Agent-driven control method and apparatus

Abstract

This invention relates to the field of intelligent agent technology and discloses an intelligent agent-driven control method and apparatus. The invention determines the intelligent agent-driven model corresponding to target requirement information and obtains the target solution problem of the intelligent agent-driven model. Attribute feature information, simulation environment information, and the target solution problem are input into the intelligent agent-driven model for learning, resulting in target learning parameters. Based on the target learning parameters, a preset solution algorithm is used to iteratively calculate the target solution result corresponding to the target solution problem multiple times until the target solution result meets the preset requirements. Finally, the optimal control parameters for the target intelligent agent to execute the target action are output. Therefore, this invention can quickly generate the optimal control parameters for the intelligent agent to execute the target action without relying on manual intervention, thereby improving the flexibility and efficiency of intelligent agent control and ensuring the accuracy of intelligent agent participation in intelligent control.

Description

Technical Field

[0001] This invention relates to the field of intelligent agent technology, and specifically to a control method and apparatus for intelligent agent driving. Background Technology

[0002] The decision-making behavior of an intelligent agent refers to the optimal or near-optimal action choice made by an intelligent agent in a system consisting of environment, rules, and goals, based on the current state information it perceives, in order to achieve its desired goal.

[0003] In related technologies, the decision-making rules of intelligent agents are generally encoded manually to control the agents. However, this method is cumbersome and prone to errors, making it difficult to meet the requirements of multi-agent and large-scale complex systems. Furthermore, manual encoding is inefficient and difficult to adapt to the optimization of multi-agent decision-making processes, thus limiting the flexibility of generating control parameters for intelligent agents. Summary of the Invention

[0004] This invention provides a control method and apparatus driven by an intelligent agent to solve the problems of low efficiency and poor flexibility caused by relying on manual encoding of decision-making behavior rules for intelligent agents.

[0005] According to a first aspect, the present invention provides a control method for intelligent agent-driven systems, the method comprising: Acquire the attribute characteristics of the target intelligent agent, the simulation environment information, and the target user's target requirement information; Determine the agent-driven model corresponding to the target requirement information, and obtain the target problem of the agent-driven model; The attribute feature information, simulation environment information, and target problem are input into the agent-driven model learning to obtain the target learning parameters. Based on the target learning parameters, the target solution result corresponding to the target problem is calculated iteratively multiple times using a preset solution algorithm until the target solution result meets the preset requirements. Then, the optimal control parameters for the target agent to perform the target action are output.

[0006] In some specific implementations, the attribute feature information includes: parameters, events, variables, and functions of the target agent; the simulation environment information includes: link information, state information, and state transition conditions of the target agent; and the target requirement information includes: number of groups, number of iterations, simulation dimension, random seed, crossover rate, and convergence parameters.

[0007] In some specific implementations, when the agent-driven model is an agent optimization model, the target learning parameters include: objective function, decision variables, constraints, and parameter boundaries; when the agent-driven model is an agent game model, the target learning parameters include: state space, action space, environmental parameters, and reward function.

[0008] In some specific implementations, when the agent-driven model is an agent optimization model, the target problem is iteratively calculated multiple times using a preset solution algorithm based on the target learning parameters until the target solution meets the preset requirements. The optimal control parameters for the target agent to execute the target action are then output, including: Based on the objective function, decision variables, constraints, parameter boundaries, and objective requirements, the system iteratively calculates the objective solution result for the objective problem using a preset solution algorithm until the objective solution result meets the preset requirements. Then, it outputs the optimal decision variables for the target agent to perform the target action.

[0009] In some specific implementations, when the agent-driven model is an agent game model, the target problem is iteratively calculated multiple times using a preset solution algorithm based on the target learning parameters until the target solution meets the preset requirements. The optimal control parameters for the target agent to execute the target action are then output, including: Based on the state space, action space, environmental parameters, reward function, and target requirement information, the target solution result corresponding to the target problem is calculated iteratively multiple times using a preset solution algorithm until the target solution result meets the preset requirements. Then, the optimal update variable for the target agent to perform the target action is output.

[0010] In some specific embodiments, the agent-driven control method of the present invention further includes: The optimal control parameters for the target agent to perform the target action are input into the behavior policy generation unit to obtain the behavior control policy for the target agent to perform the target action in the current state.

[0011] In some specific embodiments, the agent-driven control method of the present invention further includes: The behavior control strategy is fed back to the simulation platform to execute actions, and the attribute characteristics information of the target intelligent agent and the simulation environment information are updated.

[0012] Secondly, the present invention also provides a control device for intelligent agent driving, the device comprising: The acquisition module is used to acquire the attribute feature information of the target intelligent agent, the simulation environment information, and the target demand information sent by the target user; The determination module is used to determine the agent-driven model corresponding to the target requirement information and obtain the target solution problem of the agent-driven model; The learning module is used to input attribute feature information, simulation environment information, and target problem into the agent-driven model learning to obtain target learning parameters; The calculation module is used to iteratively calculate the target solution result corresponding to the target problem using a preset solution algorithm based on the target learning parameters, until the target solution result meets the preset requirements, and output the optimal control parameters for the target agent to perform the target action.

[0013] Thirdly, the present invention also provides an electronic device, comprising: The memory and the processor are interconnected and communicate with each other. The memory stores computer instructions, and the processor executes the computer instructions to perform the agent-driven control method of the first aspect or any embodiment of the first aspect.

[0014] Fourthly, the present invention also provides a computer-readable storage medium storing computer instructions for causing a computer to execute the agent-driven control method of the first aspect or any embodiment of the first aspect.

[0015] The technical solution of this invention has the following advantages: This invention discloses a control method and apparatus driven by an intelligent agent. The invention determines the intelligent agent driving model corresponding to the target requirement information and obtains the target solution problem of the intelligent agent driving model. Attribute feature information, simulation environment information, and the target solution problem are input into the intelligent agent driving model for learning, resulting in target learning parameters. Based on the target learning parameters, a preset solution algorithm is used to iteratively calculate the target solution result corresponding to the target solution problem multiple times until the target solution result meets the preset requirements. Finally, the optimal control parameters for the target intelligent agent to execute the target action are output. Therefore, this invention can quickly generate the optimal control parameters for the intelligent agent to execute the target action without relying on manual intervention, thereby improving the flexibility and efficiency of intelligent agent control and ensuring the accuracy of intelligent agent participation in intelligent control. Attached Figure Description

[0016] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0017] Figure 1 This is a flowchart illustrating a control method driven by an intelligent agent according to an embodiment of the present invention; Figure 2 This is a schematic flowchart of a control method driven by an intelligent agent according to an embodiment of the present invention; Figure 3 This is a flowchart illustrating another intelligent agent-driven control method according to an embodiment of the present invention; Figure 4 This is a flowchart illustrating another intelligent agent-driven control method according to an embodiment of the present invention; Figure 5 This is a structural block diagram of a control device driven by an intelligent agent according to an embodiment of the present invention; Figure 6 This is a schematic diagram of the hardware structure of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] It is understood that before using the technical solutions disclosed in the various embodiments of the present invention, users should be informed of the types, scope of use, and usage scenarios of the personal information involved in the present invention and their authorization should be obtained in accordance with relevant laws and regulations through appropriate means.

[0020] The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0021] The decision-making behavior of an intelligent agent refers to the optimal or near-optimal action choice made by an intelligent agent in a system consisting of environment, rules, and goals, based on the current state information it perceives, in order to achieve its desired goal.

[0022] The behavior selection process of intelligent agents typically exhibits temporal continuity, state dependence, and goal orientation, manifesting as a feedback control process involving continuous perception, evaluation, prediction, and selection within complex systems. An agent's behavior is influenced not only by its internal state (such as goal preferences and resource constraints) but also by the external environment (such as the behavior of other agents and changes in environmental resources). Therefore, the behavior of intelligent agents is highly dynamic and complex.

[0023] In related technologies, it is generally necessary to manually encode the decision-making rules of intelligent agents in order to control them. However, this method is cumbersome and prone to errors, making it difficult to meet the requirements of multi-agent and large-scale complex systems. Furthermore, manual coding is inefficient and difficult to adapt to the optimization of multi-agent decision-making processes, thus limiting the flexibility of agent policy generation.

[0024] Therefore, this embodiment provides an agent-driven control method that can be used in computer devices such as mobile phones, tablets, desktop computers, laptops, and servers. Figure 1 This is a flowchart of a control method for intelligent agent-driven operation according to an embodiment of the present invention, such as... Figure 1 As shown, the process includes the following steps: Step S101: Obtain the attribute feature information of the target intelligent agent, the simulation environment information, and the target demand information sent by the target user.

[0025] Specifically, the target agent can be the i-th agent.

[0026] In some specific implementations, the attribute feature information includes: parameters, events, variables, and functions of the target agent; the simulation environment information includes: link information, state information, and state transition conditions of the target agent; and the target requirement information includes: number of groups, number of iterations, simulation dimension, random seed, crossover rate, and convergence parameters.

[0027] For example, for the i-th agent, we acquire its attribute features such as parameters, events, variables, and functions. Simultaneously, we also acquire its connection information, state information, and state transition conditions, among other simulation environment information. The attribute features and simulation environment information of the i-th agent are then labeled as follows: .in, This represents the attribute feature information of the i-th agent. This represents the simulation environment information for the i-th intelligent agent.

[0028] For example, obtain target requirement information such as the number of groups, number of iterations, simulation dimensions, random seed, crossover rate, and convergence parameters sent by the target user, and mark the target requirement information of the i-th agent as follows: .in, Indicates the number of groups. Indicates the number of iterations. Indicates the simulation dimension.

[0029] Step S102: Determine the agent-driven model corresponding to the target requirement information, and obtain the target problem of the agent-driven model.

[0030] Specifically, the agent-driven model corresponding to the target demand information includes an agent optimization model and an agent game model. The agent optimization model is used to determine the agent's optimal decision variables, while the agent game model is used to determine the agent's optimal update variables.

[0031] The objective problem is the optimization strategy within the target application scenario. For example, in the optimization scenario of a distribution center, the target user, given variables such as "delivery route," "number of vehicles," and "time window," takes "minimum total cost" or "shortest delivery time" as the objective problem. Based on this objective problem, the implementation methods described in this application's embodiments ultimately yield an approximate optimal solution.

[0032] Step S103: Input the attribute feature information, simulation environment information and target problem into the agent-driven model learning to obtain the target learning parameters.

[0033] In some specific implementations, when the agent-driven model is an agent optimization model, the target learning parameters include: objective function, decision variables, constraints, and parameter boundaries; when the agent-driven model is an agent game model, the target learning parameters include: state space, action space, environmental parameters, and reward function.

[0034] Step S104: Based on the target learning parameters, use a preset solution algorithm to iteratively calculate the target solution result corresponding to the target solution problem multiple times until the target solution result meets the preset requirements, and output the optimal control parameters for the target agent to perform the target action.

[0035] In some specific implementations, when the agent-driven model is an agent optimization model, the target problem is iteratively calculated multiple times using a preset solution algorithm based on the target learning parameters until the target solution meets the preset requirements. The optimal control parameters for the target agent to execute the target action are then output, including: Based on the objective function, decision variables, constraints, parameter boundaries, and objective requirements, the system iteratively calculates the objective solution result for the objective problem using a preset solution algorithm until the objective solution result meets the preset requirements. Then, it outputs the optimal decision variables for the target agent to perform the target action.

[0036] like Figure 2 The diagram shown is a schematic flowchart illustrating the control method for intelligent agent-driven implementation in this application. Figure 2 First, determine whether the simulation has ended. If not, input the target agent's attribute characteristics, simulation environment information, and target user's target requirement information into the agent optimization model. Figure 2In this process, after the optimal control parameters for the target agent to execute the target action are output through the agent optimization model, operations research optimization is performed to solve the problem. Specifically, the solution can be obtained through an operations research optimization solver.

[0037] Specifically, the objective function is used In other words, decision variables are represented by Indicates that constraints are expressed using Indicate that the parameter boundary is represented by This indicates that the target requirement information mentioned above is represented as... These parameters are represented as a set of features. .

[0038] For example, when the agent optimization model is determined as the solution method, the set of elements mentioned above is used. As input, the agent optimization model is invoked to solve the objective problem. The solver then processes the decision variables... Objective function Constraints Parameter Boundaries Together with the target requirement information, the solution is integrated into a description of the optimization problem. Using a pre-defined solution algorithm, it undergoes N rounds of iterative calculation. The solver outputs a Pareto optimal solution set and returns the corresponding optimal combination of decision variables. .

[0039] For example, the first i Individual agent attribute characteristics and simulation environment information Input into the agent optimization model, and parse the first... i Information such as the task objective, performance indicators, and behavioral reward function of each intelligent agent is used to generate an objective function corresponding to the problem being solved. During the objective function generation process, keywords related to variables or functions are read... i Performance evaluation metrics related to the agent are selected, such as total metrics, task completion time (hrs), unit cost, energy consumption metrics, or system stability metrics, and are filtered through secondary confirmation by the user. Subsequently, combined with the optimization objectives set by the user (such as minimizing energy consumption, maximizing task completion, minimizing cost and risk, etc.), the objective function is generated.

[0040] When the target requirement information characterizes a scenario with multi-objective optimization needs, multiple objective functions are generated and combined into an objective function vector to support the Pareto optimization strategy. The objective functions in this application can be manually set by the target user. The generated objective function expression is denoted as... This will serve as an important basis for the population search direction in the subsequent preset solution algorithm.

[0041] For example, the first i Individual agent attribute characteristics and simulation environment information The input is fed into the agent optimization model, where variables that have a key impact on the current optimization task are identified and extracted from the complex agent attribute structure and used as decision variables input for the preset solution algorithm. Its core idea is based on the objective function described above. Using a large language model, key variables that significantly influence the target solution are identified. These variables are then output in JSON format for secondary confirmation and correction by the target user. Decision variables can also be manually set by the target user. Once confirmed, their expression is denoted as... The extracted decision variables will serve as the core search dimensions in the subsequent preset solution algorithm.

[0042] For example, the first i Individual agent attribute characteristics and simulation environment information The input is fed into the agent optimization model, which automatically identifies or assists in generating mathematical constraints that limit the target agent's behavior, ensuring the feasibility and legitimacy of the solution in actual execution. Constraints can originate from various aspects, including the target agent's internal structure, collaborative or conflictual relationships with other agents, and limitations on environmental resource availability. The model analyzes boundary conditions in agent attributes (such as maximum resource quotas, minimum action costs, and action cooldown times) and logical or arithmetic relationships between state variables (such as dependencies, non-repeatability, and exclusivity). Target users can also manually define specific constraint expressions or upload constraint logic scripts via an interface. Finally, the identified constraints are uniformly converted into a set of constraint expressions recognizable by the optimization solver, denoted as... It is used in subsequent solution stages to filter infeasible solutions and guide the search space.

[0043] For example, the first i Individual agent attribute characteristics and simulation environment information The input is fed into an agent optimization model, which is used to optimize the set of decision variables. Each variable in the model has a defined range of values, i.e., upper and lower bounds, to form a boundary description of the optimization search space. During the setting process, the range of values ​​defined for each variable in the agent optimization model is searched. If an explicit boundary definition exists (e.g., "velocity between 0 and 10", "allocation ratio between 0.0 and 1.0"), it is directly used as the upper and lower bounds for that variable. Users or system developers can also manually adjust the upper and lower bounds of variables. Finally, for each variable... (where j is the variable index) Set a lower bound With the upper realm This is summarized to form a parameter boundary set. This serves as a boundary reference for the search space in subsequent preset optimization algorithms.

[0044] In other specific implementations, when the agent-driven model is an agent game model, the target problem is iteratively calculated multiple times using a preset solution algorithm based on the target learning parameters until the target solution meets the preset requirements. The optimal control parameters for the target agent to execute the target action are then output, including: Based on the state space, action space, environmental parameters, reward function, and target requirement information, the target solution result corresponding to the target problem is calculated iteratively multiple times using a preset solution algorithm until the target solution result meets the preset requirements. Then, the optimal update variable for the target agent to perform the target action is output.

[0045] like Figure 2 The diagram shown is a schematic flowchart illustrating the control method for intelligent agent-driven implementation in this application. Figure 2 First, determine whether the simulation has ended. If not, input the target agent's attribute characteristics, simulation environment information, and target user's target demand information into the agent game model. Figure 2 In this process, after outputting the optimal control parameters for the target agent to execute the target action through the agent game model, game analysis is then performed to solve the problem. Specifically, the solution can be obtained through a game analyzer.

[0046] Specifically, the state space is used Indicates that the action space is used Indicates that environmental parameters are used This indicates that the reward function uses This indicates that the target requirement information mentioned above is represented as... These parameters are represented as a set of features. .

[0047] For example, when the agent game model is determined as the solution method, the set of elements mentioned above is used. As input, the intelligent agent game model is invoked to solve the objective problem. The solver solves the problem based on the state space. Action space Environment functions and reward function Iterative updates are performed using pre-defined algorithms such as reinforcement learning solvers, Nash equilibrium approximators, and evolutionary game simulators. In each iteration, the parameters of the agent's game model are adjusted based on the reward function feedback until the target solution meets the preset requirements. The optimal updated variable for the target agent to execute the target action is then output. .

[0048] For example, the first iIndividual agent attribute characteristics and simulation environment information The input is fed into an intelligent agent game model, which traverses the agent's attribute features to perceive or record state-related variables (such as position, speed, resource holdings, task execution stage, health status, current time window, etc.) and, combined with the defined rules in the state transition conditions, establishes a combination space of variables. Specifically, based on the first... i The state space is generated by the discrete variables of each agent, that is, by enumerating the discrete values ​​of the variables to generate a set of states. The final generated state space... Used as state input for subsequent action evaluation and strategic game.

[0049] For example, the attribute features of the i-th agent and the simulation environment information The input is fed into the intelligent agent game model, which then generates the action space. The action space is used to define the action... i A smart agent , which is the set of all possible action choices that can be executed in any state. Specifically, we analyze the executable operations defined in the game model of the i-th agent, including function calls of the agent (such as movement, cooperation, and trading), event triggers (such as responses triggered after reaching a certain location), and any operations that can cause changes to the agent's internal variables, and combine this with state transition logic to determine the availability range of actions.

[0050] The action space can be discrete (such as "forward" or "backward" for discrete tasks) or continuous (such as "speed adjustment" or "resource allocation rate" for continuous tasks). Appropriate action sets are automatically generated based on the action type. The model also labels the validity mapping between each action and state. Furthermore, this agent-based game theory model can receive user-input policy bases, game rule constraints, or action types supported by external decision-making models, dynamically tailoring or expanding the action set.

[0051] For example, the first i Individual agent attribute characteristics and simulation environment information The input is fed into the agent game model, which then generates an environment function. This environment function provides external input for state perception, behavioral reasoning, and game modeling within the agent system. In the agent game model, environmental variables (such as spatial patterns, resource point distribution, task nodes, and communication delays) in the simulation model are scanned and their relationship with the first... i A smart agent The interaction relationships between them. Secondly, it is also necessary to obtain the link relationships between the agents, specifically analyzing the... i One agent and other agents Extracting the dependencies, conflicts, or collaborations between them to the third party. i A smart agent The interactions that create an impact. Environment functions. It can be modeled as a function That is, the current number i An agent's environment changes are influenced not only by its own state and actions, but also by the actions of other agents linked to it. This environment function is used to deduce the state transition process, construct the joint policy space, and serve as the environment interface in subsequent game-solving or learning algorithms.

[0052] For example, the first i Individual agent attribute characteristics and simulation environment information The input is fed into the intelligent agent game model and combined with the objective function described above to generate the reward function. The optimization objective (such as cost minimization or efficiency maximization) in the objective function is mapped to an immediate reward. Through linear transformation, normalization, or symbol conversion, the index of the optimization objective is transformed into a positively oriented incentive. Simultaneously, behavioral consequences and environmental feedback (such as the success or failure of state transitions caused by actions, the results of interactions with others, etc.) are introduced to generate the immediate reward in the reward function. That is, the first i An agent at time step The reward value. In adversarial or cooperative game scenarios, agent game models support joint evaluation of the outcomes of multiple parties' actions, setting differentiated reward mechanisms (such as local rewards, global rewards, shared rewards, or differential rewards). The final reward function is... It can be directly called by reinforcement learning algorithms or game solving algorithms for processes such as policy gradient calculation and value function backtracking.

[0053] This application, by executing steps S101-S104 above, obtains the attribute feature information of the target intelligent agent, simulation environment information, and target demand information sent by the target user; determines the intelligent agent driving model corresponding to the target demand information, and obtains the target solution problem of the intelligent agent driving model; inputs the attribute feature information, simulation environment information, and target solution problem into the intelligent agent driving model for learning, and obtains target learning parameters; based on the target learning parameters, iteratively calculates the target solution result corresponding to the target solution problem using a preset solution algorithm until the target solution result meets the preset requirements, and outputs the optimal control parameters for the target intelligent agent to execute the target action. Therefore, this application can quickly generate the optimal control parameters for the intelligent agent to execute the target action without relying on human intervention, thereby improving the flexibility and efficiency of intelligent agent control, and also ensuring the accuracy of the intelligent agent's participation in intelligent control.

[0054] This embodiment provides an agent-driven control method that can be used in computer devices such as mobile phones, tablets, desktop computers, laptops, and servers. Figure 2This is a flowchart of a control method for intelligent agent-driven operation according to an embodiment of the present invention, such as... Figure 2 As shown, the process includes the following steps: Step S101: Obtain the attribute feature information of the target intelligent agent, the simulation environment information, and the target demand information sent by the target user.

[0055] Step S102: Determine the agent-driven model corresponding to the target requirement information, and obtain the target problem of the agent-driven model.

[0056] Step S103: Input the attribute feature information, simulation environment information and target problem into the agent-driven model learning to obtain the target learning parameters.

[0057] Step S104: Based on the target learning parameters, use a preset solution algorithm to iteratively calculate the target solution result corresponding to the target solution problem multiple times until the target solution result meets the preset requirements, and output the optimal control parameters for the target agent to perform the target action.

[0058] For details of steps S101-S104 above, please refer to the aforementioned embodiments, and they will not be repeated here.

[0059] Step S105: Input the optimal control parameters for the target agent to perform the target action into the behavior policy generation unit to obtain the behavior control policy for the target agent to perform the target action in the current state.

[0060] exist Figure 2 In this process, after operations optimization through an agent optimization model or analysis through an agent game model, a behavioral control strategy for the target action is generated, and finally, the control strategy is output through a policy mapping interface.

[0061] Specifically, the optimal decision variables generated above will be combined or optimal update variable The input is fed into the policy generation unit, which then generates the specific actions that the target agent should take in the current state.

[0062] For example, if the optimal control parameters for the target agent to perform the target action are a vector of length n, then each parameter in the vector is mapped to the corresponding nth parameter. i The behavioral decisions of each agent constitute the set of joint behavioral strategies of the agent group at the current time step.

[0063] For example, combining the optimal decision variables generated above or optimal update variable The input is fed into the policy generation unit, which, based on the numerical optimal result, generates a structured set of behavioral policies, making them executable and capable of driving the simulated target agent to produce actual behaviors. During the behavioral policy generation process, semantic mapping and structural reconstruction are performed on the elements of the optimal solution. For the numerical solution from the optimization solver... Based on the semantic definition of the original decision variables, they are mapped to parameter configurations for agent behavior; for the policy solutions from the agent game model... The "state-policy" mapping relationships in the policy sequence are organized into decision rule tables or policy functions. Finally, a structured set of behavioral policies is generated.

[0064] During processing, the behavioral policy set aims to meet the following two conditions: it must be compatible with the current agent-driven model structure, including action space matching and behavior trigger condition mapping; and it must be uniformly encapsulated into a behavioral policy interface format for easy input into the simulation platform, achieving consistency and traceability in multi-agent policy execution. Specifically, the system parses each configuration item in the policy set according to a predefined parameter mapping script and binds it to the corresponding agent's behavior control interface or configuration entry point in the simulation environment. This dynamically updates the agent's behavioral logic, resulting in a structured policy set that supports simulation environment parsing.

[0065] This application obtains the behavior control strategy for the target agent to perform the target action by inputting the optimal control parameters of the target agent to the behavior policy generation unit, thereby facilitating precise control of the agent.

[0066] This embodiment provides an agent-driven control method that can be used in computer devices such as mobile phones, tablets, desktop computers, laptops, and servers. Figure 4 This is a flowchart of a control method for intelligent agent-driven operation according to an embodiment of the present invention, such as... Figure 3 As shown, the process includes the following steps: Step S101: Obtain the attribute feature information of the target intelligent agent, the simulation environment information, and the target demand information sent by the target user.

[0067] Step S102: Determine the agent-driven model corresponding to the target requirement information, and obtain the target problem of the agent-driven model.

[0068] Step S103: Input the attribute feature information, simulation environment information and target problem into the agent-driven model learning to obtain the target learning parameters.

[0069] Step S104: Based on the target learning parameters, use a preset solution algorithm to iteratively calculate the target solution result corresponding to the target solution problem multiple times until the target solution result meets the preset requirements, and output the optimal control parameters for the target agent to perform the target action.

[0070] Step S105: Input the optimal control parameters for the target agent to perform the target action into the behavior policy generation unit to obtain the behavior control policy for the target agent to perform the target action in the current state.

[0071] For details of steps S101-S105 above, please refer to the aforementioned embodiments, and they will not be repeated here.

[0072] Step S106: Feedback the behavior control strategy to the simulation platform to execute actions, and update the attribute feature information and simulation environment information of the target intelligent agent.

[0073] Specifically, the behavior control strategy for the target agent to perform the target action in the current state is used as the agent behavior input at the current time step and fed back to the simulation platform for execution.

[0074] Based on the agent's behavior and environmental feedback rules, update the environmental state information and agent attribute information to generate the environmental state for the next time step. and the attribute state of the target agent in the next time step If the simulation has not yet reached the preset termination condition, it will enter the next simulation cycle and continue to execute. Based on the target learning parameters, the preset solution algorithm will be used to iteratively calculate the target solution result corresponding to the target solution problem multiple times until the target solution result reaches the preset requirements. Then the simulation process ends and all agent behavior records and optimization process data are output.

[0075] For example, the behavior control strategies described above are bound to specific action entities in the simulation platform via a simulation interface to ensure that the behavioral intentions output by the strategy layer can be accurately executed at the physical or logical level. Specifically, strategy parameters are mapped to predefined action execution functions (such as movement, interaction, allocation, communication, etc.) in the simulation software, completing the conversion and binding from "strategy to action". During the binding process, multiple factors such as strategy timestamps, execution conditions, and resource status are considered simultaneously to ensure the rationality and real-time performance of the action triggering logic. After the strategy action is executed, the environmental information at the current simulation moment is automatically updated. and the attribute state of the target intelligent agent If the simulation has not yet ended (e.g., the time limit has not been reached or the critical event has not been triggered), the new state is used as input to continue solving and enter the simulation and policy generation loop at the next time step; if the simulation ends, the agent parameters are no longer collected and the operation stops.

[0076] This application tightly integrates agent optimization models and agent game models, achieving automatic transformation of agent problems into optimization and game models through specific problem decomposition and solving. Compared with traditional methods relying on manual rules and coding, this significantly reduces the time cost of manual debugging and improves the flexibility of agent optimization.

[0077] In addition, by feeding back the behavior control strategy to the simulation platform to execute actions and updating the attribute feature information of the target agent and the simulation environment information, the problem of the optimization solution process being strongly bound to the simulation platform or the agent model having poor transferability is avoided, which greatly improves the independence and flexibility of the algorithm components.

[0078] This embodiment also provides a control device driven by an intelligent agent, which is used to implement the above embodiments and preferred embodiments; details already described will not be repeated. As used below, the term "module" can refer to a combination of software and / or hardware that implements a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0079] This embodiment provides a control device driven by an intelligent agent, such as... Figure 5 As shown, it includes: The acquisition module 501 is used to acquire the attribute feature information of the target intelligent agent, the simulation environment information, and the target demand information sent by the target user.

[0080] The determination module 502 is used to determine the agent-driven model corresponding to the target requirement information and obtain the target solution problem of the agent-driven model.

[0081] The learning module 503 is used to input the attribute feature information, the simulation environment information, and the target problem into the agent-driven model learning to obtain target learning parameters.

[0082] The calculation module 504 is used to calculate the target solution result corresponding to the target solution problem multiple times using a preset solution algorithm based on the target learning parameters, until the target solution result meets the preset requirements, and output the optimal control parameters for the target agent to perform the target action.

[0083] In some specific implementations, the attribute feature information includes: parameters, events, variables, and functions of the target agent; the simulation environment information includes: link information, state information, and state transition conditions of the target agent; and the target requirement information includes: number of groups, number of iterations, simulation dimension, random seed, crossover rate, and convergence parameters.

[0084] In some specific implementations, when the agent-driven model is an agent optimization model, the target learning parameters include: objective function, decision variables, constraints, and parameter boundaries; when the agent-driven model is an agent game model, the target learning parameters include: state space, action space, environmental parameters, and reward function.

[0085] In some specific implementations, when the agent-driven model is an agent optimization model, the target problem is iteratively calculated multiple times using a preset solution algorithm based on the target learning parameters until the target solution meets the preset requirements. The optimal control parameters for the target agent to execute the target action are then output, including: Based on the objective function, decision variables, constraints, parameter boundaries, and objective requirements, the system iteratively calculates the objective solution result for the objective problem using a preset solution algorithm until the objective solution result meets the preset requirements. Then, it outputs the optimal decision variables for the target agent to perform the target action.

[0086] In some specific implementations, when the agent-driven model is an agent game model, the target problem is iteratively calculated multiple times using a preset solution algorithm based on the target learning parameters until the target solution meets the preset requirements. The optimal control parameters for the target agent to execute the target action are then output, including: Based on the state space, action space, environmental parameters, reward function, and target requirement information, the target solution result corresponding to the target problem is calculated iteratively multiple times using a preset solution algorithm until the target solution result meets the preset requirements. Then, the optimal update variable for the target agent to perform the target action is output.

[0087] In some specific embodiments, the agent-driven control method of the present invention further includes: The optimal control parameters for the target agent to perform the target action are input into the behavior policy generation unit to obtain the behavior control policy for the target agent to perform the target action in the current state.

[0088] In some specific embodiments, the agent-driven control method of the present invention further includes: The behavior control strategy is fed back to the simulation platform to execute actions, and the attribute characteristics information of the target intelligent agent and the simulation environment information are updated.

[0089] The intelligent agent-driven control device provided in this embodiment of the invention can execute the intelligent agent-driven control method provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects for executing the method. Further functional descriptions of the various modules and units described above are the same as in the corresponding embodiments described above, and will not be repeated here.

[0090] Figure 6This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention.

[0091] The following is a detailed reference. Figure 6 , Figure 6 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention.

[0092] The following is a detailed reference. Figure 6 This diagram illustrates a suitable structural schematic for implementing an electronic device according to embodiments of the present invention. The electronic device may include a processor (e.g., a central processing unit, graphics processor, etc.) 601, which can perform various appropriate actions and processes based on a program stored in read-only memory (ROM) 602 or a program loaded from memory 608 into random access random access memory (RAM) 603. The RAM 603 also stores various programs and data required for the operation of the electronic device. The processor 601, ROM 602, and RAM 603 are interconnected via a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.

[0093] Typically, the following devices can be connected to I / O interface 605: input devices 606 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 607 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; memory devices 608 including, for example, magnetic tapes, hard disks, etc.; and communication devices 609. Communication device 609 allows electronic devices to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 6 Electronic devices with various devices are shown, but it should be understood that it is not required to implement or have all of the devices shown, and more or fewer devices may be implemented or have instead.

[0094] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device 609, or installed from a memory 608, or installed from a ROM 602. When the computer program is executed by the processor 601, it performs the functions defined in the intelligent agent-driven control method of the embodiments of the present invention.

[0095] Figure 6 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments of the present invention.

[0096] This invention also provides a computer-readable storage medium. The methods described above according to embodiments of the invention can be implemented in hardware or firmware, or implemented as computer code that can be recorded on a storage medium, or implemented as computer code downloaded via a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and then stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code. When the software or computer code is accessed and executed by the computer, processor, or hardware, the agent-driven control method shown in the above embodiments is implemented.

[0097] A portion of this invention can be applied as a computer program product, such as computer program instructions, which, when executed by a computer, can invoke or provide the methods and / or technical solutions according to the invention through the operation of the computer. Those skilled in the art will understand that the forms in which computer program instructions exist in a computer-readable medium include, but are not limited to, source files, executable files, installation package files, etc. Correspondingly, the ways in which computer program instructions are executed by a computer include, but are not limited to: the computer directly executing the instructions, or the computer compiling the instructions and then executing the corresponding compiled program, or the computer reading and executing the instructions, or the computer reading and installing the instructions and then executing the corresponding installed program. Here, the computer-readable medium can be any available computer-readable storage medium or communication medium accessible to a computer.

[0098] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A control method driven by an intelligent agent, characterized in that, The method includes: Acquire the attribute characteristics of the target intelligent agent, the simulation environment information, and the target user's target requirement information; Determine the agent-driven model corresponding to the target requirement information, and obtain the target problem of the agent-driven model; The attribute feature information, the simulation environment information, and the target problem are input into the agent-driven model for learning to obtain the target learning parameters; Based on the target learning parameters, the target solution result corresponding to the target problem is calculated iteratively multiple times using a preset solution algorithm until the target solution result meets the preset requirements. Then, the optimal control parameters for the target agent to perform the target action are output.

2. The method according to claim 1, characterized in that, The attribute feature information includes: parameters, events, variables, and functions of the target agent; the simulation environment information includes: link information, state information, and state transition conditions of the target agent; the target requirement information includes: number of groups, number of iterations, simulation dimension, random seed, crossover rate, and convergence parameters.

3. The method according to claim 1 or 2, characterized in that, When the agent-driven model is an agent optimization model, the target learning parameters include: objective function, decision variables, constraints, and parameter boundaries; when the agent-driven model is an agent game model, the target learning parameters include: state space, action space, environmental parameters, and reward function.

4. The method according to claim 3, characterized in that, When the agent-driven model is an agent optimization model, based on the target learning parameters, the target problem is iteratively calculated multiple times using a preset solution algorithm until the target solution result meets the preset requirements. The optimal control parameters for the target agent to execute the target action are then output, including: Based on the objective function, the decision variables, the constraints, the parameter boundaries, and the target requirement information, the target solution result corresponding to the target problem is calculated iteratively multiple times using a preset solution algorithm until the target solution result meets the preset requirements. Then, the optimal decision variables for the target agent to perform the target action are output.

5. The method according to claim 3, characterized in that, When the agent-driven model is an agent game model, based on the target learning parameters, the target problem is iteratively calculated multiple times using a preset solution algorithm until the target solution result meets the preset requirements. The optimal control parameters for the target agent to execute the target action are then output, including: Based on the state space, the action space, the environmental parameters, the reward function, and the target requirement information, the target solution result corresponding to the target solution problem is calculated iteratively multiple times using a preset solution algorithm until the target solution result meets the preset requirements. Then, the optimal update variable for the target agent to perform the target action is output.

6. The method according to claim 1, 4, or 5, characterized in that, The method further includes: The optimal control parameters for the target agent to perform the target action are input into the behavior policy generation unit to obtain the behavior control policy for the target agent to perform the target action in the current state.

7. The method according to claim 6, characterized in that, The method further includes: The behavior control strategy is fed back to the simulation platform to execute actions, and the attribute feature information and simulation environment information of the target intelligent agent are updated.

8. A control device driven by an intelligent agent, characterized in that, The device includes: The acquisition module is used to acquire the attribute feature information of the target intelligent agent, the simulation environment information, and the target demand information sent by the target user; The determination module is used to determine the agent-driven model corresponding to the target requirement information and obtain the target problem of the agent-driven model. The learning module is used to input the attribute feature information, the simulation environment information, and the target problem into the agent-driven model learning to obtain target learning parameters; The calculation module is used to calculate the target solution result corresponding to the target problem multiple times using a preset solution algorithm based on the target learning parameters, until the target solution result meets the preset requirements, and output the optimal control parameters for the target agent to perform the target action.

9. An electronic device, characterized in that, include: A memory and a processor are communicatively connected, the memory stores computer instructions, and the processor executes the computer instructions to perform the agent-driven control method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing the computer to execute the agent-driven control method according to any one of claims 1 to 7.

Images