Cross-border data transmission simulation method and system based on multi-agent grid

Abstract

The application relates to a cross-border data transmission simulation method and system based on a multi-agent grid, which comprises the following steps: collecting the preference state of each regulatory agent and constructing an initial preference state set in the form of a descending arrangement vector; performing logical conflict detection on the initial preference state set to generate an inconsistency flag and an inconsistent agent pair set; when there is an inconsistent agent pair, calculating the preference concession interval of each conflict agent based on historical negotiation data and generating negotiation mediation parameters; guiding each conflict agent to iteratively exchange and modify the priority ranking within the respective concession interval based on the negotiation mediation parameters, constantly updating the negotiation mediation parameters according to the feedback deviation vector, until the preference modification deviation degree converges to below a preset acceptance threshold, and finally obtaining a final preference modification result; synchronizing the final preference modification result to the global preference configuration space, and performing cross-border data transmission simulation decision-making based on the updated global preference state.

Description

Technical Field

[0001] This invention relates to the field of multi-agent simulation technology, specifically to a cross-border data transmission simulation method and system based on a multi-agent mesh. Background Technology

[0002] In the field of cross-border data transmission simulation technology using multi-agent meshes, simulation systems typically need to coordinate the decision-making behaviors of multiple regulatory agents from different sovereign jurisdictions. Each regulatory agent holds its own differentiated preference ranking for multiple objective indicators such as transmission delay, compliance security, and transmission cost. The system evaluates the effectiveness and stability of cross-border data transmission strategies in complex regulatory environments by simulating the collaborative decision-making process of each agent in data transmission path planning, resource scheduling, and compliance checks. Existing simulation methods generally adopt scalarization methods based on linear weighted summation or multi-objective optimization methods based on Pareto fronts when dealing with multi-agent preference conflicts. The former maps the preferences of all agents to a common metric and evaluates them as a scalar by combining them with preset weights, while the latter uses Pareto dominance relationships to characterize the trade-offs between multiple objectives and outputs a set of feasible solutions where none are dominant.

[0003] However, existing methods suffer from fundamental flaws when dealing with incommensurable conflict scenarios involving preferences: when different supervisory agents have irreconcilable priorities for target indicators, and their primary preferred objectives differ, the linear weighted summation method fails because the weights are incomparable among different agents, while the Pareto front method cannot output an executable simulation decision because it cannot make a unique choice among multiple mutually exclusive solutions. This leads to a deadlock where the simulation system is stuck with no feasible solutions to output when the preference acyclic graph cannot be topologically ordered. A deeper problem is that existing simulation methods assume the existence of a commensurable common value scale among agents, failing to identify logically exclusive conflicts between word order preference vectors at the semantic level. Consequently, they lack the ability to generate a temporarily acceptable feasible solution from a state of preference conflict without introducing meta-priorities (i.e., pre-defined priority ordering rules that all parties agree to follow). Summary of the Invention

[0004] Based on this, the purpose of the present invention is to provide a simulation method and system for cross-border data transmission based on a multi-agent grid, which can identify preference conflicts and generate provisionally consistent feasible solutions through negotiation mechanisms between agents without relying on a preset priority order.

[0005] The objective of this invention is achieved through the following solution:

[0006] In a first aspect, the present invention provides a cross-border data transmission simulation method based on a multi-agent mesh, comprising the following steps:

[0007] S1: Collect features from each regulatory agent and construct an initial preference state set from the collected preference states of each regulatory agent; wherein, the preference state of each regulatory agent is represented as a descending vector composed of the target indicators in the global target dictionary set.

[0008] S2: Perform logical conflict detection on the initial preference state set, and generate inconsistency flags and inconsistency agent pairs to represent irreconcilable preference states based on the priority ranking correlation analysis of each regulatory agent on the target indicators.

[0009] S3: When the inconsistency flag indicates the existence of inconsistent agent pairs in the set of inconsistent agent pairs, calculate the preference concession interval of each conflicting agent in the set of inconsistent agent pairs based on the adjustment tendency information of the preference ranking of each agent pair extracted from historical negotiation data, and generate the negotiation mediation parameters of each conflicting agent based on the preference concession interval.

[0010] S4: Based on the negotiation and mediation parameters, each conflict regulatory agent is guided to iteratively exchange and correct the priority ranking of each target indicator within its own preference concession range. The negotiation and mediation parameters are continuously updated according to the deviation vector fed back by each conflict regulatory agent during the iterative exchange and correction process, until the deviation of the preference correction of each conflict regulatory agent converges to below the preset acceptance threshold, and the final preference correction result of each conflict regulatory agent is obtained.

[0011] S5: Synchronize the final preference correction results of each conflict-monitoring agent to the global preference configuration space of the multi-agent mesh, and perform cross-border data transmission simulation with the updated global preference state.

[0012] In one embodiment, S2 of the cross-border data transmission simulation method based on a multi-agent mesh provided by the present invention specifically includes the following steps:

[0013] S21: The semantic deconstruction method based on word order vectors extracts semantic features from the descending order vectors of each regulatory agent in the initial preference state set. The position number of each target indicator in its descending order vector is extracted as a priority quantization value, and a priority number vector of each regulatory agent is generated.

[0014] S22: Perform pairwise correlation measurement on the priority number vectors of each regulatory agent, calculate the Spearman rank correlation coefficient of any two regulatory agents' priority number vectors, and generate the target priority correlation quantification matrix for each pair of agents.

[0015] S23: Perform logical comparison processing on each element in the target priority correlation quantification matrix with the preset logical conflict discrimination threshold, extract agent pairs whose first priority targets are different and whose Spearman rank correlation coefficient is less than the logical conflict discrimination threshold, mark them as logically irreconcilable conflict pairs, and generate inconsistency flags and inconsistent agent pair sets to characterize the state of irreconcilable preferences.

[0016] In one embodiment, S3 of the cross-border data transmission simulation method based on a multi-agent mesh provided by the present invention specifically includes the following steps:

[0017] S31: Based on the time window sliding cumulative method, the preference evolution trajectory of the historical negotiation dataset is extracted, the preference order change records of each regulatory agent in different negotiation rounds are analyzed, the adjustment ratio of each regulatory agent on the kth order preference is calculated, and the preference adjustment ratio sequence of each regulatory agent is generated.

[0018] S32: Based on the preference migration tendency quantification function, the preference adjustment ratio sequence of each regulatory agent is nonlinearly mapped to the tendency coefficient in the interval of 0 to 1, and a preference migration matrix is ​​generated that reflects the concession tendency distribution of each regulatory agent to different priorities.

[0019] S33: Perform concession space quantization operation on the elements of the preference migration matrix of the inconsistent agent to each conflicting regulatory agent in the set. Combine the original priority index of the conflicting regulatory agent in the kth order preference with the total dimension of the descending vector to calculate the maximum adjustable range of the conflicting regulatory agent in each target dimension and generate the preference concession interval of each conflicting regulatory agent.

[0020] S34: Based on the preference concession interval, perform boundary compression weighted average on the negotiation activity and concession saturation of each conflict regulatory agent, and quantify the boundary compression weighted average result as a control parameter to guide the preference ranking to migrate to a coexisting state, thereby generating negotiation and mediation parameters for each conflict agent.

[0021] In one embodiment, the calculation formula for the negotiation and mediation parameters of the cross-border data transmission simulation method based on multi-agent mesh provided by the present invention is as follows:

[0022] ;

[0023] in, The negotiation and mediation parameter for the current round t has a value range of [value range missing]. This is used to balance the level of protection for the core preferences of both parties. The closer the value is to 1, the more likely it is to prioritize satisfying the core preferences of the regulatory agent i. The closer the value is to 0, the more likely it is to prioritize satisfying the core preferences of the regulatory agent j. The base learning rate controls the basic step size for each update. and , respectively, are the L2 norms of the deviation vectors fed back by regulatory agents i and j, used to quantify the intensity of each party's deviation from the new proposal; It is a very small positive number, used to prevent the denominator from being zero; The sign of the ratio determines the direction of the adjustment coefficient's movement; the larger the absolute value of the ratio, the greater the magnitude of the movement. As a historical guiding accelerator, among which To accelerate the sensitivity coefficient, The historical average number of negotiation convergence rounds between conflict-regulating agents. This is the smooth width parameter.

[0024] In one embodiment, S4 of the cross-border data transmission simulation method based on a multi-agent mesh provided by the present invention specifically includes the following steps:

[0025] S41: Based on the preference concession range of each conflict regulatory agent, the priority ranking of each target indicator is processed by the degradation exchange under boundary constraints. The negotiation and mediation parameters are used as the weights for the concession range. The adjusted priority ranking of each conflict regulatory agent is calculated to generate the negotiation proposal matrix for this round.

[0026] S42: Based on the acceptance score function, the degree of preference deviation of each conflict regulatory agent in the negotiation proposal matrix is ​​weighted and evaluated. The acceptance score is calculated by combining the sensitivity coefficient of each target dimension. The preference adjustment requests fed back by the conflict regulatory agents whose acceptance scores are lower than the preset acceptance threshold are extracted as deviation vectors, and the deviation vector set of each conflict regulatory agent is generated.

[0027] S43: Based on the difference in the magnitude of each vector element in the set of divergence vectors, the negotiation and mediation parameters are directionally corrected. The offset direction and magnitude of the negotiation and mediation parameters are dynamically adjusted according to the difference in the L2 norm of the divergence vectors of both parties, and the updated negotiation and mediation parameters are generated.

[0028] In one embodiment, S5 of the cross-border data transmission simulation method based on a multi-agent mesh provided by the present invention specifically includes the following steps:

[0029] S51: Perform matrix aggregation processing on the final preference correction results of each conflict regulatory agent, merge the row vectors output independently by each conflict regulatory agent into a joint preference adjustment unit according to the identification index of each conflict regulatory agent in the multi-agent grid, and generate a global preference adjustment matrix;

[0030] S52: By using the state synchronization write protocol, the global preference adjustment matrix and the global preference configuration space of the multi-agent mesh are compared and atomically covered. The original conflicting preference states in the global preference configuration space are replaced with the adjusted preference states agreed upon by each conflicting agent, and an updated global preference configuration space is generated.

[0031] S53: The updated global preference configuration space is embedded into the deep reinforcement learning decision framework of the multi-agent grid. The updated global preference configuration space is used as the state input of the policy network, and the joint preference corresponding to the final preference correction result of each conflict supervisory agent is used as the scalar weight of the reward to drive the multi-agent grid to perform the cross-border data transmission simulation decision task.

[0032] Secondly, the present invention provides a cross-border data transmission simulation system based on a multi-agent mesh, which is configured with the following modules:

[0033] The initial preference set construction module is used to collect features from each regulatory agent and construct the collected preference states of each regulatory agent into an initial preference state set; wherein, the preference state of each regulatory agent is represented as a descendingly ordered vector composed of target indicators in the global target dictionary set.

[0034] The preference conflict detection module is used to detect logical conflicts in the initial preference state set. Based on the priority ranking correlation analysis of each regulatory agent on the target indicator, it generates inconsistency flags and a set of inconsistent agent pairs to represent the irreconcilable preference state.

[0035] The negotiation parameter generation module is used to calculate the preference concession interval of each conflicting agent in the inconsistent agent pair set based on the adjustment tendency information of the preference ranking of each agent pair extracted from historical negotiation data when the inconsistency flag indicates the existence of inconsistent agent pairs in the inconsistent agent pair set, and to generate the negotiation mediation parameters of each conflicting agent based on the preference concession interval.

[0036] The preference iteration correction module is used to guide each conflict regulatory agent to iteratively exchange and correct the priority ranking of each target indicator within their respective preference concession range based on the negotiation and mediation parameters. The negotiation and mediation parameters are continuously updated based on the deviation vector fed back by each conflict regulatory agent during the iterative exchange and correction process, until the deviation of the preference correction of each conflict regulatory agent converges to below the preset acceptance threshold, and the final preference correction result of each conflict regulatory agent is obtained.

[0037] The global configuration simulation module is used to synchronize the final preference correction results of each conflict supervisory agent to the global preference configuration space of the multi-agent grid, and to simulate cross-border data transmission with the updated global preference state.

[0038] Thirdly, this application provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement any of the above-mentioned cross-border data transmission simulation methods based on multi-agent meshes.

[0039] Fourthly, this application provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements any of the above-mentioned simulation methods for cross-border data transmission based on multi-agent meshes.

[0040] In summary, the cross-border data transmission simulation method based on a multi-agent grid provided in this application collects the preference states of each supervisory agent and constructs an initial preference state set in descending order vector form. It then performs semantic-level logical conflict detection on preference conflicts, effectively identifying irreconcilable states between multiple agents' preferences without requiring a pre-defined global priority ranking. Based on this, it dynamically calculates the preference concession intervals of each conflicting agent using historical negotiation data and generates negotiation and mediation parameters. This enables the generation and output of feasible solutions even when the preferences of different agents are incommensurable, overcoming the fundamental deficiency of traditional weighted summation and Pareto front paradigms in failing to output feasible solutions in multi-objective conflict scenarios. By guiding each conflicting agent to iteratively exchange and correct priority ranking within their respective preference concession intervals, and continuously and dynamically updating the negotiation and mediation parameters based on the feedback deviation vector, the multi-agent system can achieve rapid convergence in environments with irreconcilable preferences, reaching a provisional consensus feasible solution that all parties can agree to without rejection. By synchronizing the final preference correction results to the global preference configuration space and making simulation decisions based on the updated global preference state, data-driven continuous adaptive optimization can be achieved, thereby effectively improving the decision reliability and strategy output capability of the cross-border data transmission simulation system in preference conflict scenarios.

[0041] To better understand and implement this invention, the following detailed description is provided in conjunction with the accompanying drawings. Attached Figure Description

[0042] Figure 1 A flowchart illustrating a cross-border data transmission simulation method based on a multi-agent mesh, provided for an embodiment of this application;

[0043] Figure 2 This is a schematic diagram of the structure of a cross-border data transmission simulation system based on a multi-agent mesh, provided as another embodiment of this application. Detailed Implementation

[0044] To facilitate understanding of the present invention, a more complete description will be given below with reference to the accompanying drawings. Preferred embodiments of the invention are shown in the drawings. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided to provide a thorough and complete understanding of the disclosure of the invention.

[0045] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0046] In one embodiment, such as Figure 1 As shown, a simulation method for cross-border data transmission based on a multi-agent mesh is provided. This embodiment illustrates the method by applying it to a terminal. It is understood that this method can also be applied to a server, and to a system including both a terminal and a server, and is implemented through interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0047] S1: Collect features from each regulatory agent and construct an initial preference state set from the collected preference states of each regulatory agent; wherein, the preference state of each regulatory agent is represented as a descending vector composed of the target indicators in the global target dictionary set.

[0048] Specifically, the system performs full-domain feature collection for all cross-border regulatory agents within the multi-agent mesh architecture. The system pre-builds a unified standard global target dictionary set, which includes all conventional evaluation indicators corresponding to cross-border data transmission simulation operations. All indicators constitute a fixed-dimensional evaluation system, serving as a unified benchmark for constructing preferences for all regulatory agents. The system traverses all regulatory agents covering each sovereign jurisdiction, continuously collecting the data transmission decision-related features corresponding to each agent. The collected content includes the target indicator adaptation relationships and priority adaptation logic formed by each agent adapting to the regulatory rules of its own jurisdiction.

[0049] The system sorts all target indicators corresponding to a single regulatory agent according to a unified global target dictionary set dimension standard. The sorting rule follows a fixed logic of indicator priority from high to low, thereby generating a preference ranking vector for each regulatory agent. The number of dimensions of all preference ranking vectors is consistent with the number of indicators in the global target dictionary set, and the order of elements within the vector directly corresponds to the priority order of the indicators. The system uniformly collects, standardizes, and structures the preference ranking vectors generated by all independent regulatory agents, removes duplicate and invalid data, completes the integration processing of the preference data of all regulatory agents, and constructs a complete initial preference state set. The initial preference state set stores the original preference data of all agents in the multi-agent grid simulation scenario.

[0050] S2: Perform logical conflict detection on the initial preference state set, and generate inconsistency flags and inconsistency agent pairs to represent irreconcilable preference states based on the priority ranking correlation analysis of each regulatory agent on the target indicators.

[0051] Specifically, the system invokes a pre-built initial preference state set to perform a full-domain logical conflict detection operation. Using the priority ranking data of target indicators for each regulatory agent as the core basis, the system performs correlation analysis on the preference ranking of multiple agents. The system establishes a unified order relationship verification mechanism, performing a dimension-by-dimensional cross-comparison of the preference ranking vectors of any two different regulatory agents within the initial preference state set, verifying whether there is a mutually exclusive relationship between the priority judgment logic of different agents for any two sets of target indicators. Through continuous full-domain traversal comparison, the system identifies preference ranking combinations that cannot be compatible between different agents; the agent pairs corresponding to such combinations are defined as inconsistent agent pairs. The system uniformly collects and structures all identified inconsistent agent pairs, forming a complete set of inconsistent agent pairs. Based on the existence status of the inconsistent agent pair set, the system generates corresponding inconsistency flags. When the set contains valid data, the system determines that there is an irreconcilable preference state within the current multi-agent grid; when the set contains no valid data, the system determines that the preference ranking logic of all agents within the current multi-agent grid is mutually compatible. The system relies on semantic-level verification of sorting logic to complete conflict identification, replacing the traditional weight value comparison method. This enables effective identification of multi-agent preference conflicts without topological sorting, accurately locating all preference conflict points within the multi-agent grid, and providing accurate conflict data support for the negotiation and mediation process.

[0052] S3: When the inconsistency flag indicates the existence of inconsistent agent pairs in the set of inconsistent agent pairs, calculate the preference concession interval of each conflicting agent in the set of inconsistent agent pairs based on the adjustment tendency information of the preference ranking of each agent pair extracted from historical negotiation data, and generate the negotiation mediation parameters of each conflicting agent based on the preference concession interval.

[0053] Specifically, the system reads the output of the inconsistency flag. If the flag indicates the existence of a set of validly inconsistent agent pairs, it initiates the agent preference negotiation and mediation process. The system retrieves a locally stored multi-agent historical negotiation database, which stores records of preference adjustment behaviors of each regulatory agent during multiple simulation iterations. The system performs feature extraction operations on the historical negotiation dataset to uncover patterns in the ranking and adjustment behaviors of each regulatory agent in preference conflict scenarios. It also analyzes the adjustment adaptation characteristics of each agent for different target indicators, forming preference ranking and adjustment tendency information for each conflicting agent.

[0054] The system combines the matching relationship between the regulatory constraints of each agent and the current conflict scenario, distinguishes between the adjustable attributes and fixed constraint attributes of each target indicator, and performs quantitative calculations based on the extracted adjustment tendency information to determine the preference concession intervals for each conflicting agent. The preference concession intervals define the legal adjustment range of each agent's preference ranking, and all adjustment operations are limited to execution within the corresponding intervals. Relying on multiple computational bases such as the concession interval range of each agent, historical negotiation adaptation characteristics, and current conflict correlations, the system performs normalization computation processing to generate negotiation and mediation parameters for each conflicting agent. These parameters are used to constrain the execution rules and control logic of iterative preference correction, providing a standardized computational basis for the autonomous resolution of multi-agent preference conflicts.

[0055] S4: Based on the negotiation and mediation parameters, each conflict monitoring agent is guided to iteratively exchange and correct the priority ranking of each target indicator within its own preference concession range. The negotiation and mediation parameters are continuously updated according to the deviation vector fed back by each conflict monitoring agent during the iterative exchange and correction process, until the deviation of the preference correction of each conflict monitoring agent converges to below the preset acceptance threshold, and the final preference correction result of each conflict monitoring agent is obtained.

[0056] Specifically, the system loads the generated negotiation and mediation parameters, and drives each conflict monitoring agent to perform iterative exchange and correction operations on preference ranking based on the constraint logic of the parameters. All correction operations strictly correspond to the preference concession range of each agent, preventing invalid correction behaviors that exceed the constraint range. After a single iteration correction process is completed, the system compares the updated preference ranking data of each conflict monitoring agent with the initial preference ranking data. It generates a deviation vector for each agent through multi-dimensional difference calculation, which objectively reflects the degree of change of the original preference ranking by a single correction operation. The system uses the deviation vector as a basis for dynamic adjustment, continuously adaptively updating and adjusting the negotiation and mediation parameters for each agent to optimize the execution rules for the next round of iteration correction. The system cyclically executes the iterative correction and parameter update process, continuously narrowing the deviation of preference corrections for each conflict agent. The system compares the degree of deviation of each agent's correction with the preset acceptance threshold in real time. When the degree of deviation of all conflict-monitoring agents is stable within the threshold coverage range, the system terminates the iterative loop process, solidifies the preference ranking data in the current iteration state into the final preference correction result of each conflict-monitoring agent, realizes the balanced adaptation of multi-agent preference conflicts, and outputs feasible preference data that can be adapted by multiple parties.

[0057] S5: Synchronize the final preference correction results of each conflict-monitoring agent to the global preference configuration space of the multi-agent mesh, and perform cross-border data transmission simulation with the updated global preference state.

[0058] Specifically, the system retrieves the final preference correction results from each conflicting regulatory agent and synchronously inputs the corrected preference ranking data into the global preference configuration space corresponding to the multi-agent grid, completing the global configuration data update operation. The global preference configuration space stores all agent preference parameters required for the operation of the multi-agent grid simulation system, supporting parameter scheduling and logical operations in the simulation process. During data synchronization, the system replaces and updates the original preference data of conflicting agents while simultaneously retaining the initial preference data of conflict-free agents, ensuring the integrity and accuracy of all agent preference data within the global preference configuration space. Through data update operations, the system eliminates the mutual exclusion problem of preference logic within the multi-agent grid, forming a unified and conflict-free global preference state. Based on the updated global preference state, the system initiates a standardized cross-border data transmission simulation process, sequentially completing the entire simulation process, including cross-border data transmission path planning, grid-internal transmission resource scheduling, multi-region compliance status verification, and transmission strategy operation status evaluation. The system relies on conflict-free global preference parameters to complete simulation calculations, solving the simulation stagnation problem caused by preference conflicts in traditional simulation modes and improving the executability and operational stability of data transmission simulation operations in complex cross-border regulatory scenarios.

[0059] In summary, the cross-border data transmission simulation method based on a multi-agent grid provided in this application collects the preference states of each supervisory agent and constructs an initial preference state set in descending order vector form. It then performs semantic-level logical conflict detection on preference conflicts, effectively identifying irreconcilable states between multiple agents' preferences without requiring a pre-defined global priority ranking. Based on this, it dynamically calculates the preference concession intervals of each conflicting agent using historical negotiation data and generates negotiation and mediation parameters. This enables the generation and output of feasible solutions even when the preferences of different agents are incommensurable, overcoming the fundamental deficiency of traditional weighted summation and Pareto front paradigms in failing to output feasible solutions in multi-objective conflict scenarios. By guiding each conflicting agent to iteratively exchange and correct priority ranking within their respective preference concession intervals, and continuously and dynamically updating the negotiation and mediation parameters based on the feedback deviation vector, the multi-agent system can achieve rapid convergence in environments with irreconcilable preferences, reaching a provisional consensus feasible solution that all parties can agree to without rejection. By synchronizing the final preference correction results to the global preference configuration space and making simulation decisions based on the updated global preference state, data-driven continuous adaptive optimization can be achieved, thereby effectively improving the decision reliability and strategy output capability of the cross-border data transmission simulation system in preference conflict scenarios.

[0060] In one embodiment, S2 of the cross-border data transmission simulation method based on a multi-agent mesh provided by the present invention specifically includes the following steps:

[0061] S21: The semantic deconstruction method based on word order vectors extracts semantic features from the descending order vectors of each regulatory agent in the initial preference state set. The position number of each target indicator in its descending order vector is extracted as a priority quantization value, and priority number vectors of each regulatory agent are generated.

[0062] Specifically, the system invokes preset preference semantic deconstruction operation rules to perform global semantic feature extraction operations on all supervisory agent vectors stored in the initial preference state set in descending order. The system constructs a five-layer linear feature mapping network to complete structured feature transformation. The initial weights of the network follow a uniform distribution initialization rule, and the network bias parameters are uniformly set to zero. The network training uses a gradient descent iterative function to update parameters, and a dedicated feature fitting loss function ensures mapping accuracy. The loss function is used to constrain the fitting error between vector position features and priority quantization values.

[0063] Furthermore, the system performs ordinal calibration operations on the target indicators within each set of descending vectors, and completes the numerical conversion through a self-constructed priority quantization mapping formula, expressed as:

[0064]

[0065] The formula as a whole is dimensionless. The priority of a single agent's single-objective metric is standardized and quantified, where L represents the total number of objective metrics contained in the global objective dictionary set. This represents the position index of the k-th target indicator corresponding to the i-th regulatory agent in the descending order of the vector. The system uses this formula to convert the priority values ​​of all target indicators, transforming the text-based index structure into a computationally achievable numerical structure. The calculation process follows the physical principle of ordinal normalization, achieving a linear correspondence between the ranking and the quantified priority value. The system collects all priority quantified values ​​corresponding to a single regulatory agent, reorganizes the data according to the original indicator ranking order, and generates a priority index vector with unified dimensions and standardized values. The priority index vectors corresponding to all agents together constitute the global preference quantification dataset.

[0066] S22: Perform pairwise correlation measurement on the priority number vectors of each regulatory agent, calculate the Spearman rank correlation coefficient of any two regulatory agents' priority number vectors, and generate the target priority correlation quantification matrix for each agent pair.

[0067] Specifically, the system loads the priority index vectors corresponding to the global regulatory agents, performs a global pairwise correlation measurement operation, and completes the quantitative analysis of the preference association features between agents. The system constructs a four-layer correlation fitting neural network. The hidden layers of the network use linear activation functions, the initial parameters of the network follow a normal distribution initialization rule, the training process adopts an adaptive learning rate iteration mechanism, and a dedicated correlation convergence loss function is configured to constrain vector fitting bias and correlation calculation errors. The system completes vector correlation calculation through a self-constructed adaptive rank correlation measurement formula, expressed as:

[0068]

[0069] The formula as a whole is dimensionless. The degree of rank-position correlation matching between the priority vectors of the two sets of agents. This represents the priority quantification value of the k-th indicator for the i-th regulatory agent. The parameter L represents the priority quantification value of the k-th indicator for the j-th regulatory agent. The physical basis of the calculation is the normalization of the ratio of the squared deviation of the vector rank to the squared magnitude of the vector. The system traverses all agent pairings within the multi-agent grid, completes the correlation coefficient calculation for all combinations, performs numerical normalization and storage according to the agent number order, and constructs a target priority correlation quantification matrix corresponding to the agent pairing relationship in the row and column dimensions. All elements in the matrix adopt standardized dimensionless values ​​to ensure the uniformity and comparability of the correlation data across the entire domain.

[0070] S23: Perform logical comparison processing on each element in the target priority correlation quantification matrix with the preset logical conflict discrimination threshold, extract agent pairs whose first priority targets are different and whose Spearman rank correlation coefficient is less than the logical conflict discrimination threshold, mark them as logically irreconcilable conflict pairs, and generate inconsistency flags and inconsistent agent pair sets to characterize the state of irreconcilable preferences.

[0071] Specifically, the system retrieves the target priority relevance quantization matrix and a preset logical conflict discrimination threshold, and performs a full-domain element logical comparison and conflict screening operation. The system constructs a three-layer conflict discrimination classification network. The network output layer uses a linear classification function, the initial parameters are initialized with zero values, the training process uses a batch gradient descent optimization function, and a dedicated conflict discrimination loss function is configured to distinguish the classification errors between compatible preference data and conflict preference data. The system completes the conflict state determination through a self-constructed preference conflict discrimination operation formula, expressed as:

[0072]

[0073] The overall dimension of the formula is logically dimensionless. This represents the binary decision result for an agent's irreconcilable conflict of preferences. Represents the correlation coefficient of the agent, parameters This represents the system's preset logical conflict detection threshold, parameter. With parameters These represent the first priority target indicators for the two groups of agents, with the symbol 'I' representing the indicator function. The computational basis is a dual conflict judgment criterion of low correlation and core target differentiation. The system traverses all elements within the matrix, performs logical verification and filtering under the dual conditions, and extracts agent pairs that simultaneously satisfy both core target inconsistency and correlation coefficient below a threshold. These agent pairs are marked as logically irreconcilable conflict pairs. The system collects all marked conflict pair data, performs structured summary storage, and synchronously outputs the inconsistency flags for the corresponding states and the complete set of inconsistent agent pairs, thus accurately characterizing the multi-agent preference for irreconcilable states.

[0074] In one embodiment, S3 of the cross-border data transmission simulation method based on a multi-agent mesh provided by the present invention specifically includes the following steps:

[0075] S31: Based on the time window sliding cumulative method, the preference evolution trajectory of the historical negotiation dataset is extracted, the preference order change records of each regulatory agent in different negotiation rounds are analyzed, the adjustment ratio of each regulatory agent on the k-th order preference is calculated, and the preference adjustment ratio sequence of each regulatory agent is generated.

[0076] Specifically, the system employs a time-window sliding cumulative operation mechanism to extract the full-domain trajectory of preference evolution across multiple rounds within the historical negotiation dataset. The system constructs a four-layer temporal feature extraction network to adapt to temporal data processing. The network input layer corresponds to the temporal data of a single agent's multi-round negotiation, and the hidden layer uses dual temporal convolutional operation units. The network's initial parameters follow a uniform distribution initialization rule. During network training, the parameters are updated using a temporal gradient descent iterative function. A dedicated trajectory fitting loss function is configured to constrain the temporal data fitting error. This loss function is constructed based on the sum of squared temporal deviations to ensure the temporal continuity and data accuracy of the extracted preference trajectory.

[0077] Furthermore, the system uses a fixed, uniform sliding window size and updates the window traversal in negotiation rounds as the time unit. It records the preference order changes generated by each regulatory agent during the cross-border data transmission simulation negotiation process round by round, comprehensively collecting indicator ranking adjustment behavior data at different time stages. The system performs quantitative calculations using a self-developed time-series adjustment ratio calculation formula, expressed as:

[0078]

[0079] in, The formula represents the temporal adjustment ratio of a single agent's single-rank preference, and its calculation is based on the statistical normalization criterion for temporal behavior. This represents the number of times the agent's corresponding priority preference changes within a specified time window. This represents the total number of negotiation rounds within a specified time window. The system calculates the preference adjustment ratios for all regulatory agents sequentially and round by round, organizes the data structure according to the time and priority dimensions, and generates a complete preference adjustment ratio sequence with unified dimensions and corresponding time sequences.

[0080] S32: Based on the preference migration tendency quantification function, the preference adjustment ratio sequence of each regulatory agent is nonlinearly mapped to the tendency coefficient in the interval of 0 to 1, and a preference migration matrix is ​​generated that reflects the concession tendency distribution of each regulatory agent to different priority preferences.

[0081] Specifically, the system loads the preference adjustment ratio sequences corresponding to each regulatory agent and completes the nonlinear mapping transformation through a preference transfer tendency quantization function, realizing the transformation of discrete adjustment ratio data into continuous tendency coefficient data. The system constructs a five-layer nonlinear mapping neural network to complete the feature transformation. The hidden layers are uniformly set with a hyperbolic tangent activation function, and the initial weight parameters of the network follow a normal distribution initialization rule. The training process uses an adaptive momentum optimization function to achieve iterative parameter updates, and a dedicated transfer fitting loss function is configured. The loss function aims to minimize the mapping deviation, thus constraining the data distortion problem in the nonlinear mapping process. The system executes a self-developed interval normalized mapping formula to solve for the coefficients, expressed as:

[0082]

[0083] in, This represents the degree of concessionary migration tendency of the agent's corresponding priority preference. The output result is constrained to the interval between zero and one, and is a dimensionless parameter overall. This is the mapping slope control coefficient, used to control the rate at which the adjustment ratio is converted into the trend coefficient. This represents the preference adjustment ratio for each agent's corresponding ranking. The system iterates through all the ranking preference data of all agents, performs nonlinear mapping operations dimension by dimension, collects the migration tendency coefficients corresponding to all rankings, arranges the data in two dimensions according to the agent number and the ranking dimension, and constructs a preference migration matrix that can fully characterize the concession characteristics of each agent's different ranking preferences. The values ​​inside the matrix directly correspond to the degree of concession activity in different preference dimensions.

[0084] S33: Perform concession space quantization operation on the preference migration matrix elements of inconsistent agents for each conflicting regulatory agent in the set. Combine the original priority index of the conflicting regulatory agent in the k-th order preference with the total dimension of the descending arrangement vector to calculate the maximum adjustable range of the conflicting regulatory agent in each target dimension and generate the preference concession interval of each conflicting regulatory agent.

[0085] Specifically, the system retrieves the preference transfer matrices of inconsistent agents towards all conflicting supervisory agents within the set, and performs global concession space quantization calculations based on the agents' original preference ranking data. The system constructs a three-layer boundary quantization calculation network with a linear fitting structure. Initial parameters are uniformly set to zero, and the training process employs a batch gradient descent optimization function with a dedicated boundary error loss function to constrain the accuracy of adjustable amplitude calculations. The system completes dimensionality quantization based on a self-developed maximum adjustable amplitude calculation formula, expressed as:

[0086]

[0087] in, This represents the maximum adjustable space for a single agent's single-objective preference dimension. The overall formula consists of dimensionless relative amplitude parameters, and its calculation is based on the coupling criterion of ranking weight matching and migration tendency. This represents the uniform total dimension of a vector arranged in descending order. The migration tendency coefficient representing the corresponding order of the agent. This represents the original priority number corresponding to the agent's ranking. The system avoids the bias in concession space determination caused by a single data dimension by coupling the ranking weight of the original preference order with historical migration tendency characteristics through a formula. The system solves the maximum adjustable range for each objective indicator dimension of the conflicting agents one by one, defines the upper and lower limits of the preference ranking adjustment boundary for each dimension, integrates the boundary parameters of all dimensions to form a complete value range, and finally generates a standardized preference concession range for each conflicting regulatory agent that adapts to its own regulatory constraints and historical behavioral characteristics.

[0088] S34: Based on the preference concession interval, perform boundary compression weighted average on the negotiation activity and concession saturation of each conflict regulatory agent, and quantify the boundary compression weighted average result as a control parameter to guide the preference ranking to migrate to a coexisting state, thereby generating negotiation and mediation parameters for each conflict agent.

[0089] Specifically, the system loads the preference concession intervals corresponding to each conflict monitoring agent, performs dual-dimensional quantification calculations of negotiation activity and concession saturation based on interval boundary constraints, and solves for standardized negotiation and mediation parameters through a boundary compression weighted average mechanism. The system constructs a four-layer parameter control neural network, with the network output layer adapted to a parameter value range of zero to one. Initial parameters are uniformly and randomly initialized, and the training process employs an adaptive step-size iterative function and a dedicated equilibrium loss function to constrain the control error of the dual-agent preference balance. The system performs iterative update calculations to generate real-time negotiation and mediation parameters, calculated using the following formula:

[0090]

[0091] in, The negotiation and mediation parameter for the current round t has a value range of [value range missing]. This is used to balance the level of protection for the core preferences of both parties. The closer the value is to 1, the more likely it is to prioritize satisfying the core preferences of the regulatory agent i. The closer the value is to 0, the more likely it is to prioritize satisfying the core preferences of the regulatory agent j. The base learning rate controls the basic step size for each update. and , respectively, are the L2 norms of the deviation vectors fed back by regulatory agents i and j, used to quantify the intensity of each party's deviation from the new proposal; It is a very small positive number, used to prevent the denominator from being zero; The sign of the ratio determines the direction of the adjustment coefficient's movement; the larger the absolute value of the ratio, the greater the magnitude of the movement. As a historical guiding accelerator, among which To accelerate the sensitivity coefficient, The historical average number of negotiation convergence rounds between conflict-regulating agents. The system uses a smoothing width parameter. It combines historical negotiation acceleration factors with the deviation difference between the two parties to perform weighted correction, and limits the parameter iteration range through a boundary compression mechanism to continuously converge the parameters to a stable value range. Finally, it generates standardized negotiation and mediation parameters that can guide conflict preferences to migrate to a coexistence state.

[0092] In one embodiment, S4 of the cross-border data transmission simulation method based on a multi-agent mesh provided by the present invention specifically includes the following steps:

[0093] S41: Based on the preference concession range of each conflict regulatory agent, the priority ranking of each target indicator is processed by the degradation exchange under boundary constraints. The negotiation and mediation parameters are used as the weights for the concession range. The adjusted priority ranking of each conflict regulatory agent is calculated to generate the negotiation proposal matrix for this round.

[0094] Specifically, the system retrieves the preference concession intervals and negotiation parameters corresponding to each conflict monitoring agent, and performs a downgraded exchange process to prioritize various target indicators under the constraint of interval boundaries. The system constructs a four-layer boundary constraint iterative network to carry out data operations. The network input layer carries the agent's original preference ranking and concession interval data, and the hidden layer sets up linear constraint operation units. The initial weights of the network adopt a uniform distribution initialization method, and the bias parameters are uniformly set to zero. During the training process, the parameters are iteratively updated using the batch gradient descent function. A dedicated boundary constraint loss function is configured. The loss function takes the preference adjustment value not exceeding the boundary of the concession interval as the constraint objective, quantifies the excess deviation, and completes the error convergence correction.

[0095] Furthermore, the system sets the real-time generated negotiation and mediation parameters as the allocation weights for the concession magnitudes of multiple agents, and completes the quantitative calculation based on the self-developed priority adjustment ranking calculation formula, which is expressed as:

[0096]

[0097] in, This represents the priority ranking result after iterative adjustment of a single agent's single-objective metric. Represents the original priority number of the intelligent agent. The value representing the upper limit of the preference concession range is ranked. Representative negotiation and mediation parameters, The coefficients for the downgraded exchange operation are based on the weight allocation adjustment criteria under boundary value range constraints. The system traverses all target indicator dimensions of all conflict-monitoring agents, iteratively calculates the priority ranking for each dimension, organizes the preference ranking data of all agents after this round of adjustments, completes the two-dimensional data arrangement according to the correspondence between agent number and target indicator dimension, and constructs a standardized negotiation proposal matrix containing all negotiation correction results of this round, providing a complete iterative data sample for preference acceptance evaluation.

[0098] S42: Based on the acceptance score function, the degree of preference deviation of each conflict regulatory agent in the negotiation proposal matrix is ​​weighted and evaluated. The acceptance score is calculated by combining the sensitivity coefficient of each target dimension. The preference adjustment requests fed back by the conflict regulatory agents whose acceptance scores are lower than the preset acceptance threshold are extracted as deviation vectors, and the deviation vector set of each conflict regulatory agent is generated.

[0099] Specifically, the system loads global preference adjustment data from the negotiation proposal matrix and performs a weighted evaluation of the preference deviation of each conflict-monitoring agent using a standardized acceptance scoring function. The system constructs a five-layer multi-dimensional scoring neural network, with nonlinear mapping operation units configured in the hidden layers. The activation function is a hyperbolic tangent function, and the initial parameters follow a normal distribution initialization rule. An adaptive learning rate optimization algorithm is used during training, and a dedicated scoring fitting loss function is configured to constrain the computational error of the multi-dimensional weighted scoring, ensuring the objectivity and stability of the scoring results. The system introduces fixed sensitivity coefficients corresponding to each target indicator and combines indicator sensitivity with preference deviation magnitude to complete a weighted fusion calculation. The self-developed acceptance score calculation formula is as follows:

[0100]

[0101] in, The overall level of acceptance of the proposed solutions by a single agent in this round of negotiations. The sensitivity coefficient represents the k-th target indicator. This represents the priority ranking after this round of adjustments. Represents the original priority ranking. The total number of dimensions representing the target indicators is calculated based on the multi-dimensional sensitivity weighted deviation normalization criterion. The system calculates the acceptance scores of all conflicting agents, compares the results with the system's preset acceptance threshold, filters out conflicting agents whose scores do not meet the threshold, extracts the preference adjustment deviation data for each target dimension of the corresponding agent, and organizes them into structured deviation vectors. All the filtered deviation vectors are collected to construct a complete set of conflicting agent deviation vectors.

[0102] S43: Based on the difference in the magnitude of each vector element in the set of divergence vectors, the negotiation and mediation parameters are directionally corrected. The offset direction and magnitude of the negotiation and mediation parameters are dynamically adjusted according to the difference in the L2 norm of the divergence vectors of both parties, and the updated negotiation and mediation parameters are generated.

[0103] Specifically, the system reads the global vector data within the set of divergence vectors and performs directional correction of the negotiation and mediation parameters based on the difference in vector magnitudes. The system constructs a three-layer parameter correction computation network using a linear difference fitting structure. Initial parameters are uniformly set to zero. The training process relies on stochastic gradient descent for iterative optimization, and a dedicated parameter correction loss function is configured to eliminate the difference in preference bias between the two agents, thus constraining the accuracy and effectiveness of parameter correction. The system retrieves the two sets of divergence vectors corresponding to the conflicting agent pairs, calculates the vector magnitudes using fixed L2 norm operation rules, and quantifies the difference based on a self-developed parameter correction difference formula, expressed as follows: The formula represents a dimensionless difference parameter, which physically represents the difference in the intensity of the preference adjustment demands of the two conflicting agents. The parameters in the formula... and These are the squared L2 norm values ​​of the two sets of divergence vectors, calculated using the vector space deviation intensity difference criterion. The system determines the offset direction of the negotiation parameters based on the positive or negative attribute of the difference, quantifies the parameter adjustment magnitude based on the absolute value of the difference, and dynamically corrects the negotiation adjustment parameters by combining them with predetermined parameter iteration logic, avoiding the preference correction imbalance problem caused by fixed parameter iteration. The system completes this round of global parameter update and outputs new negotiation adjustment parameters that adapt to the current conflict deviation state and have bidirectional dynamic adjustment capabilities, providing a precise parameter control basis for the next round of preference iteration correction.

[0104] In one embodiment, S5 of the cross-border data transmission simulation method based on a multi-agent mesh provided by the present invention specifically includes the following steps:

[0105] S51: Perform matrix aggregation processing on the final preference correction results of each conflict regulatory agent, merge the row vectors output independently by each conflict regulatory agent into a joint preference adjustment unit according to the identification index of each conflict regulatory agent in the multi-agent grid, and generate a global preference adjustment matrix.

[0106] Specifically, the system performs global matrix aggregation processing on the final preference correction results for all conflict-monitoring agents, completing the standardization transformation from individual preference data to global structured matrix data. The system constructs a four-layer adaptive aggregation neural network to handle data computation. The network input layer receives row vector data of individual agent preferences, and the hidden layer is configured with matrix dimension calibration computation units. The initial weight parameters of the network are initialized using a uniform distribution, and the bias parameters are uniformly set to zero. During training, parameter updates are completed using a batch gradient descent iteration function. A dedicated matrix aggregation loss function is configured, with the optimization objective of minimizing vector dimension alignment error, ensuring dimensional uniformity and data integrity during the multi-agent vector aggregation process.

[0107] Furthermore, the system retrieves the preference correction row vectors independently generated by all conflict-monitoring agents. Based on the pre-defined agent identifier indexing rules within the multi-agent grid, it performs ordered merging and recombination of the disordered individual preference vectors. Structured integration is then achieved through a self-developed global matrix aggregation formula, expressed as follows: The formula outputs a two-dimensional structured matrix, with dimensions adapted to preference sorting and quantization. Physically, it represents an aggregation matrix of global conflict agent preference correction data. In the formula, parameter Ω represents the set of all conflict agent indices within the multi-agent grid, and parameter V_i represents the final preference correction row vector corresponding to the i-th conflict agent. The physical basis of the operation is the agent index dimension mapping and vector space alignment criterion. The system completes the ordered merging of all individual preference vectors, forming a joint preference adjustment unit that covers all conflict agents and has a unified and regular dimension. Based on this joint preference adjustment unit, a standardized global preference adjustment matrix is ​​generated, completely preserving the global preference data after this round of negotiation and correction.

[0108] S52: By using the state synchronization write protocol, the global preference adjustment matrix and the global preference configuration space of the multi-agent grid are compared and atomically overwritten. The original conflicting preference states in the global preference configuration space are replaced with the adjusted preference states agreed upon by all conflicting agents, and an updated global preference configuration space is generated.

[0109] Specifically, the system invokes a preset state synchronization write protocol to complete the differential verification and atomic coverage update of the global preference adjustment matrix and the multi-agent grid global preference configuration space. The system constructs a three-layer state comparison and verification network to identify data differences. The network adopts a linear difference operation structure, with initial parameters uniformly set to zero. The training process uses an adaptive step-size gradient descent optimization function and configures a dedicated state difference loss function to quantify the magnitude of the state deviation between the old and new preference data, ensuring the accuracy of data replacement. The system completes the full-domain data comparison operation through a self-developed state difference discrimination formula, expressed as:

[0110]

[0111] in, This represents the overall change in global preference data. Represents global preference adjustment matrix data. The system represents the original stored data of the global preference configuration space. The physical basis for computation is the matrix Frobenius norm differential quantization criterion, used to characterize the degree of difference in global preference states. Based on the atomic execution rules of the state synchronization write protocol, the system performs a full-coverage replacement operation on conflicting preference data with data differences, while retaining and storing stable preference data without differences. Each update process possesses uninterrupted atomic execution characteristics, avoiding problems such as missing or corrupted local data updates. After completing the global data replacement and correction, the system completely eliminates conflicting preference logic states within the multi-agent mesh, forming a new global preference configuration space that is data-unified, logically compatible, and state-stable, thus achieving the global solidification of the multi-agent preference negotiation results.

[0112] S53: The updated global preference configuration space is embedded into the deep reinforcement learning decision framework of the multi-agent grid. The updated global preference configuration space is used as the state input of the policy network, and the joint preference corresponding to the final preference correction result of each conflict supervisory agent is used as the scalar weight of the reward to drive the multi-agent grid to perform the cross-border data transmission simulation decision task.

[0113] Specifically, the system embeds the updated global preference configuration space into the deep reinforcement learning decision framework built into the multi-agent grid, completing the global deployment and task-driven initialization of the simulation decision pre-parameters. The system constructs a six-layer deep reinforcement learning policy network structure. The network input layer is configured with perceptual neurons matching the global preference dimension, the hidden layers are configured with nonlinear feature extraction units, and the output layer connects to the cross-border data transmission simulation decision action space. The initial network parameters follow a normal distribution initialization rule. During training, a proximal strategy is used to optimize the iterative function, and a dedicated simulation decision loss function is configured, with the optimization objectives being steady-state convergence of the transmission simulation task and optimal balance of multiple objective indicators.

[0114] Furthermore, the system sets the updated global preference configuration space as the sole state input source for the policy network, ensuring that the network's decision-making is fully aligned with the negotiated and corrected conflict-free preference state. Simultaneously, a weighted incentive mechanism is constructed based on a self-developed joint preference reward quantification formula, expressed as follows: , The comprehensive reward weight for the joint preference fit of multiple agents. Represents the single agent's preference fit coefficient. The vector representing the final preference correction of the agent is calculated based on the one-dimensional norm normalized weighted adaptation criterion. The system drives the deep reinforcement learning decision-making framework to adapt to the balanced preference requirements of each regulatory agent by iteratively updating the joint preference reward weight regulation strategy network. It fully executes the entire simulation decision-making task of cross-border data transmission path planning, resource scheduling, compliance verification, and policy evaluation, achieving standardized simulation operation in scenarios without preference conflicts.

[0115] 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.

[0116] Based on the same inventive concept, this application also provides a multi-agent mesh-based cross-border data transmission simulation system for implementing the aforementioned multi-agent mesh-based cross-border data transmission simulation method. The solution provided by this system is similar to the implementation scheme described in the above method. Therefore, the specific limitations in one or more embodiments of the multi-agent mesh-based cross-border data transmission simulation system provided below can be found in the limitations of the multi-agent mesh-based cross-border data transmission simulation method described above, and will not be repeated here.

[0117] Preferably, such as Figure 2 As shown, this invention provides a cross-border data transmission simulation system 600 based on a multi-agent mesh, which is configured with the following modules:

[0118] The initial preference set construction module 610 is used to collect features from each regulatory agent and construct the collected preference states of each regulatory agent into an initial preference state set; wherein, the preference state of each regulatory agent is represented as a descendingly ordered vector composed of target indicators in the global target dictionary set.

[0119] The preference conflict detection module 620 is used to detect logical conflicts in the initial preference state set. Based on the priority ranking correlation analysis of each regulatory agent on the target indicator, it generates inconsistency flags and a set of inconsistent agent pairs to represent the irreconcilable preference state.

[0120] The negotiation parameter generation module 630 is used to calculate the preference concession interval of each conflicting agent in the inconsistent agent pair set based on the adjustment tendency information of the preference ranking of each agent pair extracted from historical negotiation data when the inconsistency flag indicates that there is an inconsistent agent pair in the inconsistent agent pair set, and generate the negotiation mediation parameter of each conflicting agent based on the preference concession interval.

[0121] The preference iteration correction module 640 is used to guide each conflict regulatory agent to iteratively exchange and correct the priority ranking of each target indicator within their respective preference concession range based on the negotiation and mediation parameters. The negotiation and mediation parameters are continuously updated according to the deviation vector fed back by each conflict regulatory agent during the iterative exchange and correction process, until the deviation of the preference correction of each conflict regulatory agent converges to below the preset acceptance threshold, and the final preference correction result of each conflict regulatory agent is obtained.

[0122] The global configuration simulation module 650 is used to synchronize the final preference correction results of each conflict supervisory agent to the global preference configuration space of the multi-agent grid, and to perform cross-border data transmission simulation with the updated global preference state.

[0123] In one embodiment, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the above-described simulation method for cross-border data transmission based on a multi-agent mesh.

[0124] In one embodiment, this application also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the above-described simulation method for cross-border data transmission based on a multi-agent mesh.

[0125] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of those different embodiments or examples.

[0126] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0127] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various variations or substitutions within the technical scope disclosed in this application, and these should all be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A simulation method for cross-border data transmission based on a multi-agent mesh, characterized in that, Includes the following steps: S1: Collect features from each regulatory agent and construct an initial preference state set from the collected preference states of each regulatory agent; wherein, the preference state of each regulatory agent is represented as a descending vector composed of the target indicators in the global target dictionary set. S2: Perform logical conflict detection on the initial preference state set, and generate an inconsistency flag and a set of inconsistent agent pairs to represent the irreconcilable preference state based on the priority ranking correlation analysis of each regulatory agent on the target indicator. S3: When the inconsistency flag indicates the existence of inconsistent agent pairs in the inconsistent agent pair set, calculate the preference concession interval of each conflicting agent in the inconsistent agent pair set based on the adjustment tendency information of each agent pair preference ranking extracted from historical negotiation data, and generate negotiation mediation parameters for each conflicting agent based on the preference concession interval. S4: Based on the negotiation and mediation parameters, guide each conflict monitoring agent to iteratively exchange and correct the priority ranking of each target indicator within their respective preference concession intervals, and continuously update the negotiation and mediation parameters according to the deviation vector fed back by each conflict monitoring agent during the iterative exchange and correction process, until the deviation of the preference correction of each conflict monitoring agent converges to below the preset acceptance threshold, and obtain the final preference correction result of each conflict monitoring agent. S5: Synchronize the final preference correction results of each conflict-monitoring agent to the global preference configuration space of the multi-agent mesh, and perform cross-border data transmission simulation with the updated global preference state.

2. The method according to claim 1, characterized in that, S2 includes: S21: Based on the word order vector-based preference semantic deconstruction method, semantic features are extracted from the descending order vector of each regulatory agent in the initial preference state set. The position number of each target indicator in its descending order vector is extracted as a priority quantization value, and a priority number vector of each regulatory agent is generated. S22: Perform a pairwise correlation measurement on the priority number vectors of each regulatory agent, calculate the Spearman rank correlation coefficient of any two priority number vectors of regulatory agents, and generate the target priority correlation quantification matrix for each pair of agents. S23: Perform logical comparison processing on each element in the target priority correlation quantification matrix with the preset logical conflict discrimination threshold, extract agent pairs whose first priority targets are different and whose Spearman rank correlation coefficient is less than the logical conflict discrimination threshold, mark them as logically irreconcilable conflict pairs, and generate inconsistency flags and inconsistent agent pair sets to characterize the state of irreconcilable preferences.

3. The method according to claim 1, characterized in that, S3 includes: S31: Extract the preference evolution trajectory of the historical negotiation dataset based on the time window sliding accumulation method, analyze the preference order change records of each regulatory agent in different negotiation rounds, calculate the adjustment ratio of each regulatory agent on the kth order preference, and generate the preference adjustment ratio sequence of each regulatory agent. S32: Based on the preference migration tendency quantification function, the preference adjustment ratio sequence of each regulatory agent is nonlinearly mapped to the tendency coefficient in the interval of 0 to 1, and a preference migration matrix reflecting the concession tendency distribution of each regulatory agent to different priority preferences is generated. S33: Perform a concession space quantization operation on the preference migration matrix elements of the inconsistent agent for each conflicting regulatory agent in the set. Combine the original priority number of the conflicting regulatory agent in the kth order preference with the total number of dimensions of the descending arrangement vector to calculate the maximum adjustable range of the conflicting regulatory agent in each target dimension and generate the preference concession interval of each conflicting regulatory agent. S34: Based on the preference concession interval, perform boundary compression weighted average on the negotiation activity and concession saturation of each conflict regulatory agent, and quantify the boundary compression weighted average result as a control parameter to guide the preference ranking to migrate to a coexisting state, thereby generating negotiation and mediation parameters for each conflict agent.

4. The method according to claim 1, characterized in that, The calculation formula for the negotiation and mediation parameters is as follows: ； in, The negotiation and mediation parameter for the current round t has a value range of [value range missing]. This is used to balance the level of protection for the core preferences of both parties. The closer the value is to 1, the more likely it is to prioritize satisfying the core preferences of the regulatory agent i. The closer the value is to 0, the more likely it is to prioritize satisfying the core preferences of the regulatory agent j. The base learning rate controls the basic step size for each update. and , respectively, are the L2 norms of the deviation vectors fed back by regulatory agents i and j, used to quantify the intensity of each party's deviation from the new proposal; It is a very small positive number, used to prevent the denominator from being zero; The sign of the ratio determines the direction of the adjustment coefficient's movement; the larger the absolute value of the ratio, the greater the magnitude of the movement. As a historical guiding accelerator, among which To accelerate the sensitivity coefficient, The historical average number of negotiation convergence rounds between conflict-regulating agents. This is the smooth width parameter.

5. The method according to claim 1, characterized in that, S4 includes: S41: Based on the preference concession intervals of each conflict regulatory agent, the priority ranking of each target indicator is downgraded and exchanged under boundary constraints. The negotiation and mediation parameters are used as the weights for the concession range. The adjusted priority ranking of each conflict regulatory agent is calculated to generate the negotiation proposal matrix for this round. S42: Based on the acceptance score function, the degree of preference deviation of each conflict regulatory agent in the negotiation proposal matrix is ​​weighted and evaluated. The acceptance score is calculated by combining the sensitivity coefficient of each target dimension. The preference adjustment requests fed back by the conflict regulatory agents whose acceptance scores are lower than the preset acceptance threshold are extracted as deviation vectors, and a set of deviation vectors of each conflict regulatory agent is generated. S43: Based on the difference in magnitude of each vector element in the set of divergence vectors, the negotiation and mediation parameters are directionally corrected. The offset direction and magnitude of the negotiation and mediation parameters are dynamically adjusted according to the L2 norm difference between the divergence vectors of both parties, and the updated negotiation and mediation parameters are generated.

6. The method according to any one of claims 1-5, characterized in that, S5 includes: S51: Perform matrix aggregation processing on the final preference correction results of each conflict monitoring agent, merge the row vectors independently output by each conflict monitoring agent into a joint preference adjustment unit according to the identification index of each conflict monitoring agent in the multi-agent grid, and generate a global preference adjustment matrix; S52: By performing a difference comparison and atomic overwrite operation on the global preference adjustment matrix and the global preference configuration space of the multi-agent mesh through the state synchronization writing protocol, the original conflicting preference states in the global preference configuration space are replaced with the adjusted preference states agreed upon by each conflicting agent, and an updated global preference configuration space is generated. S53: Embed the updated global preference configuration space into the deep reinforcement learning decision framework of the multi-agent grid, use the updated global preference configuration space as the state input of the policy network, and use the joint preference corresponding to the final preference correction result of each conflict supervisory agent as the reward scalar weight to drive the multi-agent grid to perform the cross-border data transmission simulation decision task.

7. A cross-border data transmission simulation system based on a multi-agent grid, characterized in that, The system includes: The initial preference set construction module is used to collect features from each regulatory agent and construct the collected preference states of each regulatory agent into an initial preference state set; wherein, the preference state of each regulatory agent is represented as a descendingly ordered vector composed of target indicators in the global target dictionary set. The preference conflict detection module is used to detect logical conflicts in the initial preference state set. Based on the priority ranking correlation analysis of each regulatory agent on the target indicator, it generates an inconsistency flag and a set of inconsistent agent pairs to represent the irreconcilable preference state. The negotiation parameter generation module is used to calculate the preference concession interval of each conflicting agent in the inconsistent agent pair set based on the adjustment tendency information of the preference ranking of each agent pair extracted from historical negotiation data when the inconsistency flag indicates the existence of inconsistent agent pairs in the inconsistent agent pair set, and generate negotiation mediation parameters of each conflicting agent based on the preference concession interval. The preference iteration correction module is used to guide each conflict monitoring agent to iteratively exchange and correct the priority ranking of each target indicator within their respective preference concession intervals based on the negotiation and mediation parameters, and to continuously update the negotiation and mediation parameters based on the deviation vector fed back by each conflict monitoring agent during the iterative exchange and correction process, until the deviation of the preference correction of each conflict monitoring agent converges to below the preset acceptance threshold, and then obtain the final preference correction result of each conflict monitoring agent. The global configuration simulation module is used to synchronize the final preference correction results of each conflict-monitoring agent to the global preference configuration space of the multi-agent grid, and to perform cross-border data transmission simulation with the updated global preference state.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the method of any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 6.

Images