Multi-agent protection scene generation method and device, storage medium and computer device

Abstract

The multi-agent protection scene generation method and device, the storage medium and the computer device provided by the application generate a protection decision function corresponding to a first agent through a preset dynamic decision model, realize dynamic modeling of decision behaviors in a multi-agent protection scene, and then iteratively generate initial protection decisions of the first agent in different states based on the protection decision function, which makes it possible to cover diversified decision paths. On this basis, an interaction influence parameter is further introduced to correct the initial protection decisions in real time, so that the generated protection scene can more truly restore the complex interaction logic of the real world in the time sequence evolution process. Through the above iterative updating mechanism, the scheme can generate protection scene simulation data covering multiple protection scenes and having high authenticity, thereby providing more reliable and comprehensive data support for evaluation and optimization of multi-agent protection strategies.

Description

Technical Field

[0001] This application relates to the field of simulation technology, and in particular to a method, apparatus, storage medium and computer equipment for generating multi-agent protection scenarios. Background Technology

[0002] In the field of intelligent system simulation, the generation of multi-agent protection scenarios is of great significance for evaluating the decision-making ability of agents in dynamic environments and optimizing protection strategies. Taking agricultural ecological protection as an example, the combined effects of the interaction behavior between agents, dynamic environmental changes, and external incentive rules constitute a complex protection decision-making scenario.

[0003] Most existing protection scenario generation methods adopt static modeling or simple rule-driven approaches, which make it difficult to dynamically reflect the real-time interaction and influence between intelligent agents and the temporal evolution of decision-making behavior. They can usually only generate single, fixed protection scenarios, resulting in insufficient realism of the generated protection scenarios and low scenario coverage. Summary of the Invention

[0004] The purpose of this application is to at least address one of the aforementioned technical deficiencies, particularly the technical deficiency that existing protection scenario generation methods struggle to dynamically reflect the real-time interaction and influence between intelligent agents and the temporal evolution of decision-making behavior. They typically only generate single, fixed protection scenarios, resulting in insufficient realism and low scenario coverage.

[0005] Firstly, this application provides a method for generating a multi-agent protection scenario, the method comprising: Acquire the first initial data of the first intelligent agent, the second initial data of the second intelligent agent, and the incentive rules; The first initial data, the second initial data, and the incentive rule are input into a preset dynamic decision model to obtain the protection decision function corresponding to the first intelligent agent; The initial protection decision for the first agent in its current state is generated through the protection decision function. Based on the interaction influence parameters of the first intelligent agent, the initial protection decision is corrected to obtain the target protection decision; Based on the target protection decision, update the current state, and jump to execute the step of generating the initial protection decision of the first agent in the current state through the protection decision function until the iteration termination condition is met, generating multiple protection scenario simulation data.

[0006] In one embodiment, the step of inputting the first initial data, the second initial data, and the incentive rule into a preset dynamic decision model to obtain the protection decision function corresponding to the first agent includes: The discrete data in the first initial data is encoded into sequence features to form a risk preference vector, and basic information is extracted from the first initial data to generate a basic attribute vector. Based on the second initial data, predict the activity intensity of the second agent; The risk preference vector, the basic attribute vector, the activity intensity, and the incentive rule are used as the state vector of the first agent, and a knowledge-enhanced loss function is calculated. The state vector and the knowledge-enhanced loss function are input into a preset dynamic decision model to solve for the protection decision function that maximizes the benefit of the first agent under different states based on the dynamic programming algorithm.

[0007] In one embodiment, the computational knowledge-enhanced loss function includes: Acquire historical sample data and calculate the experience loss based on the historical sample data; Obtain domain knowledge rules and convert each domain knowledge rule into a penalty function to determine the knowledge penalty loss; By combining the experience loss and the knowledge penalty loss, a knowledge-enhanced loss function is obtained.

[0008] In one embodiment, the initial protection decision is modified according to the interaction influence parameters of the first agent, and the target protection decision is obtained by the following expression:

[0009] In the formula, This indicates the revised protection decision. Represents an individual in the first intelligent agent Susceptibility weighting This indicates the currently generated protection decision. Represents an individual in the first intelligent agent Interaction influence parameters, This represents a topological matrix describing the neighborhood or kinship relationships among the groups represented by the first agent. Represents an individual in the first intelligent agent For individuals The degree of acceptance of decision-making information This represents the probability of information propagation.

[0010] In one embodiment, the step of updating the current state based on the target protection decision and then jumping to execute the step of generating the initial protection decision of the first agent in the current state through the protection decision function continues until the iteration termination condition is met, generating multiple protection scenario simulation data, including: The target protection decision is executed in the interactive simulation environment formed by the first intelligent agent and the second intelligent agent, and the current state is reacquired after the target protection decision is executed; Based on the preset benefit function and the updated current state, the decision benefit of the target protection decision is calculated, and the probability of behavior change of the first agent is determined according to the decision benefit. The process jumps to the step of generating the initial protection decision of the first agent in the current state through the protection decision function until the iteration termination condition is met. Based on the data generated in each iteration, multiple protection scenario simulation data are generated.

[0011] In one embodiment, the method further includes: The behavior change probability of the first agent is extracted from the simulation data of each protection scenario to form a behavior change probability sequence; Obtain the behavior prediction model and input the behavior change probability sequence into the behavior prediction model to output the behavior change probability corresponding to each individual in the group represented by the first agent.

[0012] In one embodiment, the method further includes: In response to the incentive rule optimization instruction, the behavior change probability of the first agent is extracted from the simulation data of each protection scenario; With minimizing the probability of behavioral change of the first agent as the optimization objective, the contribution of the incentive rule to the optimization objective is evaluated; The incentive rules are optimized based on the contribution level to obtain the optimized incentive rules.

[0013] Secondly, this application provides a multi-agent protection scene generation device, the device comprising: The data acquisition module is used to acquire the first initial data of the first intelligent agent, the second initial data of the second intelligent agent, and the incentive rules. The decision determination module is used to input the first initial data, the second initial data, and the incentive rules into a preset dynamic decision model to obtain the protection decision function corresponding to the first intelligent agent; The decision generation module is used to generate the initial protection decision corresponding to the first agent in the current state through the protection decision function; The decision correction module is used to correct the initial protection decision based on the interaction influence parameters of the first intelligent agent to obtain the target protection decision; The scenario simulation module is used to update the current state according to the target protection decision, jump to execute the step of generating the initial protection decision of the first agent in the current state through the protection decision function, until the iteration termination condition is met, and generate multiple protection scenario simulation data.

[0014] Thirdly, this application provides a storage medium storing computer-readable instructions, which, when executed by one or more processors, cause the one or more processors to perform the steps of the multi-agent protection scenario generation method as described in any of the above embodiments.

[0015] Fourthly, this application provides a computer device, including: one or more processors, and a memory; The memory stores computer-readable instructions, and when the one or more processors execute the computer-readable instructions, they perform the steps of the multi-agent protection scenario generation method as described in any of the above embodiments.

[0016] As can be seen from the above technical solutions, the embodiments of this application have the following advantages: The multi-agent protection scenario generation method, apparatus, storage medium, and computer equipment provided in this application acquire first initial data of the first agent, second initial data of the second agent, and incentive rules. Multi-dimensional agent data and constraint rules are input into a preset dynamic decision model to obtain the protection decision function corresponding to the first agent, thus achieving dynamic modeling of decision-making behavior under multi-agent protection scenarios. Subsequently, based on the protection decision function, initial protection decisions for the first agent under different states are iteratively generated, effectively overcoming the limitation of traditional methods that can only generate a single fixed scenario, and providing the possibility of covering diverse decision paths. Furthermore, interaction influence parameters are introduced to correct the initial protection decisions in real time, dynamically responding to the regulatory effects of information transmission and social learning among agents on the decision-making process. This allows the generated protection scenario to more realistically reproduce the complex interaction logic of the real world during its temporal evolution. Through the above iterative update mechanism, this scheme can generate highly realistic protection scenario simulation data covering multiple protection scenarios, thereby providing more reliable and comprehensive data support for the evaluation and optimization of multi-agent protection strategies. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 A flowchart illustrating a method for generating a multi-agent protection scenario provided in an embodiment of this application; Figure 2 A flowchart illustrating how first initial data, second initial data, and incentive rules are input into a preset dynamic decision model to obtain the protection decision function corresponding to the first intelligent agent, as provided in this embodiment of the application. Figure 3 This is a schematic diagram of the structure of a multi-agent protection scene generation device provided in an embodiment of this application; Figure 4 This is an internal structural diagram of a computer device provided in an embodiment of this application. Detailed Implementation

[0019] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0020] In agricultural ecosystems, human-wildlife conflict is becoming increasingly prominent, particularly the disruption and losses caused by wildlife activity to farmers' production and lives, which has become a significant factor hindering regional sustainable development. To alleviate this problem, relevant departments typically implement incentive measures to guide farmers to adopt reasonable biosafety practices, aiming to reduce farmers' losses while protecting wildlife and maintaining farmland conditions. However, existing modeling and simulation methods for farmers' biosafety practices mostly rely on traditional statistical models, which usually only generate single, fixed biosafety scenarios, resulting in insufficient realism and low scenario coverage.

[0021] Taking the above application scenario as an example, this application provides a method for generating multi-agent protection scenarios. The following embodiments illustrate the application of this method to a multi-agent interaction system, hereinafter referred to as the system. It is understood that this system can be deployed in a computer device to execute the multi-agent protection scenario generation method provided in this application. Based on this, as... Figure 1 As shown, this application provides a method for generating a multi-agent protection scenario, the method comprising: S101: Obtain the first initial data of the first agent, the second initial data of the second agent, and the incentive rules.

[0022] The first intelligent agent refers to a computational unit with environmental perception and autonomous decision-making capabilities, abstracted from the group of farmers implementing protection decisions. The second intelligent agent refers to a computational unit with environmental perception and autonomous decision-making capabilities, abstracted from the related subjects affecting the achievement of protection goals and protection effectiveness. Furthermore, there is a two-way interaction and game-theoretic relationship between the first and second intelligent agents.

[0023] In this step, when generating a protection scenario, initial and secondary initial data are first obtained through methods such as field surveys, remote sensing detection, sensor networks, or historical statistical data. Simultaneously, incentive rules are established based on local policy documents. Specifically, the initial data refers to the dataset characterizing the basic attributes and risk preferences of the first agent, including but not limited to farmer land size, planting structure, risk preference questionnaire information, records of wildlife-related losses, and insurance information. The secondary initial data refers to the dataset characterizing the activity distribution of the second agent, including but not limited to spatial distribution trajectories, activity frequency, and seasonal fluctuation data. Incentive rules refer to external intervention mechanisms used to guide farmers to take specific protective actions, such as protection subsidy standards, loss compensation ratios, and payment delay times.

[0024] Furthermore, the first and second intelligent agents can be constructed using a simulation platform based on intelligent agent modeling. The core of constructing the first intelligent agent lies in simulating the cognitive and decision-making processes of farmers, while the core of constructing the second intelligent agent lies in simulating the spatial behavior and ecological laws of organisms.

[0025] For example, taking a multi-agent protection scenario for farmland wildlife damage as an example, the system first defines the first agent as the group of farmers who bear the responsibility of farmland protection and face the risk of wildlife damage, and the second agent as the wildlife population that causes damage to farmland crops and interacts with the farmers' protection actions. Based on this definition of the agents, the system collects basic data corresponding to the farmer group as the first initial data for the first agent. This data may include the farmers' basic family information, the area and geographical location of their contracted farmland, the scale of past protection efforts, their willingness to protect and their risk tolerance, and the types and growth cycles of farmland crops. Simultaneously, the system collects basic data corresponding to the wildlife population as the second initial data for the second agent. This data may include the species type, population size, core habitat and activity range, seasonal migration and activity patterns, and the frequency of past farmland damage. Furthermore, the system obtains the incentive rules formulated by local agricultural and forestry authorities for farmers' wildlife protection actions.

[0026] S102: Input the first initial data, the second initial data, and the incentive rules into the preset dynamic decision model to obtain the protection decision function corresponding to the first intelligent agent.

[0027] The dynamic decision model refers to an intelligent computing framework that integrates multi-source heterogeneous data and embeds domain knowledge to characterize the decision-making complexity of the first agent. It learns the decision-making patterns of a farmer group in a dynamic environment from initial data, second initial data, and incentive rules through the constraints and optimization of a knowledge-enhanced loss function. The defensive decision function represents the mapping relationship between the state of the first agent and its defensive decisions.

[0028] In this step, when the first initial data, the second initial data, and the incentive rules are obtained, these three types of data can be preprocessed. The preprocessed data is then input into a preset dynamic decision model. This model can calculate the benefit function of the first agent based on the input data. By combining various types of information and the benefit function, it dynamically fits the generation logic of the farmer's protection decision that maximizes the benefit under different states, thus obtaining the protection decision function corresponding to the first agent.

[0029] Furthermore, the state definition of the first intelligent agent is derived from the first initial data, the second initial data, and the incentive rules. Specifically, by inputting multi-dimensional initial data and incentive rules into the dynamic decision model, a protection decision function is generated. This fully integrates the core information of the scenario and leverages the dynamic modeling advantages of the dynamic decision model to transform fragmented and diverse input data into a reusable protection decision function. This enables the first intelligent agent to make adaptive protection responses that conform to domain knowledge based on its own state, environmental risks, and incentive rules, thereby providing decision-making dependencies for the subsequent generation of various protection scenarios.

[0030] S103: Generate the initial protection decision for the first agent in the current state through the protection decision function.

[0031] Here, the current state refers to the current state vector of the first agent, which includes the basic attribute vector of the first agent, the risk preference vector, the activity intensity of the second agent, and the incentive rules. The initial protection decision refers to the protection behavior decision generated by the first agent based solely on its current state vector by calling the protection decision function, which includes, but is not limited to, information such as the level of protection investment, whether to purchase insurance, and specific protection measures.

[0032] In this step, the relevant information is first obtained according to the predefined composition fields of the state vector to construct the current state vector of the first agent. Then, this state vector is input into the protection decision function, and calculations and inferences are performed according to the built-in mapping relationship and decision logic of the function to obtain the initial protection decision corresponding to the first agent in the current state.

[0033] For example, in the scenario of protecting farmland wildlife, the current state of the first agent can be reflected in the basic attributes, risk preferences, wildlife activity, crop growth status and incentive rules of the farmer group at a certain moment. After inputting this state into the protection decision function of the first agent, the corresponding initial protection decision can be generated, such as whether to take protection measures and what kind of protection method to adopt.

[0034] S104: Based on the interaction influence parameters of the first intelligent agent, the initial protection decision is corrected to obtain the target protection decision.

[0035] Among them, the interaction influence parameter refers to an indicator used to quantify the degree of influence of decisions made by other individuals or the social environment on the first agent in a social network. Target protection decision refers to the protective behavior decision generated by the first agent after comprehensively considering the current state and the social interaction influence.

[0036] In this step, the interaction influence parameters of each individual in the group represented by the first agent can be obtained first. Based on the interaction influence parameters of each individual, the interaction influence parameters of the first agent are formed, and the initial protection decision is modified accordingly to obtain the modified initial protection decision, i.e., the target protection decision.

[0037] For example, in the scenario of protecting farmland from wildlife damage, assuming that the initial protection decision of the first agent is to set up a physical fence alone, the interaction influence parameters are reflected in the protection behavior of surrounding farmers (such as whether neighboring farmers jointly protect the farmland and the type of protection facilities), the willingness of farmers to cooperate in protection, etc. The initial decision is corrected based on the interaction influence parameters. For example, if the surrounding farmers all choose to jointly build a protective net, the initial decision to set up a physical fence alone may be corrected to jointly build a protective net with the surrounding farmers and add a target protection decision of infrared monitoring equipment, thereby improving the protection effect.

[0038] It is understandable that since the degree of influence of decisions on individuals in their social networks generally does not change significantly in the short term, the interaction influence parameters of each individual can be pre-calculated and stored, and then directly retrieved and reused when needed.

[0039] S105: Based on the target protection decision, update the current state and determine whether the iteration termination condition is met.

[0040] The iteration termination condition refers to the termination condition of the cyclic simulation based on the protection decision function. It can be set to reach a preset number of iterations or to the scenario evolution tending to stabilize. This application does not impose specific restrictions on this.

[0041] In this step, a target protection decision is executed within the simulation environment formed by the first and second agents. Driven by this target protection decision, the basic attributes of the first agent, the strength feedback of the second agent, and the incentive rules involved in the current state are adjusted, thereby generating a new current state corresponding to the first agent. It can be understood that by using the target protection decision as the trigger for change, and following the logical correlation and temporal continuity of state evolution, the updated current state can accurately reflect the scene changes after the decision is executed.

[0042] S106: If not satisfied, then proceed to the step of generating the initial protection decision of the first agent in the current state through the protection decision function.

[0043] S107: If satisfied, generate multiple protection scenario simulation data.

[0044] Among them, the protection scenario simulation data refers to the dataset used to describe the interaction between the first agent and the second agent in the corresponding protection scenario. Each protection scenario simulation data corresponds to a protection scenario, and the protection scenario simulation data includes, but is not limited to, the state of the first agent, the target protection decision, and the protection benefits.

[0045] In this step, when the iteration termination condition is met, the various data from each simulation iteration can be integrated and summarized to obtain the protection scenario simulation data corresponding to each simulation iteration. It can be understood that when multiple protection scenario simulation data are generated, the evolution of each protection scenario can be visualized based on each simulation data, in order to evaluate the long-term evolution trend and protection effect of farmers' protection behavior under different incentive rules and different environments.

[0046] In the above embodiments, by acquiring the first initial data of the first agent, the second initial data of the second agent, and incentive rules, multi-dimensional agent data and constraint rules are input into a preset dynamic decision model to obtain the protection decision function corresponding to the first agent, thus realizing dynamic modeling of decision-making behavior in multi-agent protection scenarios. Subsequently, based on the protection decision function, initial protection decisions for the first agent under different states are iteratively generated, effectively overcoming the limitation of traditional methods that can only generate a single fixed scenario, and providing the possibility of covering diverse decision paths. Furthermore, interaction influence parameters are introduced to correct the initial protection decisions in real time, dynamically responding to the regulatory effect of interactive behaviors such as information transmission and social learning between agents on the decision, enabling the generated protection scenario to more realistically reproduce the complex interactive logic of the real world during temporal evolution. Through the above iterative update mechanism, this scheme can generate protection scenario simulation data covering multiple protection scenarios with high realism, thereby providing more reliable and comprehensive data support for the evaluation and optimization of multi-agent protection strategies.

[0047] like Figure 2 As shown, in one embodiment, the first initial data, the second initial data, and the incentive rules are input into a preset dynamic decision model to obtain the protection decision function corresponding to the first intelligent agent, including: S201: Encode the discrete data in the first initial data into sequence features to form a risk preference vector, and extract basic information from the first initial data to generate a basic attribute vector.

[0048] S202: Based on the second initial data, predict the activity intensity of the second agent.

[0049] S203: Use the risk preference vector, basic attribute vector, activity intensity, and incentive rules as the state vector of the first agent, and calculate the knowledge-enhanced loss function.

[0050] S204: Input the state vector and knowledge-enhanced loss function into the preset dynamic decision model, and solve the protection decision function that maximizes the benefit of the first agent under different states based on the dynamic programming algorithm.

[0051] Among them, the risk preference vector is used to quantify the behavioral tendencies of the first agent when facing risks. The basic attribute vector is used to quantify the static characteristics of the first agent, such as age, cultivated land area, and household income. The activity intensity is used to quantify the activity level of the second agent within a specific spatiotemporal range, which may pose a threat to farmland. The knowledge-enhanced loss function is a composite function formed by incorporating domain knowledge into the experience loss function.

[0052] In this embodiment, discrete features, such as risk preference questionnaires, can first be extracted from the first initial data and converted into sequential features using methods such as one-hot encoding, sequence embedding, or discrete feature mapping. These sequential features are then subjected to dimensionality reduction by an encoder to obtain a risk preference vector that expresses the risk acceptance level of the first agent. Simultaneously, static features that do not change with decision-making are selected from the first initial data and standardized and vectorized to generate the basic attribute vector of the first agent. Then, using historical spatiotemporal distributions and environmental information included in the second initial data, the activity intensity data of the second agent over a period of time can be calculated through regression prediction or time-series prediction.

[0053] Subsequently, the risk preference vector, basic attribute vector, activity intensity, and incentive rules are concatenated and dimensionally aligned. Different combinations generate various state vectors, representing the different states of the first agent. Based on this, and incorporating domain prior knowledge, a knowledge-enhanced loss function including a knowledge regularization term is calculated. Finally, the constructed state vectors and the knowledge-enhanced loss function are input into a pre-built dynamic decision model. Dynamic programming is used to iteratively optimize and seek the best solution for this model, deriving a protective decision function that maximizes the agent's gains in different states. This protective decision function outputs matching protective decisions based on different states.

[0054] It is understandable that by embedding domain knowledge, the decision-making results that may be contrary to common sense by a purely data-driven model can be effectively avoided. By combining dynamic programming algorithm to solve for the optimal benefit, the dynamic decision-making model can take into account both data patterns and real-world logic during the learning process, and finally generate a protective decision function that is more in line with the actual scenario and more in line with the main benefit goal.

[0055] In one embodiment, calculating the knowledge-enhanced loss function includes: S1: Obtain historical sample data and calculate the experience loss based on the historical sample data.

[0056] S2: Obtain domain knowledge rules and convert each domain knowledge rule into a penalty function to determine the knowledge penalty loss.

[0057] S3: Integrate experience loss and knowledge penalty loss to obtain a knowledge-enhancing loss function.

[0058] Historical sample data refers to the set of sample data consisting of the past states of the first intelligent agent, target protection decisions, and actual execution decisions. Experience loss refers to the fitting error of the dynamic decision-making model to the historical sample data. Domain knowledge rules refer to the prior common sense, behavioral constraints, normative logic, and evolutionary laws extracted from the professional domain to which the protection scenario belongs, verified through practice, used to constrain the model's decisions to conform to realistic rationality. The penalty function is a numerically calculable mathematical mapping relationship transformed from specified domain knowledge rules. Knowledge penalty loss is the total loss term calculated by integrating the penalty functions corresponding to all domain knowledge rules.

[0059] In this embodiment, historical sample data from past protection scenarios are collected and organized. This data includes historical state information, corresponding protection decisions, and actual execution results. The predicted protection decisions are compared with the actual decisions, and an empirical loss is obtained through error calculation to measure the model's fitting accuracy and decision bias at the data level. Subsequently, validated domain knowledge rules are extracted from protection-related professional fields. Each domain knowledge rule is numerically mapped and transformed into a computable penalty function. All penalty functions are aggregated to obtain the knowledge penalty loss, which characterizes the degree of deviation between the protection decision and domain common sense. Finally, the empirical loss and the knowledge penalty loss are fused according to set weights to form a knowledge-enhanced loss function.

[0060] For example, the fused knowledge-enhanced loss function can be expressed as:

[0061] In the formula, This represents the knowledge-enhanced loss function. Indicates loss of experience. This indicates a knowledge penalty loss. This represents the weight parameter, which is an empirical value.

[0062] Specifically, the model first calculates the experience loss based on historical sample data to ensure that the model's decisions are consistent with historical patterns. Then, it constructs a knowledge penalty loss through domain knowledge rules to make the decisions conform to real-world constraints and professional logic. The knowledge-enhanced loss function obtained by integrating the two can not only improve the model's fitting accuracy and generalization ability, but also effectively avoid generating protective decisions that violate common sense, are unenforceable, or are unreasonable.

[0063] In one embodiment, the initial protection decision is modified according to the interaction influence parameters of the first intelligent agent, and the target protection decision is obtained by the following expression:

[0064] In the formula, This indicates the revised protection decision. Represents an individual in the first intelligent agent Susceptibility weighting This indicates the currently generated protection decision. Represents an individual in the first intelligent agent Interaction influence parameters, This represents a topological matrix describing the neighborhood or kinship relationships among the groups represented by the first agent. Represents an individual in the first intelligent agent For individuals The degree of acceptance of decision-making information This represents the probability of information propagation.

[0065] in, , Represents an individual The basic attribute vector, Represents an individual The risk preference vector, This represents the activity intensity of the second agent at time t. This represents the incentive rule; in one instance, the incentive rule... S represents the subsidy intensity, and I represents the insurance payout ratio. For mixed policy weights.

[0066] Specifically, the interaction effect parameter can be calculated using the following expression:

[0067] In the formula, Represents an individual in the topological matrix With individuals Connection strength between them Represents the relationship between individuals in the topological matrix. A collection of individuals who are neighbors or related by kinship. Represents an individual Decision information.

[0068] In one embodiment, based on the target protection decision, the current state is updated, and the process jumps to the step of generating the initial protection decision of the first agent in the current state through the protection decision function, until the iteration termination condition is met, generating multiple protection scenario simulation data, including: S1: Execute target protection decisions in the interactive simulation environment formed by the first and second intelligent agents, and reacquire the current state after executing the target protection decisions.

[0069] S2: Based on the preset benefit function and the updated current state, calculate the decision benefit of the target protection decision, and determine the probability of behavior change of the first agent based on the decision benefit.

[0070] S3: Jump to execute the step of generating the initial protection decision of the first agent in the current state through the protection decision function until the iteration termination condition is met. Based on the data generated in each iteration, generate multiple protection scenario simulation data.

[0071] Among them, the benefit function is used to quantify the execution effect of the target protection decision, and the behavior change probability is used to characterize the probability of the first agent making a behavior change based on the decision benefit in the current protection scenario, such as the probability of abandoning farmland.

[0072] In this embodiment, within an interactive simulation environment comprised of a first and a second intelligent agent, the currently obtained target protection decision is first implemented in the environment to simulate the effect of the decision. After the decision is executed, information such as the state of the first intelligent agent and the activity intensity feedback of the second intelligent agent is re-collected and integrated to obtain an updated current state. Then, based on this updated current state, a preset benefit function is substituted to calculate the decision benefit corresponding to this target protection decision. Based on this decision benefit, the probability of subsequent behavioral changes by the first intelligent agent in the current protection scenario is determined according to the magnitude of the decision benefit, i.e., the probability of behavioral change. Then, the process jumps back to the step of generating the initial protection decision for the first intelligent agent in the current state through the protection decision function, and continues iteratively until the iteration termination condition is met. At this point, the decision data, state data, benefit data, and behavioral change data generated in each iteration are organized and structured, ultimately forming multiple protection scenario simulation data. Furthermore, the benefit function can be set by comprehensively considering data such as basic agricultural income, policy subsidies, and accident compensation ratios.

[0073] Specifically, in a dual-agent interactive simulation environment, target protection decisions are executed and the current state is updated. The decision benefits are calculated by combining the benefit function and the probability of behavior change is determined, making the entire iterative process more in line with the real decision-making logic of the agents and the evolution law of the scenario. After the iteration termination condition is met, multiple protection scenario simulation data are generated based on the full round data. This protection scenario simulation data can cover a variety of protection scenarios and has high realism, thus providing more reliable and comprehensive data support for the evaluation and optimization of multi-agent protection strategies.

[0074] In one embodiment, the multi-agent protection scenario generation method further includes: S1: Extract the behavior change probability of the first agent from the simulation data of each protection scenario to form a behavior change probability sequence.

[0075] S2: Obtain the behavior prediction model and input the behavior change probability sequence into the behavior prediction model to output the behavior change probability of each individual in the group represented by the first agent.

[0076] Among them, the behavior prediction model is a binary classification model used for predicting individual behavior.

[0077] In this embodiment, the behavior change probability of the first agent is extracted from the simulation data of each protection scenario, and then a sequence is formed, namely the behavior change probability sequence. This behavior change probability sequence is then input into a pre-trained behavior prediction model to obtain the behavior change probability for each individual in the group represented by the first agent. Furthermore, machine learning algorithms, such as random forests and XGBoost, can be used to train the behavior prediction model.

[0078] Specifically, behavior change probabilities are extracted from multiple sets of protection scenario simulation data and formed into time series, which realizes the capture of the behavior change pattern of the first intelligent agent group. Then, the group time series probability is transformed into individual behavior change probability by using a behavior prediction model. This not only makes up for the shortcomings of traditional simulation that can only obtain group-level probabilities and cannot accurately characterize individual behavior tendencies, but also improves the accuracy of individual behavior change probability prediction through time series analysis.

[0079] In one embodiment, the multi-agent protection scenario generation method further includes: S1: In response to the incentive rule optimization instruction, extract the behavior change probability of the first agent from the simulation data of each protection scenario.

[0080] S2: With minimizing the probability of behavioral change of the first agent as the optimization objective, evaluate the contribution of the incentive rule to the optimization objective.

[0081] S3: Optimize the incentive rules based on contribution to obtain the optimized incentive rules.

[0082] In this embodiment, after receiving the incentive rule optimization instruction, the simulation data of each protection scenario can be traversed and parsed. The behavior change probability of the corresponding first agent in each round of evolution can be extracted according to the scenario dimension. The quantitative collection of the behavior stability characteristics of the first agent under different incentive rules is completed. Then, with the goal of minimizing the behavior change probability of the first agent, the behavior change probability corresponding to different incentive rule values, combinations and triggering conditions is compared and analyzed one by one. The contribution of each incentive rule to the optimization goal is evaluated by correlation calculation, contribution rate ranking and other methods. High contribution, low contribution and even negative impact rule clauses are distinguished. On this basis, positive incentive rules with high contribution are strengthened and retained, rules with low contribution are adjusted, and rules with negative impact are eliminated or reconstructed, so as to obtain the optimized incentive rules, which provide a more scientific and reasonable rule basis for subsequent scenario simulation and decision inference.

[0083] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0084] The multi-agent protection scene generation device provided in the embodiments of this application is described below. The multi-agent protection scene generation device described below and the multi-agent protection scene generation method described above can be referred to in correspondence.

[0085] like Figure 3 As shown, this application provides a multi-agent protection scene generation device 300, the device comprising: Data acquisition module 301 is used to acquire the first initial data of the first intelligent agent, the second initial data of the second intelligent agent, and the incentive rules; The decision determination module 302 is used to input the first initial data, the second initial data and the incentive rules into the preset dynamic decision model to obtain the protection decision function corresponding to the first intelligent agent; Decision generation module 303 is used to generate the initial protection decision corresponding to the first agent in the current state through the protection decision function; The decision correction module 304 is used to correct the initial protection decision based on the interaction influence parameters of the first intelligent agent to obtain the target protection decision; The scenario simulation module 305 is used to update the current state according to the target protection decision, jump to execute the step of generating the initial protection decision of the first agent in the current state through the protection decision function, until the iteration termination condition is met, and generate multiple protection scenario simulation data.

[0086] In the above embodiments, by acquiring the first initial data of the first agent, the second initial data of the second agent, and incentive rules, multi-dimensional agent data and constraint rules are input into a preset dynamic decision model to obtain the protection decision function corresponding to the first agent, thus realizing dynamic modeling of decision-making behavior in multi-agent protection scenarios. Subsequently, based on the protection decision function, initial protection decisions for the first agent under different states are iteratively generated, effectively overcoming the limitation of traditional methods that can only generate a single fixed scenario, and providing the possibility of covering diverse decision paths. Furthermore, interaction influence parameters are introduced to correct the initial protection decisions in real time, dynamically responding to the regulatory effect of interactive behaviors such as information transmission and social learning between agents on the decision, enabling the generated protection scenario to more realistically reproduce the complex interactive logic of the real world during temporal evolution. Through the above iterative update mechanism, this scheme can generate protection scenario simulation data covering multiple protection scenarios with high realism, thereby providing more reliable and comprehensive data support for the evaluation and optimization of multi-agent protection strategies.

[0087] In one embodiment, the decision determination module includes: The vector generation submodule is used to encode the discrete data in the first initial data into sequence features to form a risk preference vector, and to extract basic information from the first initial data to generate a basic attribute vector. The intensity prediction submodule is used to predict the activity intensity of the second agent based on the second initial data; The loss calculation submodule is used to calculate the knowledge-enhanced loss function using the risk preference vector, basic attribute vector, activity intensity, and incentive rules as the state vector of the first agent. The function solving submodule is used to input the state vector and knowledge-enhanced loss function into a preset dynamic decision model, so as to solve the protection decision function that maximizes the benefit of the first agent under different states based on the dynamic programming algorithm.

[0088] In one embodiment, the loss calculation submodule includes: The first computing unit is used to acquire historical sample data and calculate the experience loss based on the historical sample data. The second computational unit is used to acquire domain knowledge rules and convert each domain knowledge rule into a penalty function to determine the knowledge penalty loss. The loss fusion unit is used to fuse experience loss and knowledge penalty loss to obtain a knowledge-enhanced loss function.

[0089] In one embodiment, the decision correction module includes:

[0090] In the formula, This indicates the revised protection decision. Represents an individual in the first intelligent agent Susceptibility weighting This indicates the currently generated protection decision. Represents an individual in the first intelligent agent Interaction influence parameters, This represents a topological matrix describing the neighborhood or kinship relationships among the groups represented by the first agent. Represents an individual in the first intelligent agent For individuals The degree of acceptance of decision-making information This represents the probability of information propagation.

[0091] In one embodiment, the scene simulation module includes: The state update submodule is used to execute target protection decisions in the interactive simulation environment formed by the first agent and the second agent, and to reacquire the current state after executing the target protection decisions; The probability determination submodule is used to calculate the decision benefit of target protection decision based on the preset benefit function and the updated current state, and to determine the probability of behavior change of the first intelligent agent based on the decision benefit. The iterative simulation submodule is used to jump to execute the step of generating the initial protection decision of the first agent in the current state through the protection decision function until the iteration termination condition is met. Based on the data generated in each iteration, it generates multiple protection scenario simulation data.

[0092] In one embodiment, the multi-agent protection scenario generation device further includes: The first acquisition module is used to extract the behavior change probability of the first intelligent agent from the simulation data of each protection scenario, and form a behavior change probability sequence. The probability prediction module is used to acquire the behavior prediction model and input the behavior change probability sequence into the behavior prediction model to output the behavior change probability of each individual in the group represented by the first agent.

[0093] In one embodiment, the multi-agent protection scenario generation device further includes: The second acquisition module is used to extract the behavior change probability of the first agent from the simulation data of each protection scenario in response to the incentive rule optimization instruction. The contribution evaluation module is used to evaluate the contribution of the incentive rules to the optimization objective, which is to minimize the probability of behavioral change of the first agent. The rule optimization module is used to optimize the incentive rules based on contribution, resulting in optimized incentive rules.

[0094] The division of modules in the above-described multi-agent protection scenario generation device is merely illustrative. In other embodiments, the multi-agent protection scenario generation device can be divided into different modules as needed to complete all or part of its functions. Each module in the above-described multi-agent protection scenario generation device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0095] In one embodiment, this application also provides a storage medium storing computer-readable instructions that, when executed by one or more processors, cause the one or more processors to perform the steps of the multi-agent protection scenario generation method as described in any of the above embodiments.

[0096] In one embodiment, this application also provides a computer device storing computer-readable instructions that, when executed by one or more processors, cause the one or more processors to perform the steps of the multi-agent protection scenario generation method as described in any of the above embodiments.

[0097] Indicatively, such as Figure 4 As shown, Figure 4 This is a schematic diagram of the internal structure of a computer device 400 provided in an embodiment of this application. The computer device 400 can be provided as a server. (Refer to...) Figure 4 The computer device 400 includes a processing component 402, which further includes one or more processors, and memory resources represented by memory 401 for storing instructions, such as application programs, that can be executed by the processing component 402. The application programs stored in memory 401 may include one or more modules, each corresponding to a set of instructions. Furthermore, the processing component 402 is configured to execute instructions to perform the multi-agent protection scene generation method of any of the above embodiments.

[0098] The computer device 400 may also include a power supply component 403 configured to perform power management of the computer device 400, a wired or wireless network interface 404 configured to connect the computer device 400 to a network, and an input / output (I / O) interface 405. The computer device 400 may operate on an operating system stored in memory 401, such as Windows Server™, Mac OS X™, Unix™, Linux™, Free BSD™, or similar.

[0099] Those skilled in the art will understand that Figure 4 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0100] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising a…" does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element. In this document, the singular forms "a," "an," and "the" may also include the plural forms unless the context clearly indicates otherwise. It should also be understood that the terms “comprising / including” or “having” specify the presence of the stated features, wholes, steps, operations, components, parts or combinations thereof, but do not exclude the possibility of the presence or addition of one or more other features, wholes, steps, operations, components, parts or combinations thereof. Meanwhile, the term “and / or” as used in this specification includes any and all combinations of the associated listed items.

[0101] The various embodiments in this specification are described in a progressive manner. Each embodiment focuses on the differences from other embodiments. The various embodiments can be combined as needed, and the same or similar parts can be referred to each other.

[0102] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for generating a multi-agent protection scenario, characterized in that, The method includes: Acquire the first initial data of the first intelligent agent, the second initial data of the second intelligent agent, and the incentive rules; The first initial data, the second initial data, and the incentive rule are input into a preset dynamic decision model to obtain the protection decision function corresponding to the first intelligent agent; The initial protection decision for the first agent in its current state is generated through the protection decision function. Based on the interaction influence parameters of the first intelligent agent, the initial protection decision is corrected to obtain the target protection decision; Based on the target protection decision, update the current state, and jump to execute the step of generating the initial protection decision of the first agent in the current state through the protection decision function until the iteration termination condition is met, generating multiple protection scenario simulation data.

2. The multi-agent protection scene generation method according to claim 1, characterized in that, The step of inputting the first initial data, the second initial data, and the incentive rule into a preset dynamic decision model to obtain the protection decision function corresponding to the first agent includes: The discrete data in the first initial data is encoded into sequence features to form a risk preference vector, and basic information is extracted from the first initial data to generate a basic attribute vector. Based on the second initial data, predict the activity intensity of the second agent; The risk preference vector, the basic attribute vector, the activity intensity, and the incentive rule are used as the state vector of the first agent, and a knowledge-enhanced loss function is calculated. The state vector and the knowledge-enhanced loss function are input into a preset dynamic decision model to solve for the protection decision function that maximizes the benefit of the first agent under different states based on the dynamic programming algorithm.

3. The multi-agent protection scene generation method according to claim 2, characterized in that, The computational knowledge-enhanced loss function includes: Acquire historical sample data and calculate the experience loss based on the historical sample data; Obtain domain knowledge rules and convert each domain knowledge rule into a penalty function to determine the knowledge penalty loss; By combining the experience loss and the knowledge penalty loss, a knowledge-enhanced loss function is obtained.

4. The multi-agent protection scene generation method according to claim 1, characterized in that, Based on the interaction influence parameters of the first intelligent agent, the initial protection decision is corrected according to the following expression to obtain the target protection decision: In the formula, This indicates the revised protection decision. Represents an individual in the first intelligent agent Susceptibility weighting This indicates the currently generated protection decision. Represents an individual in the first intelligent agent Interaction influence parameters, This represents a topological matrix describing the neighborhood or kinship relationships among the groups represented by the first agent. Represents an individual in the first intelligent agent For individuals The degree of acceptance of decision-making information This represents the probability of information propagation.

5. The multi-agent protection scene generation method according to claim 1, characterized in that, The step of updating the current state based on the target protection decision, and then proceeding to the step of generating the initial protection decision of the first agent in the current state through the protection decision function, continues until the iteration termination condition is met, generating multiple protection scenario simulation data, including: The target protection decision is executed in the interactive simulation environment formed by the first intelligent agent and the second intelligent agent, and the current state is reacquired after the target protection decision is executed; Based on the preset benefit function and the updated current state, the decision benefit of the target protection decision is calculated, and the probability of behavior change of the first agent is determined according to the decision benefit. The process jumps to the step of generating the initial protection decision of the first agent in the current state through the protection decision function until the iteration termination condition is met. Based on the data generated in each iteration, multiple protection scenario simulation data are generated.

6. The multi-agent protection scene generation method according to any one of claims 1 to 5, characterized in that, The method further includes: The behavior change probability of the first agent is extracted from the simulation data of each protection scenario to form a behavior change probability sequence; Obtain the behavior prediction model and input the behavior change probability sequence into the behavior prediction model to output the behavior change probability corresponding to each individual in the group represented by the first agent.

7. The multi-agent protection scene generation method according to any one of claims 1 to 5, characterized in that, The method further includes: In response to the incentive rule optimization instruction, the behavior change probability of the first agent is extracted from the simulation data of each protection scenario; With minimizing the probability of behavioral change of the first agent as the optimization objective, the contribution of the incentive rule to the optimization objective is evaluated; The incentive rules are optimized based on the contribution level to obtain the optimized incentive rules.

8. A multi-agent protection scene generation device, characterized in that, The device includes: The data acquisition module is used to acquire the first initial data of the first intelligent agent, the second initial data of the second intelligent agent, and the incentive rules. The decision determination module is used to input the first initial data, the second initial data, and the incentive rules into a preset dynamic decision model to obtain the protection decision function corresponding to the first intelligent agent; The decision generation module is used to generate the initial protection decision corresponding to the first agent in the current state through the protection decision function; The decision correction module is used to correct the initial protection decision based on the interaction influence parameters of the first intelligent agent to obtain the target protection decision; The scenario simulation module is used to update the current state according to the target protection decision, jump to execute the step of generating the initial protection decision of the first agent in the current state through the protection decision function, until the iteration termination condition is met, and generate multiple protection scenario simulation data.

9. A storage medium, characterized in that: The storage medium stores computer-readable instructions, which, when executed by one or more processors, cause the one or more processors to perform the steps of the multi-agent protection scenario generation method as described in any one of claims 1 to 7.

10. A computer device, characterized in that, include: One or more processors, and memory; The memory stores computer-readable instructions, which, when executed by the one or more processors, perform the steps of the multi-agent protection scenario generation method as described in any one of claims 1 to 7.

Images