Construction method of interactive simulation model among multiple subjects, interactive simulation method among multiple subjects and related device

Abstract

The invention discloses a construction method of an interactive simulation model among multiple subjects, an interactive simulation method among multiple subjects, a computer program product, electronic equipment and a computer readable storage medium, and the construction method of the interactive simulation model among multiple subjects comprises the following steps: determining a plurality of subjects in a target interactive scene, attribute data corresponding to the main bodies are obtained; determining a behavior interaction rule among the plurality of main bodies and an association relationship between the behavior interaction rule and an environment state; utilizing the attribute data, the behavior interaction rule and the incidence relation to construct a multi-body interaction simulation model with the deviation smaller than or equal to a deviation threshold value; wherein the deviation is the deviation between a simulation result output by the multi-agent interaction simulation model and an expected interaction result, and the behavior interaction rule can dynamically change based on the deviation. The model simulation precision can be improved.

Description

Technical Field

[0001] The present application relates to the field of multi-agent interaction simulation, and in particular to a method for constructing a multi-agent interaction simulation model, a multi-agent interaction simulation method, a computer program product, an electronic device, and a computer-readable storage medium. Background Art

[0002] While multi-agent modeling in related fields, such as social governance, smart education, and rural credit, has been widely applied in data integration, event response, and static rule modeling, achieving some progress, it still faces significant technical shortcomings in simulating the interactions between different agents in complex social systems. Currently, the accuracy of simulations of multi-agent interactions in various interaction scenarios is poor. Summary of the Invention

[0003] To address the above technical issues, the present application provides a method for constructing a multi-agent interaction simulation model, a multi-agent interaction simulation method, a computer program product, an electronic device, and a computer-readable storage medium. The technical solutions are as follows:

[0004] According to a first aspect of the present application, a method for constructing a multi-agent interaction simulation model is provided, the method comprising:

[0005] Determine several subjects in the target interaction scenario, and obtain attribute data corresponding to each of the several subjects;

[0006] Determining behavioral interaction rules among the plurality of subjects, and an association between the behavioral interaction rules and an environmental state;

[0007] Constructing a multi-agent interaction simulation model with a deviation less than or equal to a deviation threshold using the attribute data, the behavioral interaction rules, and the association relationship;

[0008] Among them, the deviation is the deviation between the simulation result output by the multi-subject interaction simulation model and the expected interaction result, the behavioral interaction rule can change dynamically based on the deviation, the simulation result is the simulation data of the target association relationship changing over time, the expected interaction result is the expected data of the target association relationship changing over time, and the target association relationship is the relationship between the interaction behavior between the several subjects and the environmental state.

[0009] According to a second aspect of the present application, a multi-agent interaction simulation method is provided, the method comprising:

[0010] Determine several subjects in the target interaction scenario, and obtain attribute data corresponding to each of the several subjects;

[0011] Determining behavioral interaction rules among the plurality of subjects, and an association between the behavioral interaction rules and an environmental state;

[0012] The attribute data, the behavioral interaction rules and the association relationship are input into the multi-agent interaction simulation model as described in the first aspect, and the simulation results are output through the multi-agent interaction simulation model. The simulation results are simulation data of the target association relationship changing over time, and the target association relationship is the relationship between the interaction behavior between the several subjects and the environmental state.

[0013] According to a third aspect of the present application, a computer program product is provided. The computer program product includes a computer program. When the computer program is executed by a processor, the method according to the first aspect or the second aspect is implemented.

[0014] According to a fourth aspect of the present application, an electronic device is provided, comprising:

[0015] processor;

[0016] a memory for storing processor-executable instructions;

[0017] The processor is configured to implement the method as described in the first aspect or the second aspect.

[0018] According to a fifth aspect of the present application, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the method described in the first aspect or the second aspect are implemented.

[0019] The technical solution provided in this application first clarifies the various interactive subjects in the target interaction scenario, collects the attribute data of each subject, and sets behavioral interaction rules that are correlated between different subjects and the environmental state. Based on this, a multi-subject interaction simulation model is constructed. The set behavioral interaction rules are dynamically adjusted based on the deviation between the simulation results output by the multi-subject interaction simulation model and the expected interaction results. Finally, a constructed multi-subject interaction simulation model with a deviation less than or equal to the deviation threshold is obtained, which can accurately simulate simulation data of the relationship between the interactive behaviors of several subjects and the environmental state that changes over time, thereby improving the model simulation accuracy.

[0020] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following briefly introduces the drawings required for use in the embodiments or related technical descriptions. Obviously, the drawings described below are only some embodiments recorded in this application. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0022] Figure 1 It is a schematic diagram of a specific simulation scenario in the related art;

[0023] Figure 2 This is a flowchart of a method for constructing a multi-agent interaction simulation model according to an embodiment of the present application;

[0024] Figure 3 This is a schematic diagram of a specific simulation scenario of an embodiment of the present application;

[0025] Figure 4 This is a flowchart of a multi-agent interaction simulation method according to an embodiment of the present application;

[0026] Figure 5 It is a structural diagram of an electronic device according to an embodiment of the present application. DETAILED DESCRIPTION

[0027] In order to enable those skilled in the art to better understand the technical solutions in this application, the technical solutions in the embodiments of this application will be described in detail below in conjunction with the drawings in the embodiments of this application. Obviously, the described embodiments are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art should fall within the scope of protection of this application.

[0028] Although multi-agent modeling in related technologies has been widely used in data integration, event response, and static rule modeling in areas such as social governance, smart education, and rural credit, and has made some progress, it still faces significant technical shortcomings in simulating the behavioral interactions of different agents in complex social systems. Currently, the simulation accuracy of multi-agent interactions in various interaction scenarios is poor. For example:

[0029] First, most of the related technologies remain at the static modeling level, lacking high-fidelity modeling and dynamic feedback regulation mechanisms for multi-agent interaction behaviors. Figure 1 As shown, the simulation methods of related technologies are mostly based on fixed rules or expert-set thresholds, which makes it difficult to adjust the subject behavior rules in real time according to changes in system status (such as Figure 1, assuming that there are three subjects in the target environment, "subject 1", "subject 2", "subject 3", the related technology often assumes that the three subjects behave in the target environment with fixed rules during simulation), which cannot adapt to the rapid evolution of social scenes such as public opinion, education or rural credit, resulting in rigid simulation models, insufficient prediction accuracy and response flexibility. In addition, in the existing multi-agent based city simulation method, although the micro-selection theory is used to describe individual behavior, the modeling granularity focuses on discrete selection behavior, lacks a closed-loop coupling mechanism with the system macro state, and is difficult to realize dynamic functions such as strategy feedback and resource redistribution in the governance system. It can be seen that the simulation of the interaction between multiple subjects in various interaction scenarios in the related technology has the problems of poor flexibility, lack of dynamic feedback mechanism, difficulty in accurately depicting behavior evolution, and insufficient universality and scalability, resulting in poor simulation accuracy.

[0030] To solve the above problems, the application provides a method for constructing a multi-agent interaction simulation model, which can improve the simulation accuracy of the model. As shown in Figure 2 The method comprises the following steps:

[0031] S201, determining a plurality of subjects in a target interaction scenario, and obtaining attribute data corresponding to each of the plurality of subjects.

[0032] S202, determining the behavior interaction rules between the plurality of subjects, and the association relationship between the behavior interaction rules and the environment state.

[0033] S203, using the attribute data, the behavior interaction rules and the association relationship to construct a multi-agent interaction simulation model with a deviation less than or equal to a deviation threshold.

[0034] Wherein, the deviation is the deviation between the simulation result output by the multi-agent interaction simulation model and the expected interaction result, the behavior interaction rules can be dynamically changed based on the deviation, the simulation result is the simulation data of the target association relationship changing with time, the expected interaction result is the expected data of the target association relationship changing with time, and the target association relationship is the relationship between the interaction behavior of the plurality of subjects and the environment state.

[0035] The technical solution provided in the embodiment of the present application first clarifies the various interactive subjects in the target interaction scenario, collects the attribute data of each subject, and sets behavioral interaction rules that are correlated between different subjects and environmental states. Based on this, a multi-subject interaction simulation model is constructed. The set behavioral interaction rules are dynamically adjusted based on the deviation between the simulation results output by the multi-subject interaction simulation model and the expected interaction results. Finally, a constructed multi-subject interaction simulation model with a deviation less than or equal to the deviation threshold is obtained, which can accurately simulate simulation data of the relationship between the interactive behaviors of several subjects and the environmental states that changes over time, thereby improving the model simulation accuracy.

[0036] It is understandable that the target interaction scenario can have multiple specific implementations, and different target interaction scenarios can also have different subjects. After determining the specific target interaction scenario, the specific subjects participating in the interaction can be determined based on the target interaction scenario.

[0037] As an embodiment, each subject in the target interaction scenario can represent an intelligent entity with independent behavioral logic and decision-making capabilities. The type of subject can have multiple specific implementations. As an example, the type of subject can include individual user subjects, organizational management subjects, institutional regulatory subjects or special function subjects, or other types of subjects. As another example, individual user subjects can include subjects such as farmers, students, community residents, and ordinary netizens. Organizational management subjects can include subjects such as teachers, education managers, community managers, and social platform operators. Institutional regulatory subjects can include subjects such as governments or higher-level regulatory agencies. Special function subjects can include subjects such as opinion leaders, virtual users, and corporate platforms.

[0038] It is worth noting that the above introduction to the specific implementation of the subject type is only an exemplary display. In actual application, other specific implementations are not excluded and are not specifically limited to this.

[0039] The several subjects in the above-mentioned target interaction scenario can respectively correspond to different role functions, and the attribute data corresponding to the several subjects can include static attributes and dynamic behavior variables. The static attributes can reflect the prior fixed characteristics of the subject, and the dynamic behavior variables can dynamically change based on the deviation between the simulation results output by the above-mentioned multi-subject interaction simulation model and the expected interaction results.

[0040] The static attributes and dynamic behavioral variables included in the aforementioned attribute data can have various specific implementations. As an example, for the aforementioned individual user subjects, static attributes may include: age, gender, education level, occupational role, etc.; for the aforementioned organizational management subjects or institutional regulatory subjects, static attributes may include: management resources, policy-making capabilities, etc.; for the aforementioned special function subjects, static attributes may include: technical usability, functional design parameters, etc. As another example, dynamic behavioral variables can be data used to describe the subject's behavioral intentions or state transitions during the model evolution process. For example, dynamic behavioral variables may include: credit literacy, learning ability, behavioral improvement rate, technology acceptance, adoption willingness, interaction frequency, or public opinion trend, posting probability, content update intensity, etc.

[0041] It is understandable that the selection of the above attribute data must meet the principles of observability and controllability to facilitate subsequent interactive behavior modeling and feedback mechanism embedding.

[0042] There are various ways to obtain the attribute data corresponding to each of the aforementioned agents. As an example, historically collected data associated with each of the aforementioned agents can be obtained, and then, based on principal component analysis and clustering, the attribute data corresponding to each of the agents can be extracted from the acquired historically collected data. As another example, the attribute data can be fixed at the initial moment of acquisition and dynamically change during model evolution based on the deviation between the simulation results output by the multi-agent interaction simulation model and the expected interaction results.

[0043] As another example, the attribute data can also be obtained or constructed through questionnaire surveys, historical data extraction, or simulation generation based on empirical distribution.

[0044] The above-mentioned target interaction scenarios can be implemented in multiple ways. As an example, the above-mentioned target interaction scenarios include: rural credit system evaluation scenarios, smart education scenarios, community governance scenarios or public opinion governance scenarios; in the above-mentioned rural credit system evaluation scenarios, the several subjects include at least farmers, social organizations and credit platforms; the attribute data of the several subjects include at least: farmers' historical credit records, information representing farmers' literacy in using credit platforms and information representing social behavioral norms; in the smart education scenarios, the several subjects include at least students, teachers and education managers; the attribute data include at least: information representing students' learning ability, information representing teachers' teaching familiarity and information representing education managers' resource allocation tendencies; in the community governance scenarios, the several subjects include at least residents, community managers and potential criminals; the attribute data include at least: information representing residents' perceived usefulness of smart access control, information representing community managers' publicity investment and information representing the decision-making logic of potential criminals; in the public opinion governance scenarios, the several subjects include at least netizens, the government and the platform; the attribute data include at least: information representing netizens' posting behavior, information representing the government's regulatory intensity and information representing the platform's execution ability.

[0045] It is understandable that the behavioral interaction rules between the above-mentioned subjects can be used to guide the dynamic generation and response mechanism of the behaviors between different subjects in the simulation process. There are many ways to determine the behavioral interaction rules between several subjects. As an example, the behavioral interaction rules can be constructed based on the multi-agent modeling (Agent-Based Modeling, ABM) technology. It is understandable that ABM technology can be used as an important tool for complex system research due to its excellent simulation ability and predictive performance. Figure 3 As shown, ABM can effectively characterize the behavioral logic and interactions of individuals (agents) or organizations in complex environments (environments) by introducing intelligent agents with autonomous decision-making capabilities and combining rule constraints with dynamic feedback mechanisms. It is widely used in scenarios such as policy evaluation, technology diffusion, and social behavior intervention. ABM technology can serve as the core support for intelligent social governance systems. When ABM technology is applied in public opinion governance, the design and behavior of virtual users can be dynamically adjusted based on the network context to guide the orderly development of public sentiment. When ABM technology is applied in smart education, the interaction between teachers and students and the iteration of teaching strategies can be updated in real time with the help of model data, thereby optimizing the allocation of teaching resources and evaluation mechanisms. When ABM technology is applied in rural credit development, factors such as credit game between entities, concealment behavior, and regulatory pressure can be expressed in the form of rule constraints and integrated into the model evolution trajectory, enabling the system to dynamically reflect the behavioral characteristics and risk levels of farmers.

[0046] As another example, another way to determine the behavioral interaction rules between several subjects may include: the behavioral interaction rules can be constructed based on the Technology Acceptance Model (TAM), which is a core theoretical framework in the field of information systems for explaining and predicting users' acceptance behavior of new technologies. The core idea of ​​TAM is that users' willingness to adopt new technologies is mainly driven by two key factors: perceived usefulness (PU) and perceived ease of use (PEOU). The core components of TAM include perceived usefulness (PU), perceived ease of use (PEOU), behavioral intention (BI) and actual use (Actual Use), as well as external variables. Among them, perceived usefulness (PU) refers to the degree to which users subjectively believe that using a certain technology can improve their work performance or life efficiency. For example, if students believe that online learning platforms can improve their learning efficiency, they are more likely to continue using them. Perceived ease of use (PEOU) refers to the subjective perception of the difficulty of learning or operating a certain technology. For example, if a mobile payment application has a simple interface and intuitive operation, users will have higher acceptance. PU and PEOU jointly influence user attitudes, which in turn determines behavioral intentions (whether to plan to use it), and ultimately translates into actual usage behavior. That is, the logical chain is: external variables → PEOU → PU → attitude → behavioral intention → actual use. External variables (External Variables) include system design, user characteristics, social norms, policy environment, etc., which play a role by indirectly influencing PU and PEOU. For example, technical training can improve users' perception of system usability.

[0047] As another example, another way to determine the behavioral interaction rules among several subjects may include: the behavioral interaction rules can be constructed based on a Markov Chain. The Markov Chain is a random process with "no memory" (Markov property). Its core feature is that the probability distribution of the next state of the system depends only on the current state and is independent of the historical state.

[0048] As an example, behavioral interaction rules can be set for each type of subject based on the types of subjects and their corresponding roles and functions, combined with behavioral models (such as the aforementioned ABM, TAM, and Markov chains). As another example, the behavioral logic within these interaction rules can typically be driven by the following three mechanisms: ① Intrinsic individual motivations: for example, farmers pursuing credit returns, students pursuing academic performance, residents seeking safety and convenience, and netizens pursuing a desire for expression or influence; ② External environmental stimuli: such as policy pressure, community norms, platform rules, publicity investment, and system design, which directly influence the subject's behavioral activation threshold or choice preferences; ③ Group influence mechanisms: These mechanisms promote the diffusion or suppression of social behavior through imitation, comparison, feedback and punishment, and other mechanisms, such as the role model effect, guidance from opinion leaders, and the usage rate of surrounding farmers. As another example, the behavioral logic can be formalized as a decision function for the subject. One implementation of this function could be: state input → decision rule → behavior output. Examples include: "credit literacy + normative pressure → willingness to participate"; "perceived usefulness + imitation effect → technology adoption decision"; and "platform public opinion intensity + social needs → posting frequency." As another example, the decision function can take the form of a linear combination (e.g., weighted sum), a threshold function (step / sigmoid), a probability transition (Markov chain), or a game strategy (multi-agent game), etc., without specific limitation. It is understood that the behavioral interaction rules should support parameter adjustment, causal interpretation, and reproducibility of simulation deduction.

[0049] The relationship between the behavioral interaction rules among the aforementioned entities and the environmental state can be determined in a variety of ways. As an example, a state response function and feedback adjustment coefficient can be first determined. Based on this state response function, the environmental state can be mapped in real time to the behavioral interaction parameters associated with the behavioral interaction rules. The sensitivity and intensity of this mapping can be controlled by the feedback adjustment coefficient. Based on the mapped behavioral interaction parameters, the relationship between the behavioral interaction rules among the several entities and the environmental state can be determined. Determining the relationship between the behavioral interaction rules and the environmental state based on the state response function and feedback adjustment coefficient enables the model to dynamically perceive and regulate changes in the environmental state, supporting the continuity and stability of multi-cycle evolutionary simulations.

[0050] When constructing the aforementioned multi-agent interaction simulation model, a closed-loop logic of behavior-state-feedback can be followed to provide the model with adaptive adjustment capabilities. First, a closed-loop mapping of agent-variable-behavior can be constructed: a correspondence between agent types and their attribute variables can be established, and the static and dynamic attributes of the agent can be input into the decision function to generate behavioral outputs. Simultaneously, the impact path of the behavioral results on the environmental state can be set, allowing the behavioral output to feedback and act on environmental state changes, providing a foundation for subsequent behavioral evolution. Next, the aforementioned state response function and feedback adjustment coefficient can be introduced to design a state response function, mapping key system states (such as rating error, satisfaction, system pressure, etc.) into influencing inputs for behavioral rules, allowing the behavioral logic to dynamically adjust with environmental states. The feedback adjustment coefficient can be set to control the rate and intensity of the impact of state changes on the behavioral logic, thereby achieving adaptive updates of the parameters or input structure in the decision function. Finally, a closed-loop structure of "state-behavior-feedback-re-behavior" is formed, in which the environmental state is updated through the behavioral results. The environmental state is fed back to the behavioral interaction rules through the state response function and the feedback adjustment coefficient, and the behavioral path is dynamically modified to achieve synchronous adjustment of behavioral evolution and environmental response, thus constructing a sustainable and dynamically controllable multi-agent interaction simulation model.

[0051] Before the multi-agent interaction simulation model is run, data loading, parameter setting, and simulation configuration can be performed, and a periodic simulation process can be executed. As an example, during the startup phase of the multi-agent interaction simulation model, initial values ​​for static and dynamic attribute variables (i.e., the static attributes and dynamic behavior variables included in the attribute data) can be set for each agent. For example, initial values ​​can be obtained through questionnaires or field surveys to obtain individual characteristics and behavioral preferences, which is suitable for small-sample precision modeling. Initial values ​​can also be extracted from behavior logs or databases by sampling historical data, which is suitable for model calibration and verification. Initial values ​​can also be generated from empirical distributions, that is, generating large-scale simulated individuals based on preset statistical distributions, which is suitable for complex system simulation. Initial values ​​can be assigned through rule configuration, that is, static settings are used for some fixed parameters (such as role labels and initial permissions). For attribute variables with feedback structures (such as perceived risk, policy response, and technology trust), initial weights can be assigned based on policy goals or intervention scenarios to enhance the adaptability and dynamic control capabilities of the multi-agent interaction simulation model.

[0052] As another example, for multi-agent interaction simulation models under different target interaction scenarios, key parameters controlling model evolution can be set based on the simulation objectives and specific application scenarios. These parameters include: public opinion propagation rate, platform regulatory intensity; credit score error, policy enforcement; educational technology adoption threshold, learning adaptation rate, etc. These parameters can influence the functional structure of behavioral interaction rules, the sensitivity of feedback adjustment coefficients, and the overall operational path of the model. As another example, the operation of a multi-agent interaction simulation model can be based on periodic updates, executing the following process: ① Agent behavior generation: Each agent generates behavioral output based on its current attribute state and decision function; ② Interaction and feedback execution: The behavioral results affect the environmental state and, through feedback mechanisms, affect the attributes of the next round of agents; ③ State variable recording and tracking: Real-time recording of the evolutionary trajectory of key variables such as credit level, emotional state, adoption behavior, and policy response, providing a basis for subsequent analysis, model tuning, and evaluation.

[0053] The multi-agent interaction simulation model described above can be optimized in a variety of ways. As an example, the model's adaptability to real-world scenarios can be verified first: the simulation results output by the multi-agent interaction simulation model can be compared and analyzed with actual survey data, historical statistical data, or typical scenario cases to evaluate the model's fit in terms of behavioral trends, variable distribution, feedback paths, and so on. This can then determine the explanatory power and adaptability of the model structure and parameter configuration to the target interaction scenario. The model parameters and behavioral interaction rules can then be adjusted in reverse: based on the deviation between the simulation results and the real-world data, key parameters in the multi-agent interaction simulation model (such as feedback adjustment weights, behavioral sensitivity, and environmental response rate) can be corrected, or the decision function structure in the behavioral interaction rules can be adjusted. Optimization can be performed through sensitivity analysis, error function evaluation, or automated calibration mechanisms to improve the model's predictive accuracy and stability. Finally, after the multi-agent interaction simulation model is operational and structurally adjusted, its output data (i.e., simulation results) can be further analyzed as follows: ① Tracking behavioral variable evolution: Analyzing the dynamic trends over time in various agent behaviors (such as adoption, participation, and posting); ② Tracking state variable changes: Recording and comparing the cyclical evolution paths of the system's core state variables (such as credit level, learning performance, sense of security, and public opinion); ③ Identifying key features and abnormal phases: Discovering inflection points, mutation intervals, or abnormal system fluctuations during the model's operation, providing decision-making references for policy evaluation and mechanism design. Through this analysis and optimization process, the model can continuously improve its ability to realistically depict complex social systems, enhance its predictive support for policy intervention, behavioral regulation, and system evolution trends, and thus further enhance the simulation accuracy of the multi-agent interaction simulation model.

[0054] As an example, if the expected data (i.e., expected interaction result) of the relationship between the interaction behavior of the plurality of subjects and the environmental state (i.e., target association relationship) changing over time and the simulation result (i.e., simulation data of the relationship between the interaction behavior of the plurality of subjects and the environmental state changing over time) output by the multi-subject interaction simulation model both include a plurality of data types, corresponding weights are set for different data types, and the deviation between the simulation result output by the multi-subject interaction simulation model and the expected interaction result is determined by weighted summation based on the weights corresponding to various data types.

[0055] The specific multi-subject interaction simulation scenarios of the embodiments of the present application will be exemplarily described below in combination with four specific target interaction scenarios (scenario one, scenario two, scenario three, and scenario four):

[0056] I. Rural credit system evaluation scenario:

[0057] 1. Data generation and initialization

[0058] According to the investigation data, the virtual data sets of farmers, social organizations, and credit platforms are generated. The farmer data includes historical credit records, literacy in using credit platforms, social behavior norms, etc. Virtual data can be generated through scripts written in Python, and the generated results can be stored in Comma-Separated Values (CSV) files for subsequent use.

[0059] 2. Farmer credit behavior modeling

[0060] The formula update_alpha() can be used to simulate the dynamic change of the literacy of farmers in using the credit platform. The improvement of literacy is jointly determined by the endogenous learning ability of farmers and the exogenous support of social organizations. The dynamic change of the literacy of farmers in using the credit platform can be defined by the following formula:

[0061] α i,t+1 =α i,t +β ext ·S o,t +β int ·α i,t (1)

[0062] Wherein, α i,t represents the literacy level of i subjects at time t; β ext represents the exogenous learning rate, β int represents the endogenous learning rate, and S o,t represents the support level of social organizations.

[0063] 3. Credit platform scoring mechanism

[0064] The credit platform generates credit scores based on the credit data provided by farmers. The scores are related to farmers' social behavior and data quality. The scoring error is gradually reduced through the optimization feedback mechanism and can be expressed by the following formula:

[0065] Δ t =Δ t-1 -γ·(Observed Effect-Target Effect) (2)

[0066] Among them, Δ t represents the scoring error or system deviation at time t; Observed Effect represents the actual effect (such as credit score, learning performance, etc.) obtained through model simulation or data observation; Target Effect represents the system expectation or policy target effect; γ is the optimization rate of the credit platform.

[0067] 4. Simulation results

[0068] Through the iteration of the multi-agent interaction simulation model, the changing trend of farmers' behavior after the promotion of the credit platform was simulated, and the evolution trajectory of farmers' participation strategies was generated.

[0069] 2. Modeling and simulation of smart education scenarios:

[0070] 1. Educational Dataset Generation

[0071] Based on actual survey results, a virtual dataset of students, teachers, and education administrators is generated, including information on student learning ability, teacher teaching familiarity, and administrator resource allocation preferences. Rules can be used to ensure the rationality and representativeness of the data.

[0072] 2. Modeling student learning behavior

[0073] The student's learning state can be modeled using a Markov process, and its state transition probability is defined as: P(S t+1 |S t )=f(E t ,R t ) (3)

[0074] Among them, S t ,S t+1 They represent the status of the student at time t and time t+1 respectively, E t For educational resource investment, R t Access to learning resources for students.

[0075] 3. Decision Modeling for Education Administrators

[0076] Education administrators adjust resource allocation strategies based on student performance and education budgets. The decision formula is:

[0077] It+1 =I t +δ·(Budget Effect+Policy Effect) (4)

[0078] Among them, I t ,I t+1 They represent the policy or resource intervention intensity at time t and time t+1 respectively; Budget Effect represents the impact of budget input effect, Policy Effect represents the impact of policy adjustment, and δ is the manager's adjustment rate.

[0079] 4. Simulation results

[0080] Through the simulation of a multi-agent interaction simulation model, the impact of different education policies on student learning outcomes and system resource utilization efficiency was analyzed, and a significant improvement trend in student performance was found under a high budget.

[0081] 3. Modeling and simulation of community governance scenarios:

[0082] 1. Community data generation

[0083] Generate a virtual dataset of residents, community managers, and potential criminals, covering residents' perceived usefulness of smart access control, community managers' publicity investment, and potential criminals' decision-making logic.

[0084] 2. Manager Promotion Strategy Modeling

[0085] Managers’ publicity input affects residents’ perceived usefulness through the following formula:

[0086] PU t+1 =PU t +λ·(Promotion Effect) (5)

[0087] Among them, PU t ,PU t+1 represents the residents' perceived usefulness of the smart access control system at time t and t+1; PromotionEffect represents the effect of the promotional investment of the community manager or platform; λ represents the effectiveness of the promotional investment.

[0088] 3. Crime rate evolution modeling

[0089] The logit model is used to describe the criminal tendency of criminals. The expression is as follows:

[0090]

[0091] Among them, C trepresents the sense of security of residents or society at time t; Security Effect represents the impact of security measures or intelligent systems in reducing crime risks; k is the impact coefficient.

[0092] 4. Simulation results

[0093] The results show that the continuous investment strategy is significantly effective in reducing crime rates and improving technology acceptance, while the effect of the decreasing strategy gradually weakens.

[0094] 4. Modeling and simulation of public opinion governance scenarios:

[0095] 1. Public opinion data generation

[0096] Generate a virtual data set of netizens, governments, and platforms, including netizens' posting behavior, government supervision, and platform execution.

[0097] 2. Public Opinion Evolution Modeling

[0098] The platform's public opinion control power is defined as:

[0099] C t =P t ·G t (7)

[0100] Among them, P t For platform execution, G t For government supervision.

[0101] 3. Simulation results

[0102] The simulation results show that coordination between the platform and the government can effectively slow down the deterioration of public opinion, and moderate subsidy policies can further enhance governance effects.

[0103] Based on the multi-agent interaction simulation model constructed by the construction method of the multi-agent interaction simulation model described in any of the above embodiments, that is, the preset multi-agent interaction simulation model, the embodiment of the present application also provides a multi-agent interaction simulation method, such as Figure 4 As shown, the method includes the following steps:

[0104] S401: Determine several subjects in a target interaction scenario, and obtain attribute data corresponding to the several subjects.

[0105] S402: Determine the behavior interaction rules among the plurality of subjects, and the association relationship between the behavior interaction rules and the environment state.

[0106] S403: Input the attribute data, the behavior interaction rules, and the association relationship into a preset multi-agent interaction simulation model, and output a simulation result through the multi-agent interaction simulation model.

[0107] The simulation result is simulation data of a target correlation changing over time, and the target correlation is a relationship between interaction behaviors among the plurality of agents and an environment state.

[0108] The construction manner of the preset multi-agent interaction simulation model can refer to the construction method of the multi-agent interaction simulation model described in any of the embodiments above, and details are not described herein again.

[0109] Corresponding to the method embodiments described above, the present application also provides a computer program product, comprising a computer program, which, when executed by a processor, implements the construction method of the multi-agent interaction simulation model or the multi-agent interaction simulation method described in any of the embodiments above.

[0110] The present application also provides an electronic device, such as Figure 5 As shown in the figure, the electronic device comprises:

[0111] a processor 501;

[0112] a memory 502 for storing processor-executable instructions;

[0113] The processor 501 is configured to implement the construction method of the multi-agent interaction simulation model or the multi-agent interaction simulation method described in any of the embodiments above.

[0114] The present application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the construction method of the multi-agent interaction simulation model or the multi-agent interaction simulation method described in any of the embodiments above.

[0115] The above is only a specific embodiment of the present application, and it should be noted that for those skilled in the art, without departing from the principles of the present application, a number of improvements and refinements can be made, which should also be considered as the protection scope of the present application.

Claims

1. A method for constructing a multi-agent interaction simulation model, characterized in that: The method comprises: Determine several subjects in the target interaction scenario, and obtain attribute data corresponding to each of the several subjects; Determining behavioral interaction rules among the plurality of subjects, and an association between the behavioral interaction rules and an environmental state; Constructing a multi-agent interaction simulation model with a deviation less than or equal to a deviation threshold using the attribute data, the behavioral interaction rules, and the association relationship; Among them, the deviation is the deviation between the simulation result output by the multi-subject interaction simulation model and the expected interaction result, the behavioral interaction rule can change dynamically based on the deviation, the simulation result is the simulation data of the target association relationship changing over time, the expected interaction result is the expected data of the target association relationship changing over time, and the target association relationship is the relationship between the interaction behavior between the several subjects and the environmental state.

2. The method according to claim 1, characterized in that The acquiring of attribute data corresponding to the plurality of subjects includes: Obtaining historical collected data associated with each of the plurality of subjects; The attribute data is extracted from the historically collected data based on principal component analysis and clustering.

3. The method according to claim 1, characterized in that The several subjects respectively correspond to different role functions; The attribute data includes static attributes and dynamic behavior variables, wherein the static attributes can reflect the prior fixed characteristics of the subject, and the dynamic behavior variables can dynamically change based on the deviation.

4. The method according to claim 1, wherein The association between the behavior interaction rules and the environment state is determined in the following way: Determine the state response function and feedback adjustment coefficient; Based on the state response function, the environmental state is mapped in real time to the behavior interaction parameters associated with the behavior interaction rules; wherein the sensitivity and intensity of the mapping are controlled by the feedback adjustment coefficient; The association relationship is determined based on the behavioral interaction parameters.

5. The method according to claim 1, wherein The target interaction scenarios include: rural credit system evaluation scenarios, smart education scenarios, community governance scenarios, or public opinion governance scenarios; In the rural credit system evaluation scenario, the multiple entities include at least farmers, social organizations, and credit platforms; the attribute data include at least: farmers' historical credit records, information representing farmers' literacy in using credit platforms, and information representing social behavioral norms; In the smart education scenario, the multiple subjects include at least students, teachers, and education administrators; the attribute data includes at least: information representing students' learning ability, information representing teachers' teaching familiarity, and information representing education administrators' resource allocation tendencies; In the community governance scenario, the multiple subjects include at least residents, community managers, and potential criminals; the attribute data includes at least: information representing residents' perceived usefulness of smart access control, information representing community managers' publicity investment, and information representing the decision-making logic of potential criminals; In the public opinion governance scenario, the several subjects include at least netizens, the government and the platform; the attribute data includes at least: information representing the posting behavior of netizens, information representing the government's regulatory strength and information representing the platform's execution ability.

6. The method according to claim 1, characterized in that The method further comprises: If the expected interaction result and the simulation result both include multiple data types, corresponding weights are set for different data types, and the deviation is determined by weighted summation based on the weights corresponding to the various data types.

7. A multi-agent interaction simulation method, characterized in that: The method comprises: Determine several subjects in the target interaction scenario, and obtain attribute data corresponding to each of the several subjects; Determining behavioral interaction rules among the plurality of subjects, and an association between the behavioral interaction rules and an environmental state; The attribute data, the behavioral interaction rules and the association relationship are input into the multi-agent interaction simulation model in the method according to any one of claims 1 to 6, and the simulation results are output through the multi-agent interaction simulation model, wherein the simulation results are simulation data of the target association relationship changing over time, and the target association relationship is the relationship between the interaction behavior between the several subjects and the environmental state.

8. A computer program product comprising a computer program, wherein when the computer program is executed by a processor, the computer program implements the steps of the method according to any one of claims 1 to 7.

9. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to implement the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images