related equipment thereof, belongs to the technical field of artificial intelligence, and is applied to task coordination scenes in the financial field or the health medical field. According to the method, the strategy space of each agent is discretized into a finite set, a revenue matrix based on historical data and domain knowledge is constructed, and a support enumeration method is adopted to efficiently solve Nash equilibrium, so that a selected strategy combination is ensured to have antagonistic stability and Pareto optimal characteristics; a rule engine and a strategy library are designed, a strategy is dynamically mapped based on state characteristics such as user emotion and task progress, a weighted voting mechanism and a priority rule are introduced to realize conflict coordination of multiple agents, the stability of strategy combination is further ensured in combination with a Nash equilibrium verification mechanism, and the robustness and consistency of the system in a dynamic environment are guaranteed. The method has remarkable advantages in scenes such as complex task scheduling, intelligent customer service collaboration and multi-robot collaboration, and has good expandability.

Description

Technical Field

[0001] This application belongs to the field of artificial intelligence technology, specifically relating to a multi-agent coordination method based on dynamic Nash equilibrium and related equipment. Background Technology

[0002] Against the backdrop of digital transformation and the rapid development of artificial intelligence technology, the need for multi-agent collaborative decision-making in complex systems (such as intelligent customer service, autonomous driving, and smart cities) is becoming increasingly urgent. However, traditional single-agent or simple rule-driven systems often face problems such as inefficiency, conflicting objectives, and poor user experience when dealing with dynamic, multi-objective, and highly conflicting scenarios. For example, in intelligent customer service scenarios, multiple agents (dialogue advancement agent, emotional soothing agent, risk control agent, etc.) need to simultaneously satisfy multiple objectives, but these objectives may conflict with each other. This multi-objective conflict is particularly prominent in highly sensitive fields such as finance and healthcare. Current mainstream multi-agent coordination methods rely on manually preset fixed rules (such as priority ranking), which cannot adapt to real-time changing environments, leading to rigid decision-making. Furthermore, when objectives conflict, "voting" or "forced priority" is often used, which may sacrifice the reasonable needs of some roles. Therefore, there is an urgent need for an intelligent coordination mechanism that can balance the interests of all parties and dynamically adapt to environmental changes. Summary of the Invention

[0003] The purpose of this application is to propose a multi-agent coordination method, device, computer equipment, and storage medium based on dynamic Nash equilibrium, in order to solve the technical problem that current mainstream multi-agent coordination methods rely on fixed rules preset by humans, cannot adapt to real-time changing environments, lead to rigid decision-making, and may sacrifice the reasonable needs of some roles when there is a conflict of objectives.

[0004] To address the aforementioned technical problems, this application provides a multi-agent coordination method based on dynamic Nash equilibrium, employing the following technical solution: A multi-agent coordination method based on dynamic Nash equilibrium includes: Obtain the current state characteristics of the multi-agent system, and define the role, objective, and discretized policy space of each agent, where the discretized policy space is a finite set of policies; For each agent, a domain-knowledge-based payoff matrix is ​​constructed, where the payoff matrix records the payoff values ​​under different policy combinations. Solve for the Nash equilibrium value based on the discretized strategy space and payoff matrix; Based on the Nash equilibrium value, the policy support probability distribution of each agent is calculated by enumeration, and a system of equations is established to ensure that the expected returns are equal on the supporting policies, thereby determining the stable policy combination. A dynamic coordination mechanism is designed based on the current state characteristics. The dynamic coordination mechanism and stable strategy combination are mapped through a rule engine. Conflicts are resolved by weighted voting and priority rules. The weight of the weighted voting is adjusted and normalized in real time according to the current state characteristics. For stable policy combinations, Nash equilibrium verification is performed. Based on the Nash equilibrium verification results, each agent is traversed to determine if there is a better policy. If so, a new policy combination is selected, and a convergence mechanism is introduced to monitor the policy change amplitude to avoid oscillations, until the change amplitude of all agents is lower than the preset threshold.

[0005] To address the aforementioned technical problems, this application also provides a multi-agent coordination device based on dynamic Nash equilibrium, employing the following technical solution: A multi-agent coordination device based on dynamic Nash equilibrium, comprising: The agent definition module is used to obtain the current state characteristics of the multi-agent system and define the role, objective, and discretized policy space of each agent, wherein the discretized policy space is a finite set of policies; The revenue matrix module is used to construct a domain-knowledge-based revenue matrix for each agent, where the revenue matrix records the revenue values ​​under different policy combinations. The Nash equilibrium module is used to solve for the Nash equilibrium value based on the discretized policy space and payoff matrix. The strategy combination module is used to calculate the policy support probability distribution of each agent by enumeration based on the Nash equilibrium value, and to establish a system of equations to ensure that the expected returns are equal on the supporting policies, thereby determining a stable strategy combination. The coordination mapping module is used to design a dynamic coordination mechanism based on the current state characteristics. It maps the combination of dynamic coordination mechanism and stable strategy through the rule engine, and uses weighted voting and priority rules to resolve conflicts. The weight of the weighted voting is adjusted and normalized in real time according to the current state characteristics. The policy update module is used to perform Nash equilibrium verification on stable policy combinations. Based on the Nash equilibrium verification results, it iterates through each agent to determine whether there is a better policy. If there is, a new policy combination is selected, and a convergence mechanism is introduced to monitor the policy change amplitude to avoid oscillations, until the change amplitude of all agents is lower than the preset threshold.

[0006] To address the aforementioned technical problems, this application also provides a computer device that employs the following technical solution: A computer device includes a memory and a processor, the memory storing computer-readable instructions, the processor executing the computer-readable instructions to implement the steps of the multi-agent coordination method based on dynamic Nash equilibrium as described in any of the preceding claims.

[0007] To address the aforementioned technical problems, this application also provides a computer-readable storage medium, employing the technical solution described below: A computer-readable storage medium storing computer-readable instructions, which, when executed by a processor, implement the steps of the multi-agent coordination method based on dynamic Nash equilibrium as described in any one of the preceding descriptions.

[0008] Compared with the prior art, the embodiments of this application have the following main advantages: This application discloses a multi-agent coordination method and related equipment based on dynamic Nash equilibrium, belonging to the field of artificial intelligence technology, and applicable to task collaboration scenarios in the financial or healthcare sectors. The proposed multi-agent coordination mechanism based on dynamic Nash equilibrium abandons reinforcement learning and gradient optimization methods, effectively avoiding problems such as training instability, convergence difficulties, and high computational overhead. By discretizing the policy space of each agent into a finite set, a payoff matrix based on historical data and domain knowledge is constructed, and a support enumeration method is used to efficiently solve the Nash equilibrium, ensuring that the selected policy combination possesses adversarial stability and Pareto optimality. To further improve system adaptability, a rule engine and policy library are designed, dynamically mapping policies based on user emotions, task progress, and other state features. A weighted voting mechanism and priority rules are introduced to achieve conflict coordination among multiple agents. Combined with a Nash equilibrium verification mechanism, the stability of the policy combination is further ensured, guaranteeing the robustness and consistency of the system in dynamic environments. This application has significant advantages in scenarios such as complex task scheduling, intelligent customer service collaboration, and multi-robot collaboration, and possesses good scalability. Attached Figure Description

[0009] To more clearly illustrate the solutions in this application, the accompanying drawings used in the description of the embodiments of this application will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1 An exemplary system architecture diagram is shown, in which this application can be applied; Figure 2 A flowchart is shown as an embodiment of a multi-agent coordination method based on dynamic Nash equilibrium according to this application; Figure 3 It shows Figure 2 A flowchart of an embodiment of step S204; Figure 4 A schematic diagram of the structure of an embodiment of a multi-agent coordination device based on dynamic Nash equilibrium according to this application is shown; Figure 5 It shows Figure 4 A schematic diagram of a embodiment of the strategy combination module 404; Figure 6 A schematic diagram of the structure of one embodiment of a computer device according to this application is shown. Detailed Implementation

[0011] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the terminology used herein in the specification of the application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application; the terms "comprising" and "having," and any variations thereof, in the specification, claims, and foregoing drawings of this application, are intended to cover non-exclusive inclusion. The terms "first," "second," etc., in the specification, claims, or foregoing drawings of this application are used to distinguish different objects, not to describe a particular order.

[0012] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0013] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings.

[0014] like Figure 1 As shown, system architecture 100 may include terminal device 101, network 102, and server 103. Terminal device 101 may be a laptop 1011, tablet 1012, or mobile phone 1013. Network 102 is used as a medium to provide a communication link between terminal device 101 and server 103. Network 102 may include various connection types, such as wired, wireless communication links, or fiber optic cables.

[0015] Users can use terminal device 101 to interact with server 103 via network 102 to receive or send messages, etc. Various communication client applications can be installed on terminal device 101, such as web browser applications, shopping applications, search applications, instant messaging tools, email clients, social media platform software, etc.

[0016] Terminal device 101 can be various electronic devices with a display screen and support web browsing. In addition to laptops 1011, tablets 1012, or mobile phones 1013, terminal device 101 can also be an e-book reader, an MP3 player (Moving Picture Experts Group Audio Layer III), an MP4 player (Moving Picture Experts Group Audio Layer IV), a laptop computer, and a desktop computer, etc.

[0017] Server 103 can be a server that provides various services, such as a backend server that provides support for the pages displayed on terminal device 101.

[0018] It should be noted that the multi-agent coordination method based on dynamic Nash equilibrium provided in this application is generally executed by a server / terminal device, and correspondingly, the multi-agent coordination device based on dynamic Nash equilibrium is generally set in the server / terminal device.

[0019] It should be understood that Figure 1 The number of terminal devices, networks, and servers shown is merely illustrative; the system can have any number of terminal devices, networks, and servers depending on implementation needs.

[0020] Continue to refer to Figure 2 The diagram illustrates a flowchart of an embodiment of a multi-agent coordination method based on dynamic Nash equilibrium according to this application. The multi-agent coordination method based on dynamic Nash equilibrium includes the following steps: S201, Obtain the current state characteristics of the multi-agent system, and define the role, objective, and discretized policy space of each agent, wherein the discretized policy space is a finite set of policies; Specifically, the first step involves real-time data collection of the operating environment of the multi-agent system to obtain current state characteristics, including but not limited to environmental state parameters (such as task progress, resource utilization, risk level, and system load), individual behavioral characteristics (such as agent response time, strategy execution history, and behavioral preferences), and external input signals (such as user emotion ratings and task urgency). To ensure that the decision-making objectives of each agent are differentiated and coordinated, this step requires clarifying the role definition and task objectives of each agent. For example, in an intelligent customer service system, the task-driving agent aims to complete the task with the shortest response time, the emotion-regulating agent aims to improve user satisfaction, and the risk-controlling agent aims to ensure compliance and minimize risk. After defining roles and objectives, the continuous policy space is transformed into a finite policy set through discretization. Discretization can be based on policy clustering, benefit stratification, or expert experience, dividing continuous control variables (such as response intensity and intervention timing) into several discrete intervals to form a finite policy set (aggressive policies, balanced policies, conservative policies, etc.) to reduce the complexity of policy solution.

[0021] S202, construct a domain-knowledge-based payoff matrix for each agent, where the payoff matrix records the payoff values ​​under different policy combinations; Specifically, by combining domain knowledge and historical interaction data, a corresponding payoff matrix is ​​constructed for each agent. The rows and columns of the payoff matrix represent different strategy combinations, and the elements record the payoff value that the agent can obtain under a specific strategy combination. Multi-dimensional features are introduced during payoff modeling, such as immediate task payoff, long-term trust payoff, user feedback score, and risk penalty coefficient. Payoff calculation can use a weighted linear combination model, i.e.: Payoff value = α × Immediate payoff + β × Long-term payoff - γ × Risk cost, where α, β, and γ are adjustable weight parameters used to reflect the trade-off relationship between different objectives. For example, a payoff matrix is ​​defined for each agent, recording the payoff values ​​under different strategy combinations. The profit function = User satisfaction × 0.8 + Long-term trust level × 0.2

[0022] Domain knowledge can be introduced through expert annotation rules, empirical formulas, or statistical regression models. For example, in a financial customer service scenario, if a user is at a high risk level, the reward weight of the risk control agent will be significantly increased; if the user's emotion score is low, the reward contribution of the emotion regulation agent will be higher. Furthermore, the system can dynamically update the reward matrix based on historical data, enabling it to adapt to environmental changes and the evolution of agent behavior patterns. This reward matrix is ​​constructed through...

[0023] S203, Solve for the Nash equilibrium value based on the discretized strategy space and payoff matrix; Specifically, based on a defined finite set of policies and the payoff matrices of each agent, the equilibrium point of the system is calculated using the Nash equilibrium solution method in game theory. The solution process can be implemented through enumeration, support set search, or linear programming-based methods. First, assuming the probability distribution of each agent's policy selection within its discretized policy space, a joint policy space for all agents is constructed. Next, for each agent, with the policy distributions of other agents fixed, its expected payoff under different policies is calculated, resulting in a payoff vector. According to the Nash equilibrium condition, if a certain policy combination ensures that any agent's unilateral policy change will not increase its expected payoff, then this combination is the Nash equilibrium point. An iterative algorithm can be introduced during the solution process to gradually adjust the policy distributions of each agent until the equilibrium condition is met. To improve solution efficiency, subgames can be solved locally first (e.g., agents can be grouped by functional modules) and then integrated into a global equilibrium. Furthermore, constraints (such as resource limits or risk boundaries) can be introduced to ensure that the obtained Nash equilibrium satisfies both local optima and system-level executability and stability.

[0024] S204. Based on the Nash equilibrium value, the policy support probability distribution of each agent is calculated by enumeration, and a system of equations is established to ensure that the expected returns are equal on the supporting policies, thereby determining the stable policy combination. Specifically, after obtaining the Nash equilibrium value, the policy support probability distribution of each agent is further calculated using an enumeration method to obtain stable policy combinations. First, a complete enumeration of the finite policy set for each agent is performed to generate all possible policy combinations, and their corresponding joint payoffs are calculated. Then, by traversing the payoff matrix, candidate policy sets satisfying the Nash equilibrium condition are identified. For each candidate policy set, a system of expected payoff equations is established to ensure that the expected payoffs of agents on their supporting policies are equal, thereby maintaining the equilibrium state. Mathematically, for any agent i, the expected payoffs of its supporting policies a and b should satisfy E_i(a) = E_i(b); otherwise, the corresponding probability distributions need to be adjusted. This system of equations can be obtained through linear algebra methods or numerical solutions. The obtained probability distribution represents the probability of each policy being selected in the equilibrium state, i.e., the policy support probability distribution of the agent. To ensure the robustness of the results, the stability of the solution space can be checked, eliminating policy combinations that do not satisfy payoff balance or have excessive fluctuations. Ultimately, the resulting stable policy combination will serve as the policy basis for system execution, guiding coordinated decision-making among multiple agents.

[0025] S205, a dynamic coordination mechanism is designed based on the current state characteristics. The dynamic coordination mechanism and stable strategy combination are mapped through the rule engine, and weighted voting and priority rules are used to resolve conflicts. The weight of the weighted voting is adjusted and normalized in real time according to the current state characteristics. Specifically, a dynamic coordination mechanism based on a rule engine is constructed to achieve an adaptive mapping between stable policy combinations and real-time environmental states. When conflicts are detected among multiple agents in policy selection, the system triggers a weighted voting mechanism for coordination and decision-making. Specifically, firstly, the weight parameters of each agent are determined based on current state characteristics (such as task urgency, risk level, user sentiment, etc.). Weights can be calculated using rule functions or machine learning models, such as w_i=f(sentiment score, risk level, task progress), and normalized to ensure the sum of all weights is 1. Then, each agent submits a vote based on its own policy preferences, and the system calculates the weighted score for each candidate policy combination, for example, the total score: in, The weights of Agent(i) Let be the payoff value of the strategy combination in state (s).

[0026] When multiple policy scores are close or conflict, a decision is made based on priority rules, prioritizing the policy of the higher-priority role (such as the compliance control agent). Furthermore, to prevent local optima from causing overall performance degradation, a "dynamic priority boosting mechanism" can be introduced into the rule engine, adjusting priority labels in real time based on state changes. Ultimately, the coordination result is mapped to specific actions and fed back to the rule engine to update the knowledge base, achieving continuous adaptive optimization of the system.

[0027] S206. Perform Nash equilibrium verification on stable policy combinations. Based on the Nash equilibrium verification results, iterate through each agent to determine if there is a better policy. If there is, reselect the policy combination and introduce a convergence mechanism to monitor the policy change amplitude to avoid oscillations, until the change amplitude of all agents is lower than the preset threshold.

[0028] Specifically, the selected stable policy combination undergoes Nash equilibrium verification and convergence monitoring to ensure the long-term stability of the system. First, the current stable policy combination is traversed agent-by-agent. With other agents' policies fixed, the potential gains of a single agent deviating from the current policy are calculated. If an agent is found to have an alternative policy that could improve its gains, the current combination is considered not to have reached a true equilibrium state, and the policy combination needs to be reselected or adjusted. Adjustments can employ local search or probabilistic update mechanisms to balance exploration and stability. Simultaneously, to prevent frequent oscillations between multiple equilibrium points, a convergence mechanism is introduced to monitor the policy change magnitude Δ. Δ can be defined as a quantitative index of the difference in agent policy distribution between two consecutive iterations. When the Δ of all agents is less than a preset threshold, the system is considered to have converged; otherwise, iteration continues. To further improve convergence speed and stability, a damping coefficient or temperature decay function can be introduced to regulate the policy update magnitude. Once the system reaches a stable state, the final policy combination will be confirmed as a global dynamic Nash equilibrium solution.

[0029] For example, the system executes priority rules, checks whether the policies of the core intelligent agent (such as the compliance agent) conflict, executes the highest priority policy first, and performs a weighted vote on the policies of the remaining agents to select the policy combination with the highest weighted score. For the final strategy combination Iterate through each agent (Agent(i)) and verify whether a policy exists. Make > If it exists, then a new strategy combination is selected, where, Describes the profit function of Agent(i). In order to consider new strategies for change, This represents the equilibrium strategy combination for all agents other than Agent(i). This represents the current equilibrium strategy of Agent(i); Furthermore, a convergence threshold or cooling-off period is introduced to prevent the strategy portfolio from oscillating infinitely between multiple equilibrium points. For each Agent(i), its policy space is A = { , Then the strategy value after the t-th iteration Each element Indicates the selection action The probability, and the quantitative indicator of the magnitude of change Δ =The strategy value after the t-th iteration The policy value after the (t-1)th iteration A threshold is preset based on requirements to represent the minimum allowable range of policy changes, and the policy change range Δ for all agents is calculated after each policy adjustment. If the Δ of all agents If the value is less than or equal to the preset threshold, convergence is considered achieved, and adjustment is stopped; otherwise, the iterative adjustment strategy continues.

[0030] Furthermore, the steps of obtaining the current state characteristics of the multi-agent system and defining the role, objective, and discretized policy space of each agent specifically include: The current state features are extracted from real-time data through a pre-trained model, where the current state features include at least sentiment score and risk level; Define a role and a goal for each agent, where the goal is a locally optimized objective. The continuous policy of each agent is converted into a finite set of policies, wherein the finite set of policies includes at least aggressive policies, balanced policies, and deferred policies.

[0031] In this embodiment, features are first extracted from the real-time input data of the multi-agent system based on a pre-trained deep representation model (such as BERT, Transformer, or graph neural network model) to construct a current state feature vector. This feature vector comprehensively reflects the system's environmental situation and interaction state at the current moment, specifically including emotion rating, risk level, task progress, resource utilization, and external environmental fluctuation parameters. Subsequently, for each agent, the system defines a corresponding objective function based on its functional role in the task system (such as task advancement, emotion regulation, risk control, resource allocation, etc.). The objective function is usually the agent's local optimal goal and can be adjusted by weighting through parameterization. To reduce the complexity of policy solution and improve decision response speed, continuous policy control variables are converted into a finite set of policies through discretization. Common discretization types include aggressive policies (prioritizing short-term gains), balanced policies (considering both short-term and long-term goals), and deferred policies (risk minimization-oriented). Through the discretization of the policy space and the explicit definition of role goals, the system achieves comparability among multiple agents and consistency of policy constraints.

[0032] Through the above steps, accurate state modeling and policy space standardization of complex multi-agent environments can be achieved, significantly improving the system's coordinated solution efficiency and policy convergence stability.

[0033] Further, please refer to Figure 3 The steps involved in determining a stable policy combination, specifically, include: calculating the policy support probability distribution for each agent using an enumeration method based on the Nash equilibrium value, and establishing a system of equations to ensure that expected returns are equal across supporting policies. S301, enumerate the finite policy set for each agent, construct the set of all possible policy combinations, and record the corresponding joint reward vector; S302, based on the payoff matrix, calculates the individual payoff of each agent under each policy combination, and obtains the relative payoff of each agent under different policies through normalization. S303, Based on the enumeration results, extract candidate policy combinations that satisfy the Nash equilibrium condition, and calculate the support policy set for each candidate policy combination; S304. For each agent's support policy set, establish a set of expected return equations, where the set of expected return equations is used to constrain the expected returns of all support policies to be equal under equilibrium conditions. S305, the probability distribution values ​​corresponding to each support strategy are obtained by solving the linear equation system or the Lagrange multiplier method, thus obtaining the policy support probability distribution of each agent. S306, perform a stability test on the strategy support probability distribution. If there are strategies that do not meet the payoff balance, remove them and renormalize the probability distribution. S307, under the condition of satisfying the balance of expected returns of all agents, determine the final stable strategy combination.

[0034] In this embodiment, after obtaining the initial Nash equilibrium value, the system performs a comprehensive search of the finite policy set for each agent using an enumeration method to ensure that no possible equilibrium solution is overlooked in the discretized policy space. Specifically, firstly, based on the policy sets of each agent, all possible policy combinations are generated and the corresponding joint payoff vectors are calculated to form a global payoff matrix. Subsequently, the system calculates the individual payoff of each agent under each policy combination based on the payoff matrix, and obtains a more comparable relative payoff index through normalization processing, so that agents with different objective functions can be analyzed for equilibrium under the same payoff scale. On this basis, the system extracts a set of candidate policies that satisfy the Nash equilibrium condition, i.e., there is no combination in which any single agent can improve its own payoff by unilaterally changing its policy. For each candidate combination, the support policy set is further calculated, and an expected payoff equation system is established. This equation system constrains the expected payoffs of the support policies to be equal under equilibrium conditions. The solution stage can use a linear equation system solution method or an optimization solution based on the Lagrange multiplier method to obtain the probability distribution values ​​corresponding to each support policy. After obtaining the initial distribution, the system performs a stability check. If a strategy is found to violate the payoff balance condition or cause excessive payoff fluctuations, it is discarded and the distribution is renormalized until the expected payoff difference among all agents is below a set threshold. The final stable strategy combination serves as the input for the dynamic coordination mechanism and can be directly used for system strategy execution and dynamic weight mapping.

[0035] Through the above steps, global equilibrium solutions and stable policy selection in the multi-agent policy space can be achieved, ensuring that the coordination results achieve the optimal balance between revenue and system stability.

[0036] Furthermore, a dynamic coordination mechanism is designed based on the current state characteristics. A rule engine maps the dynamic coordination mechanism to a combination of stable strategies, and a conflict resolution process using weighted voting and priority rules is employed. Specifically, this includes: Based on the current state characteristics, environmental variables, agent behavior characteristics, and global system constraints are extracted to construct the input parameter set of the dynamic coordination mechanism; A rule engine is used to establish a mapping relationship between current state characteristics and stable policy combinations. Each mapping rule includes a triggering condition, a priority label, and a corresponding coordination action. When multiple agents propose conflicting policy choices, a weighted voting mechanism is triggered to aggregate the policy preferences of each agent. The voting weights are dynamically calculated based on the importance of each agent's role in the system, its historical performance, and its risk level, and then the weights are normalized. The weighted voting results are determined by combining priority rules. When there are ties or similar benefits, the decision is made based on the priority label. The consistency of the coordination result determined by the voting decision is verified. If there is a conflict with the global constraints, the rule engine is invoked to rematch the backup coordination strategy. The final coordinated output is used as the system execution instruction, and the current state characteristics and the final coordinated result are stored together to update the rule engine knowledge base.

[0037] In this embodiment, the system uses a dynamic coordination mechanism to resolve policy conflicts and optimize real-time decisions among multiple agents. First, key input parameters, including environmental load, task progress, risk level, and emotion score, are extracted from environmental data, agent behavior logs, and system constraints based on current state characteristics, constructing a set of input parameters for the coordination mechanism. Then, a rule engine is used to establish a mapping relationship between state characteristics and stable policy combinations. Each rule includes triggering conditions, priority labels, and corresponding coordination actions. For example, when a risk level is detected to be higher than a threshold, the risk control agent's policy response is triggered first; when a user's emotion score is low, the emotion-soothing strategy is activated first. When multiple agents propose conflicting policies, the system automatically initiates a weighted voting mechanism, aggregating the policy preferences of each agent and calculating a weighted score to form a voting result. The voting weights are dynamically adjusted based on factors such as the agent's role importance in the system, historical performance, and risk level, and normalization ensures voting fairness. If scores are identical or benefits are close, the system will make a final decision based on the priority label. Next, the winning strategy undergoes consistency verification to ensure it does not violate global constraints, such as resource limits or task dependencies. If verification fails, the rules engine will automatically invoke a backup coordination strategy to guarantee system continuity. Finally, the determined coordination result is not only translated into system execution instructions but also synchronously stored in the rules engine's knowledge base.

[0038] Through the above steps, adaptive coordination and dynamic decision optimization of policy conflicts among multiple agents can be achieved, thereby improving the real-time responsiveness and overall collaborative stability of the system.

[0039] Furthermore, when multiple agents propose conflicting policy choices, a weighted voting mechanism is triggered to aggregate the policy preferences of each agent. This process specifically includes: The system detects the current policy output of each agent under a stable policy combination. If there is a conflict between the profit target or the action to be executed, it is determined that a conflict event has occurred and a weighted voting mechanism is initiated. Identify the agents involved in the conflict, extract their role attributes, priority parameters, and historical voting records, and construct a set of voting participants; Based on the risk level, emotion score, and resource utilization rate contained in the current state characteristics, calculate the dynamic voting weight of each agent participating in the conflict; Obtain the selection signals submitted by each participating agent based on its own policy preferences, and calculate the weighted score of each candidate policy according to the dynamic voting weight; The weighted scores of all candidate strategies are sorted, and the strategy with the highest score is selected as the strategy coordination result.

[0040] In this embodiment, during the multi-agent collaboration process, when the system detects conflicting policy outputs under a stable policy combination, a weighted voting mechanism is immediately triggered to dynamically resolve the conflict. To ensure the robustness of the voting process, the agents involved in the conflict are first identified, and their role attributes, priority parameters, and historical voting records are extracted to form a set of voting participants, making the decision traceable and interpretable. Subsequently, based on variables such as risk level, sentiment score, and resource occupancy rate in the current state characteristics, the voting weight of each agent involved in the conflict is dynamically calculated. These weights need to be normalized through a state-driven decision model to ensure that the sum is 1 and that it is robust to extreme states. The policy preference signals submitted by the agents are thus mapped onto candidate policies. The system defines the score of each candidate policy as a weighted average of its predicted benefit value and conflict cost under the corresponding voting weight, forming a weighted score. Subsequently, the weighted scores of all candidate strategies are ranked, and the strategy with the highest score is selected first. In case of ties or boundary situations, the system will introduce priority rules (such as the strategy priority of the core agent, historical stability, and contribution to the global goal) for arbitration to avoid strategy fluctuations caused by a single vote. Throughout the process, the stability of the voting results is continuously monitored during runtime. If distribution drift or drastic weight changes occur several times in a row, an adaptive threshold adjustment or rollback mechanism will be triggered to prevent the system from entering a state of oscillation. To improve interpretability, the system logs the voting process and outputs intermediate results of conflict resolution, and provides conflict heatmaps when necessary to help developers and operations personnel quickly locate the root cause of the conflict and make policy-level optimizations. In addition, this mechanism supports distributed processing of concurrent conflict events, ensuring real-time performance and scalability in large-scale multi-agent scenarios. Through the above design, policy coordination in conflict scenarios will achieve fast, stable, and traceable conflict resolution while preserving the core global goal.

[0041] Through the above steps, rapid and traceable strategy coordination in conflict situations is achieved, improving the system's adaptability and stability to dynamic environments.

[0042] Furthermore, Nash equilibrium verification is performed on stable policy combinations. Based on the Nash equilibrium verification results, each agent is traversed to determine if a better policy exists. If so, a new policy combination is selected, and a convergence mechanism is introduced to monitor the magnitude of policy changes to avoid oscillations, until the magnitude of changes for all agents is lower than a preset threshold. Specifically, this includes the following steps: Based on the current stable strategy combination, calculate the expected reward value of each agent and record it as the benchmark reward vector; For each agent, traverse the set of feasible policies and, with the policies of other agents remaining unchanged, calculate the potential gain of the target agent after deviating alone. Here, the target agent is any agent in the system. The potential reward of each agent is compared with the benchmark reward. If there is a strategy that can improve the reward, it is marked as an improved strategy and the corresponding reward gain is recorded. Based on the profit gains of the improveable strategies and the global system constraints, select a strategy combination that significantly improves profits without compromising global stability for replacement and update. After each policy update, the magnitude of the change in the payoff of all agents is recalculated, and a convergence monitoring mechanism is introduced to track the dynamic trend of the payoff of each agent. If the change in the payoff of any agent exceeds a preset threshold in multiple consecutive iterations, the damping adjustment mechanism is activated to adaptively reduce the policy update step size. When the change in the payoff of all agents is lower than the preset threshold, and the strategy combination remains stable after multiple iterations, the system is determined to have reached convergence and the final Nash equilibrium strategy combination is output.

[0043] In this embodiment, the key to implementing the Nash equilibrium verification and convergence mechanism for stable policy combinations lies in constructing a complete iterative policy modification process to ensure a balance between local search and global stability. First, based on the current stable policy combination, the system needs to accurately calculate the baseline expected payoff vector for each agent. Considering multi-objective weights, external constraints, and resource limitations, each dimension of the vector can correspond to an objective function (such as satisfaction, trust, energy consumption, completion time, etc.), and these dimensions are standardized for cross-agent comparison. Then, with other agent policies remaining unchanged, each agent's optional policies are enumerated one by one, calculating the potential payoff gain if the agent deviates from its current policy. This step requires the payoff matrix and the opponent's policy distribution as input, utilizing the opponent's fixed policy approximation to avoid computational explosion in high-dimensional global enumeration. By comparing the baseline payoff, agents with improvement potential and their corresponding payoff gains are identified. Then, using global constraints (such as system stability, conflict cost, failure risk threshold, etc.) as constraints, policy replacement combinations that significantly improve payoff without compromising global stability are selected. To ensure robustness, the system re-evaluates the magnitude of payoff changes for all agents after each policy update, outputs convergence trend indicators, and uses visualized change trajectories (such as convergence curves of probability distributions and fluctuation ranges of payoff vectors) to aid in the analysis of the convergence process. If the magnitude of payoff change for any agent consistently exceeds a threshold over several iterations, a damping adjustment mechanism is triggered to adaptively decrease the policy update step size, reducing the risk of oscillations during iterations. If the magnitude of payoff changes for all agents remains below the threshold for an extended period and the policy combination remains stable, convergence is determined, and the final Nash equilibrium policy combination is output, while simultaneously recording the convergence time, number of iterations, and convergence path. This process not only enables refined exploration of local improvements but also ensures that new policies do not disrupt the overall stability and coordination of the system by introducing global constraints, while providing traceable convergence evidence and interpretable convergence paths.

[0044] Through the above steps, a controllable convergence and an interpretable improvement path for a stable strategy are achieved.

[0045] Furthermore, based on the profit gains of the improveable strategies and the global constraints of the system, the step of selecting a strategy combination that significantly improves profits without compromising global stability for replacement and updating specifically includes: Agents marked as having improvable strategies are screened, and the incremental benefit of each candidate strategy under the current global strategy combination and its impact on the overall objective function of the system are calculated. Based on the global system constraints, a joint optimization model of revenue gain and system stability constraints is established. The system stability constraints include resource limitations, risk tolerance, and coordination consistency indicators. Solve the joint optimization model, select a set of candidate strategies that significantly improve returns and satisfy global constraints, and calculate the global stability coefficient for each candidate strategy; Based on the trade-off between global stability coefficient and profit gain, a multi-objective ranking algorithm is used to determine the optimal combination of replacement strategies. The optimal replacement strategy combination is compared with the existing stable strategy combination. If the overall return increases by more than a threshold and the system stability does not decrease, then the strategy replacement update is performed.

[0046] In this embodiment, the core of the replacement and update step based on the gains of the improveable strategy and the global system constraints lies in jointly optimizing local improvements and global stability to ensure that the collaborative performance of the entire multi-agent system in a dynamic environment is not compromised by local optimization. First, when screening agents marked as having improveable strategies, the marginal gain increment of each candidate strategy under the current global strategy combination needs to be calculated in detail. The impact of this increment on the global system is then evaluated in conjunction with the system objective function (such as overall satisfaction, resource utilization, task completion rate, delay cost, etc.). Second, a joint optimization model needs to be established, integrating the gains and global stability constraints as objectives and constraints. The stability constraints should cover multiple dimensions such as resource constraints, risk tolerance, coordination consistency, and cost thresholds for hedging conflicts, and adjustable weight coefficients should be set to reflect the current system's priorities and risk preferences. The model is then solved to obtain a set of candidate strategies, and a global stability coefficient is calculated for each candidate strategy. The stability coefficient can be determined based on quantitative indicators such as the fluctuation of the global objective function, cross-agent consistency measurement, and the duration of hedging conflicts. Next, multi-objective ranking algorithms (such as the Analytic Hierarchy Process, TOPSIS, and entropy weighting) are used to weigh the gains and stability coefficients, ranking the optimal replacement strategy combinations. Finally, the optimal combination is compared with the current stable strategy. If the new combination reaches a set threshold in global gain without reducing global stability, the replacement update can be executed. Simultaneously, the system should record key intermediate results, change trajectories, and triggering conditions during the replacement process. At the implementation level, the atomicity of strategy replacement, rollback mechanisms, and impact assessment on the next round of Nash equilibrium solution need to be considered to ensure that the replacement does not introduce new instabilities or conflicts. This process should also have robust anomaly detection and fault tolerance to avoid single-point failures causing global update failures. Through the above design, the replacement update can achieve more efficient collaborative decision-making while ensuring global stability.

[0047] Through the above steps, controllable replacement and updates and stability protection under global constraints are achieved, thereby improving the collaborative robustness of the system.

[0048] In the above embodiments, this application discloses a multi-agent coordination method based on dynamic Nash equilibrium, belonging to the field of artificial intelligence technology, and applied to task collaboration scenarios in the financial or healthcare fields. The multi-agent coordination mechanism based on dynamic Nash equilibrium proposed in this application abandons reinforcement learning and gradient optimization methods, effectively avoiding problems such as training instability, convergence difficulties, and high computational overhead. By discretizing the policy space of each agent into a finite set, a payoff matrix based on historical data and domain knowledge is constructed, and the support enumeration method is used to efficiently solve the Nash equilibrium, ensuring that the selected policy combination possesses adversarial stability and Pareto optimality. To further improve system adaptability, a rule engine and policy library are designed, dynamically mapping policies based on user emotions, task progress, and other state features. A weighted voting mechanism and priority rules are introduced to achieve conflict coordination among multiple agents. Combined with a Nash equilibrium verification mechanism, the stability of the policy combination is further ensured, guaranteeing the robustness and consistency of the system in dynamic environments. This application has significant advantages in scenarios such as complex task scheduling, intelligent customer service collaboration, and multi-robot collaboration, and possesses good scalability.

[0049] In this embodiment, the electronic device (e.g., based on the dynamic Nash equilibrium multi-agent coordination method) operates on it. Figure 1 The server shown can receive instructions or acquire data via wired or wireless connection. It should be noted that the aforementioned wireless connection methods may include, but are not limited to, 3G / 4G connections, WiFi connections, Bluetooth connections, WiMAX connections, Zigbee connections, UWB (ultra-wideband) connections, and other currently known or future wireless connection methods.

[0050] It should be emphasized that, to further ensure the privacy and security of the above strategy combination information, the above strategy combination information can also be stored in a node of a blockchain.

[0051] The blockchain referred to in this application is a novel application model of computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanisms, and encryption algorithms. Essentially, a blockchain is a decentralized database, a chain of data blocks linked together using cryptographic methods. Each data block contains information about a batch of network transactions, used to verify the validity of the information (anti-counterfeiting) and generate the next block. A blockchain can include an underlying blockchain platform, a platform product service layer, and an application service layer.

[0052] The embodiments of this application can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence (AI) refers to the theories, methods, technologies, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0053] Foundational technologies for artificial intelligence generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, operating / interactive systems, and mechatronics. AI software technologies mainly encompass computer vision, robotics, biometrics, speech processing, natural language processing, and machine learning / deep learning.

[0054] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by instructing related hardware through computer-readable instructions. These computer-readable instructions can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above methods. The aforementioned storage medium can be a non-volatile storage medium such as a magnetic disk, optical disk, or read-only memory (ROM), or random access memory (RAM).

[0055] It should be understood that although the steps in the flowcharts of the accompanying figures are shown sequentially as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the accompanying figures may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the sub-steps or stages of other steps.

[0056] Further reference Figure 4 As a response to the above Figure 2 To implement the method shown, this application provides an embodiment of a multi-agent coordination device based on dynamic Nash equilibrium. This device embodiment is similar to... Figure 2 Corresponding to the method embodiments shown, this device can be specifically applied to various electronic devices.

[0057] like Figure 4 As shown, the multi-agent coordination device 400 based on dynamic Nash equilibrium described in this embodiment includes: The agent definition module 401 is used to obtain the current state characteristics of the multi-agent system and define the role, objective and discretized policy space of each agent, wherein the discretized policy space is a finite set of policies; The revenue matrix module 402 is used to construct a domain knowledge-based revenue matrix for each agent, wherein the revenue matrix records the revenue values ​​under different policy combinations. Nash equilibrium module 403 is used to solve for the Nash equilibrium value based on the discretized policy space and the payoff matrix; The strategy combination module 404 is used to calculate the policy support probability distribution of each agent by enumeration based on the Nash equilibrium value, and to establish a system of equations to ensure that the expected returns are equal on the supporting policies, thereby determining a stable strategy combination. The coordination mapping module 405 is used to design a dynamic coordination mechanism based on the current state characteristics. It maps the combination of dynamic coordination mechanism and stable strategy through the rule engine, and uses weighted voting and priority rules to resolve conflicts. The weight of the weighted voting is adjusted and normalized in real time according to the current state characteristics. The policy update module 406 is used to perform Nash equilibrium verification on stable policy combinations. Based on the Nash equilibrium verification results, it iterates through each agent to determine whether there is a better policy. If there is, it reselects the policy combination and introduces a convergence mechanism to monitor the policy change amplitude to avoid oscillations until the change amplitude of all agents is lower than the preset threshold.

[0058] Furthermore, the agent definition module 401 specifically includes: The feature extraction unit is used to extract current state features from real-time data through a pre-trained model, wherein the current state features include at least sentiment score and risk level; The agent definition unit is used to define roles and objectives for each agent, where the objectives are independently optimized local objectives; The policy conversion unit is used to convert the continuous policy of each agent into a finite set of policies, wherein the finite set of policies includes at least aggressive policies, balanced policies, and deferred policies.

[0059] Further, please refer to Figure 5 The strategy combination module 404 specifically includes: The policy enumeration unit 501 is used to enumerate the finite policy set for each agent, construct the set of all possible policy combinations, and record the corresponding joint reward vector. Individual benefit unit 502 is used to calculate the individual benefit of each agent under each strategy combination based on the benefit matrix, and obtain the relative benefit of each agent under different strategies through normalization processing. The candidate computation unit 503 is used to extract candidate strategy combinations that satisfy the Nash equilibrium condition based on the enumeration results, and to compute a support strategy set for each candidate strategy combination. The benefit equation unit 504 is used to establish a set of expected benefit equations for each agent's set of support policies, wherein the set of expected benefit equations is used to constrain the expected benefits of all support policies to be equal under equilibrium conditions. The probability distribution unit 505 is used to obtain the probability distribution value corresponding to each support policy by solving a system of linear equations or by the Lagrange multiplier method, so as to obtain the policy support probability distribution of each agent. Stability testing unit 506 is used to test the stability of the strategy support probability distribution. If there is a strategy that does not meet the payoff balance, it is removed and the probability distribution is renormalized. The payoff balancing unit 507 is used to determine the final stable strategy combination under the condition that the expected payoffs of all agents are balanced.

[0060] Furthermore, the coordination mapping module 405 specifically includes: The input parameter unit is used to extract environmental variables, agent behavior features, and global system constraints based on the current state characteristics, and to construct the input parameter set of the dynamic coordination mechanism. The mapping relationship unit is used to establish a mapping relationship between the current state characteristics and the stable policy combination using the rule engine. Each mapping rule includes a triggering condition, a priority label, and a corresponding coordination action. The weighted voting unit is used to trigger the weighted voting mechanism when multiple agents propose conflicting policy choices, and to summarize the policy preferences of each agent. The voting weight unit is used to dynamically calculate the voting weight based on the importance of each agent's role in the system, its historical performance, and its risk level, and to normalize the weight. The decision-making unit is used to determine the weighted voting results in combination with priority rules. When there are ties or similar benefits, the decision-making is carried out based on the priority label. The consistency verification unit is used to verify the consistency of the coordination result determined by the voting decision. If there is a conflict with the global constraints, the rule engine is called to rematch the backup coordination strategy. The rule update unit is used to take the final determined coordination output as the system execution instruction and store the current state characteristics and the final coordination result together to update the rule engine knowledge base.

[0061] Furthermore, the weighted voting unit specifically includes: The action conflict subunit is used to detect the current policy output of each agent under a stable policy combination. If there is a conflict between the benefit objective or the execution of actions, it is determined that a conflict event has occurred and a weighted voting mechanism is initiated. The conflict identification subunit is used to identify the intelligent agents involved in the conflict, extract role attributes, priority parameters and historical voting records, and construct a set of voting participants; The dynamic voting subunit is used to calculate the dynamic voting weight of each agent participating in the conflict based on the risk level, emotion score and resource utilization rate contained in the current state characteristics. The weighted calculation subunit is used to obtain the selection signals submitted by each participating agent based on its own policy preferences, and to calculate the weighted score of each candidate policy according to the dynamic voting weight. The weighted sorting subunit is used to sort the weighted scores of all candidate strategies and select the strategy with the highest score as the strategy coordination result.

[0062] Furthermore, the policy update module 406 specifically includes: The benchmark payout unit is used to calculate the expected payout value of each agent based on the current stable policy combination and record it as the benchmark payout vector. The potential reward unit is used to traverse the set of feasible policies for each agent and calculate the potential reward of the target agent after deviating alone, while keeping the policies of other agents unchanged. The target agent is any agent in the system. The profit comparison unit is used to compare the potential profit of each agent with the benchmark profit. If there is a strategy that can improve the profit, it is marked as an improved strategy and the corresponding profit gain is recorded. The strategy selection unit is used to select a strategy combination that significantly improves returns without compromising global stability for replacement and updating, based on the gains of the improveable strategies and the global constraints of the system. The convergence monitoring unit is used to recalculate the change in the payoff of all agents after each policy update, and introduces a convergence monitoring mechanism to track the dynamic trend of the payoff of each agent. The damping adjustment unit is used to activate the damping adjustment mechanism if the change in the payoff of any agent exceeds a preset threshold in multiple consecutive iterations, so as to adaptively reduce the policy update step size. The strategy iteration unit is used to determine that the system has reached convergence when the change in the payoff of all agents is lower than a preset threshold and the strategy combination remains stable after multiple rounds of iteration, and outputs the final Nash equilibrium strategy combination.

[0063] Furthermore, the strategy selection unit specifically includes: The agent selection subunit is used to select agents marked as having improvable policies, and calculate the incremental benefit of each candidate policy under the current global policy combination and its impact on the overall objective function of the system. The joint optimization subunit is used to establish a joint optimization model of revenue gain and system stability constraints based on the global system constraints. The system stability constraints include resource constraints, risk tolerance, and coordination consistency indicators. The stability calculation subunit is used to solve the joint optimization model, select a set of candidate strategies that significantly improve returns and meet global constraints, and calculate the global stability coefficient for each candidate strategy. The trade-off subunit is used to determine the optimal combination of replacement strategies based on the trade-off between the global stability coefficient and the benefit gain, using a multi-objective sorting algorithm. The replacement and update subunit is used to compare the optimal replacement strategy combination with the existing stable strategy combination. If the overall return increases by more than a threshold and the system stability does not decrease, then the strategy replacement and update is performed.

[0064] In the above embodiments, this application discloses a multi-agent coordination device based on dynamic Nash equilibrium, belonging to the field of artificial intelligence technology, and applied to task collaboration scenarios in the financial or healthcare fields. The multi-agent coordination mechanism based on dynamic Nash equilibrium proposed in this application abandons reinforcement learning and gradient optimization methods, effectively avoiding problems such as training instability, convergence difficulties, and high computational overhead. By discretizing the policy space of each agent into a finite set, a payoff matrix based on historical data and domain knowledge is constructed, and the support enumeration method is used to efficiently solve the Nash equilibrium, ensuring that the selected policy combination possesses adversarial stability and Pareto optimality. To further improve system adaptability, a rule engine and policy library are designed, dynamically mapping policies based on user emotions, task progress, and other state features. A weighted voting mechanism and priority rules are introduced to achieve conflict coordination among multiple agents. Combined with a Nash equilibrium verification mechanism, the stability of the policy combination is further ensured, guaranteeing the robustness and consistency of the system in dynamic environments. This application has significant advantages in scenarios such as complex task scheduling, intelligent customer service collaboration, and multi-robot collaboration, and possesses good scalability.

[0065] To address the aforementioned technical problems, embodiments of this application also provide a computer device. Please refer to [link / reference needed]. Figure 6 , Figure 6 This is a basic structural block diagram of the computer device in this embodiment.

[0066] The computer device 6 includes a memory 61, a processor 62, and a network interface 63 that are interconnected via a system bus. It should be noted that only the computer device 6 with memory 61, processor 62, and network interface 63 is shown in the figure; however, it should be understood that it is not required to implement all the components shown, and more or fewer components can be implemented alternatively. Those skilled in the art will understand that the computer device described here is a device capable of automatically performing numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes, but is not limited to, microprocessors, application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), digital signal processors (DSPs), embedded devices, etc.

[0067] The computer device can be a desktop computer, laptop, handheld computer, or cloud server, etc. The computer device can interact with the user via a keyboard, mouse, remote control, touchpad, or voice control.

[0068] The memory 61 includes at least one type of readable storage medium, including flash memory, hard disk, multimedia card, card-type memory (e.g., SD or DX memory), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 61 may be an internal storage unit of the computer device 6, such as the hard disk or memory of the computer device 6. In other embodiments, the memory 61 may also be an external storage device of the computer device 6, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the computer device 6. Of course, the memory 61 may also include both the internal storage unit and its external storage device of the computer device 6. In this embodiment, the memory 61 is typically used to store the operating system and various application software installed on the computer device 6, such as computer-readable instructions based on the dynamic Nash equilibrium multi-agent coordination method. In addition, the memory 61 can also be used to temporarily store various types of data that have been output or will be output.

[0069] In some embodiments, the processor 62 may be a central processing unit (CPU), controller, microcontroller, microprocessor, or other data processing chip. The processor 62 is typically used to control the overall operation of the computer device 6. In this embodiment, the processor 62 is used to execute computer-readable instructions stored in the memory 61 or to process data, for example, to execute computer-readable instructions of the multi-agent coordination method based on dynamic Nash equilibrium.

[0070] The network interface 63 may include a wireless network interface or a wired network interface, which is typically used to establish communication connections between the computer device 6 and other electronic devices.

[0071] This application also provides an implementation method, namely, a computer device including a memory and a processor, wherein the memory stores computer-readable instructions, and the processor executes the computer-readable instructions to implement the steps of the multi-agent coordination method based on dynamic Nash equilibrium as described above.

[0072] This application also provides another embodiment, namely, providing a computer-readable storage medium storing computer-readable instructions that can be executed by at least one processor to cause the at least one processor to perform the steps of the multi-agent coordination method based on dynamic Nash equilibrium as described above.

[0073] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0074] This application can be used in a wide variety of general-purpose or special-purpose computer system environments or configurations. Examples include: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics devices, network PCs, minicomputers, mainframe computers, and distributed computing environments including any of the above systems or devices. This application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform specific tasks or implement specific abstract data types. This application can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0075] It should be noted that the software tools or components not belonging to this company that appear in the various embodiments of this application are merely illustrative examples and do not represent actual use.

[0076] Obviously, the embodiments described above are only some embodiments of this application, not all embodiments. The accompanying drawings show preferred embodiments of this application, but do not limit the patent scope of this application. This application can be implemented in many different forms; rather, the purpose of providing these embodiments is to provide a more thorough and comprehensive understanding of the disclosure of this application. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing specific embodiments, or make equivalent substitutions for some of the technical features. Any equivalent structures made using the content of this application's specification and drawings, directly or indirectly applied to other related technical fields, are similarly within the scope of patent protection of this application.

Claims

1. A multi-agent coordination method based on dynamic Nash equilibrium, characterized in that, include: Obtain the current state characteristics of a multi-agent system, and define the role, objective, and discretized policy space of each agent, wherein the discretized policy space is a finite set of policies; A domain-knowledge-based payoff matrix is ​​constructed for each agent, wherein the payoff matrix records the payoff values ​​under different policy combinations; The Nash equilibrium value is determined based on the discretized strategy space and the payoff matrix. Based on the Nash equilibrium value, the policy support probability distribution of each agent is calculated by enumeration, and a system of equations is established to ensure that the expected returns are equal across supporting policies, thereby determining a stable policy combination. A dynamic coordination mechanism is designed based on the current state characteristics. The dynamic coordination mechanism and the stable strategy combination are mapped through a rule engine. Conflicts are resolved using weighted voting and priority rules. The weight of the weighted voting is adjusted and normalized in real time according to the current state characteristics. Nash equilibrium verification is performed on the stable policy combination. Based on the Nash equilibrium verification result, each agent is traversed to determine whether there is a better policy. If there is, a new policy combination is selected, and a convergence mechanism is introduced to monitor the policy change amplitude to avoid oscillation, until the change amplitude of all agents is lower than a preset threshold.

2. The multi-agent coordination method based on dynamic Nash equilibrium as described in claim 1, characterized in that, The steps of obtaining the current state characteristics of the multi-agent system and defining the role, objective, and discretized policy space of each agent specifically include: The current state features are extracted from real-time data using a pre-trained model, wherein the current state features include at least an emotion score and a risk level; For each agent, define the role and the objective, wherein the objective is an independently optimized local objective; The continuous policy of each agent is converted into the finite policy set, wherein the finite policy set includes at least aggressive policy, balanced policy and deferred policy.

3. The multi-agent coordination method based on dynamic Nash equilibrium as described in claim 1, characterized in that, The step of calculating the policy support probability distribution for each agent using an enumeration method based on the Nash equilibrium value, and establishing a system of equations to ensure that expected returns are equal across supporting policies, thereby determining a stable policy combination, specifically includes: For each agent, a finite set of policies is enumerated to construct a set of all possible policy combinations, and the corresponding joint reward vector is recorded. Based on the payout matrix, the individual payout of each agent under each strategy combination is calculated, and the relative payout of each agent under different strategies is obtained through normalization. Based on the enumeration results, candidate policy combinations that satisfy the Nash equilibrium condition are extracted, and a support policy set is calculated for each candidate policy combination. For each of the agent's support policy sets, a set of expected return equations is established, wherein the set of expected return equations is used to constrain the expected returns of all support policies to be equal under equilibrium conditions. The probability distribution values ​​corresponding to each support strategy are obtained by solving a system of linear equations or by using the Lagrange multiplier method, thus obtaining the policy support probability distribution of each agent. The stability of the probability distribution supporting the strategy is tested. If there is a strategy that does not meet the payoff balance, it is removed and the probability distribution is renormalized. Determine the final stable strategy combination while satisfying the balance of expected returns for all agents.

4. The multi-agent coordination method based on dynamic Nash equilibrium as described in claim 1, characterized in that, The steps of designing a dynamic coordination mechanism based on the current state characteristics, mapping the dynamic coordination mechanism and the stable strategy combination through a rule engine, and resolving conflicts using weighted voting and priority rules specifically include: Based on the current state characteristics, environmental variables, agent behavior characteristics, and global system constraints are extracted to construct the input parameter set of the dynamic coordination mechanism; The rule engine is used to establish a mapping relationship between the current state features and the stable policy combination, wherein each mapping rule includes a triggering condition, a priority label and a corresponding coordination action; When multiple agents propose conflicting policy choices, a weighted voting mechanism is triggered to aggregate the policy preferences of each agent. The voting weights are dynamically calculated based on the importance of each agent's role in the system, its historical performance, and its risk level, and the weights are then normalized. The weighted voting results are determined by combining priority rules. When there are ties or similar benefits, the decision is made based on the priority label. The consistency of the coordination result determined by the voting decision is verified. If there is a conflict with the global constraints, the rule engine is invoked to rematch the backup coordination strategy. The final determined coordination output is used as the system execution instruction, and the current state characteristics and the final coordination result are stored together to update the rule engine knowledge base.

5. The multi-agent coordination method based on dynamic Nash equilibrium as described in claim 4, characterized in that, The step of triggering a weighted voting mechanism to aggregate the policy preferences of each agent when multiple agents propose conflicting policy choices specifically includes: The current policy output of each agent under the stable policy combination is detected. If there is a conflict between the benefit objective or the action to be executed, it is determined that a conflict event has occurred and a weighted voting mechanism is initiated. Identify the agents involved in the conflict, extract their role attributes, priority parameters, and historical voting records, and construct a set of voting participants; Based on the risk level, emotion score, and resource utilization rate included in the current state characteristics, calculate the dynamic voting weight of each agent participating in the conflict; Obtain the selection signals submitted by each participating agent based on its own policy preferences, and calculate the weighted score of each candidate policy according to the dynamic voting weight. The weighted scores of all candidate strategies are sorted, and the strategy with the highest score is selected as the strategy coordination result.

6. The multi-agent coordination method based on dynamic Nash equilibrium as described in claim 1, characterized in that, The step of performing Nash equilibrium verification on the stable policy combination, traversing each agent based on the Nash equilibrium verification result to determine if a better policy exists, and if so, reselecting a policy combination, and introducing a convergence mechanism to monitor the policy change amplitude to avoid oscillations, until the change amplitude of all agents is lower than a preset threshold, specifically includes: Based on the current stable strategy combination, calculate the expected reward value of each agent and record it as the benchmark reward vector; For each agent, the set of feasible strategies is traversed. Under the condition that the strategies of other agents remain unchanged, the potential gain of the target agent after deviating alone is calculated, wherein the target agent is any agent in the system. The potential gains of each agent are compared with the baseline gains. If there is a strategy that can improve the gains, it is marked as an improvable strategy, and the corresponding gain is recorded. Based on the profit gains of the improved strategies and the global system constraints, select a strategy combination that significantly improves profits without compromising global stability for replacement and update. After each policy update, the change in the payoff of all agents is recalculated, and a convergence monitoring mechanism is introduced to track the dynamic trend of the payoff of each agent. If the change in the reward of any agent exceeds a preset threshold in multiple consecutive iterations, a damping adjustment mechanism is activated to adaptively reduce the policy update step size. When the change in the returns of all the agents is lower than the preset threshold, and the strategy combination remains stable after multiple iterations, the system is determined to have reached convergence and the final Nash equilibrium strategy combination is output.

7. The multi-agent coordination method based on dynamic Nash equilibrium as described in claim 6, characterized in that, The step of selecting a strategy combination that significantly improves returns without compromising global stability for replacement and updating based on the gain of the improveable strategy and the global constraints of the system specifically includes: Agents marked as having improvable strategies are screened, and the incremental benefit of each candidate strategy under the current global strategy combination and its impact on the overall objective function of the system are calculated. Based on the global system constraints, a joint optimization model of the revenue gain and system stability constraints is established, wherein the system stability constraints include resource constraints, risk tolerance, and coordination consistency index. The joint optimization model is solved to select a set of candidate strategies that significantly improve returns and satisfy global constraints, and the global stability coefficient is calculated for each candidate strategy. Based on the trade-off between the global stability coefficient and the profit gain, a multi-objective ranking algorithm is used to determine the optimal combination of replacement strategies. The optimal replacement strategy combination is compared with the existing stable strategy combination. If the overall return increases by more than a threshold and the system stability does not decrease, then the strategy replacement update is performed.

8. A multi-agent coordination device based on dynamic Nash equilibrium, characterized in that, include: The agent definition module is used to obtain the current state characteristics of the multi-agent system and define the role, objective, and discretized policy space of each agent, wherein the discretized policy space is a finite set of policies; The revenue matrix module is used to construct a domain knowledge-based revenue matrix for each agent, wherein the revenue matrix records the revenue values ​​under different strategy combinations; The Nash equilibrium module is used to solve for the Nash equilibrium value based on the discretized strategy space and the payoff matrix. The strategy combination module is used to calculate the policy support probability distribution of each agent based on the Nash equilibrium value by enumeration, and to establish a system of equations to ensure that the expected returns are equal on the supporting policies, thereby determining a stable strategy combination. The coordination mapping module is used to design a dynamic coordination mechanism based on the current state characteristics, map the combination of the dynamic coordination mechanism and the stable strategy through a rule engine, and resolve conflicts using weighted voting and priority rules. The weight of the weighted voting is adjusted and normalized in real time according to the current state characteristics. The policy update module is used to perform Nash equilibrium verification on the stable policy combination. Based on the Nash equilibrium verification result, it iterates through each agent to determine whether there is a better policy. If there is, a new policy combination is selected, and a convergence mechanism is introduced to monitor the policy change amplitude to avoid oscillation, until the change amplitude of all agents is lower than a preset threshold.

9. A computer device, characterized in that, The system includes a memory and a processor, wherein the memory stores computer-readable instructions, and the processor executes the computer-readable instructions to implement the steps of the multi-agent coordination method based on dynamic Nash equilibrium as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-readable instructions, which, when executed by a processor, implement the steps of the multi-agent coordination method based on dynamic Nash equilibrium as described in any one of claims 1 to 7.

Images