Multi-agent cooperative confrontation simulation method based on wide area network synchronization

Abstract

This invention discloses a multi-agent cooperative adversarial simulation method based on wide area network (WAN) synchronization. Addressing the problems of low clock synchronization accuracy, proneness to timing errors, high bandwidth consumption, and insufficient fault tolerance in WAN scenarios, existing technologies construct a three-tiered architecture: global simulation control, regional collaborative decision-making, and local simulation inference. It employs a layered master-slave clock synchronization combined with delay compensation to achieve unified clock references across all nodes, establishes a clock-state dual-loop hybrid synchronization control mechanism, and completes cooperative adversarial inference through a distributed architecture that decouples decision-making and inference. Simultaneously, it sets up fault-tolerant synchronization and conflict resolution mechanisms. This invention can achieve high-precision synchronization, high-continuity operation, and high-reliability inference for large-scale cross-regional multi-agent adversarial simulations in complex WAN environments, adapting to the needs of distributed simulation applications in multiple fields.

Description

Technical Field

[0001] This invention relates to the field of distributed simulation and multi-agent cooperative technology, and in particular to a multi-agent cooperative adversarial simulation method based on wide area network synchronization. Background Technology

[0002] Multi-agent cooperative adversarial simulation is a core simulation technology based on a virtual simulation environment, enabling multiple intelligent agents with autonomous decision-making capabilities to complete cooperative interaction, adversarial game, and strategy verification. It is widely used in various fields such as military confrontation exercises, unmanned swarm cooperative control, intelligent game strategy optimization, and emergency response simulation. Meanwhile, multi-agent cooperative adversarial simulation technology based on wide area network synchronization is a key supporting technology for realizing cross-regional, multi-node, and large-scale distributed simulation. It can break the geographical limitations of local area networks, realize the joint access and collaborative simulation of simulation nodes distributed in different geographical areas, and greatly expand the scale, coverage, and realism of simulation scenarios. It is an important research and application direction in the current field of distributed simulation.

[0003] Most existing multi-agent cooperative adversarial simulation technologies are designed for closed local area network (LAN) environments. Their core architectures typically employ centralized clock synchronization and centralized control. The basic technical process involves a single core server serving as the sole global clock source and control center. All simulation terminal nodes directly synchronize clocks and interact with the core server for state data. A fixed simulation step size is used for periodic synchronization of all simulation state data. The core server generates collaborative decision-making instructions and manages the simulation process for all agents. While this type of technology can operate stably in the low-latency, high-stability environment of a LAN, it suffers from significant technical drawbacks in wide area network (WAN) scenarios. Firstly, the centralized single-clock-source architecture struggles to adapt to the long-distance transmission latency, latency jitter, and uplink / downlink asymmetry characteristics of WAN links. First, the inability to achieve high-precision clock synchronization across all nodes easily leads to simulation timing errors. Second, the fixed-step full data synchronization mode consumes extremely high bandwidth, which can easily cause data packet loss and synchronization delays in wide area network (WAN) bandwidth-constrained environments, making it impossible to balance simulation synchronization real-time performance and operational efficiency. Third, the centralized control and decision-making architecture cannot adapt to the real-time collaboration needs of widely distributed multi-agent systems, and cross-regional interaction latency leads to low collaborative decision-making efficiency, making it difficult to support large-scale agent adversarial simulations. Fourth, the lack of fault-tolerant synchronization mechanisms and simulation state conflict resolution mechanisms for WAN link fluctuations and interruptions makes it easy for simulation interruptions and inconsistent situational data to occur when links are abnormal, resulting in a serious lack of credibility and continuity in adversarial simulations, and failing to meet the usage requirements of large-scale cross-regional multi-agent collaborative adversarial simulations in WAN environments. Summary of the Invention

[0004] To address the aforementioned deficiencies in existing technologies, the present invention aims to provide a multi-agent cooperative adversarial simulation method based on wide area network synchronization. This method solves the problems of low clock synchronization accuracy, easy timing errors, high bandwidth consumption, low cooperative efficiency, poor fault tolerance, and insufficient situational consistency in multi-agent cooperative adversarial simulation under wide area network environments. It achieves high-precision synchronization, high-continuity operation, and high-reliability inference in large-scale cross-regional multi-agent cooperative adversarial simulation.

[0005] To address the aforementioned technical problems, this invention provides the following technical solution: a multi-agent cooperative adversarial simulation method based on wide area network synchronization, comprising the following steps: S1 constructs a multi-agent collaborative adversarial simulation layered architecture in a wide area network environment. The layered architecture includes, from top to bottom, a global simulation control layer 1, a regional collaborative decision-making layer 2, and a local agent simulation inference layer 3. The global simulation control layer 1 is deployed on the core backbone nodes of the wide area network, the regional collaborative decision-making layer 2 is deployed on the edge computing nodes of each geographical partition, and the local agent simulation inference layer 3 is deployed on each terminal simulation node. Each terminal simulation node corresponds to carrying at least one simulation agent instance. S2 performs hierarchical master-slave clock synchronization in the wide area network. The core node of the global simulation control layer 1 is used as the global master clock source, and the edge computing nodes in each region are used as regional slave clock sources. After the regional slave clock sources complete the first-level clock synchronization with the global master clock source, they provide the second-level clock synchronization service for their subordinate terminal simulation nodes. Based on the dynamic prediction results of the wide area network link delay, bidirectional transmission delay compensation is completed, and the sub-millisecond clock reference of all simulation nodes is unified. S3 establishes a dual-ring synchronization control mechanism for multi-agent simulation. The inner ring is a clock synchronization ring, which continuously maintains the clock reference consistency of all simulation nodes based on the hierarchical master-slave clock synchronization in step S2. The outer ring is a simulation state synchronization ring, which adopts a hybrid synchronization mode that combines time step-driven and event-triggered methods to complete the wide-area network distributed synchronization of multi-agent simulation state, adversarial behavior data, and global adversarial situation data. S4 executes distributed collaborative adversarial simulation, with each terminal simulation node based on a unified local clock reference to complete the local behavior inference, environmental interaction physics calculation, and local situational awareness of the corresponding simulated agent; the regional collaborative decision layer 2, based on synchronized regional situational data, completes the generation and issuance of collaborative adversarial decision instructions for the multiple agents under its jurisdiction; the global simulation control layer 1 completes the unified verification of the global adversarial situation, the global step size control of the simulation process, and the global execution of adversarial rules. S5 performs fault-tolerant synchronization and conflict resolution for wide area network link fluctuations. Through node periodic state snapshot mechanism, dynamic redundant synchronization path switching and multi-source situational data consistency verification, it completes real-time correction of simulation timing disorder and state conflict, ensuring the continuity, timing correctness and situational credibility of multi-agent cooperative adversarial simulation in wide area network environment.

[0006] Further, in step S2, the hierarchical master-slave clock synchronization specifically includes: using a bidirectional timestamp exchange mechanism to complete the link transmission delay measurement between master and slave nodes; using an improved unscented Kalman filter algorithm to continuously and dynamically predict the delay jitter and packet loss rate of the wide area network link; and using feedforward delay compensation for the clock synchronization signal based on the prediction results; at the same time, deviation correction is performed for the asymmetric transmission delay of the uplink and downlink of the wide area network, and the clock synchronization error of the fully simulated node is controlled within 1ms.

[0007] Furthermore, in step S3, the hybrid synchronization mode combining time step-driven and event-triggered synchronization is specifically as follows: a globally unified minimum simulation step size is set as the baseline synchronization period, and a full node key state forced synchronization is triggered once after each baseline synchronization period ends; at the same time, an event trigger threshold is set, and when the adversarial behavior command, local situational change, or environmental interaction result of the simulated agent exceeds the preset event trigger threshold, incremental emergency synchronization is immediately triggered without waiting for the baseline synchronization period to arrive.

[0008] Furthermore, in step S3, the simulation state synchronization loop adopts an incremental data synchronization mechanism, which only synchronizes the incremental simulation state data, adversarial behavior instructions and situational characteristic data that have changed in the current cycle, and does not synchronize the full simulation state data; at the same time, a globally unified clock stamp is attached to all synchronized data, and the timing sorting and disorder correction of synchronized data are completed based on the clock stamp.

[0009] Furthermore, in step S4, an execution architecture that decouples collaborative decision-making from local inference is adopted: the regional collaborative decision-making layer 2 only issues the target instructions, task constraints and collaborative rules for collaborative confrontation to the terminal simulation nodes under its jurisdiction, without interfering with the specific behavior calculation of the simulated intelligent agent; after receiving the decision instructions, the terminal simulation nodes autonomously complete the kinematic calculation, behavior decision-making, environmental collision detection and local situational awareness of the intelligent agent locally, and only synchronize the key behavior results and situational data to the upper-level nodes through the wide area network.

[0010] Furthermore, in step S4, the global simulation control layer 1 sets up a global simulation fence synchronization mechanism. Only after all regional edge computing nodes have completed the reporting and verification of key states in the current benchmark synchronization cycle will the global simulation step size advancement command be released to avoid simulation timing disorder. At the same time, a timeout fault tolerance mechanism is set up. For unresponsive nodes that exceed the preset timeout threshold, the valid state data of the previous cycle of the node is used for interpolation to complete the process, so as not to block the continuous advancement of the global simulation process.

[0011] Further, in step S5, the node periodic state snapshot mechanism is as follows: after each baseline synchronization cycle ends, each terminal simulation node generates a local simulation state snapshot and synchronizes it to the storage of the edge computing node in its region. The edge computing node summarizes the snapshots of the nodes under its jurisdiction and synchronizes them to the global simulation control layer 1 for archiving. When the wide area network link is interrupted and restored, the interrupted node completes local state rollback and resynchronization based on the most recent globally consistent state snapshot to avoid the simulation state from being interrupted.

[0012] Further, in step S5, the consistency verification and conflict resolution of the multi-source situational data specifically involves: the global simulation control layer 1 and the regional collaborative decision-making layer 2 performing consistency verification on the situational data reported by each node; when there is a conflict in the situational data of different nodes, weighted fusion is performed based on the clock stamp priority of the data, the confidence weight of the node data, and the data acquisition accuracy level to generate globally consistent adversarial situational benchmark data; at the same time, the corrected situational benchmark data is synchronized to each relevant node to complete the consistency correction of the local state of the node.

[0013] Furthermore, the simulated intelligent agents include red team adversarial intelligent agents, blue team adversarial intelligent agents, and neutral environment intelligent agents. Each simulated intelligent agent belongs to a terminal simulation node in a different geographical region within the wide area network. The intelligent agents can complete cross-node and cross-regional collaborative interaction and adversarial behavior triggering through the wide area network. The global simulation control layer 1 synchronizes the simulation process data of all nodes to complete the determination of the winner and loser of the adversarial parties and the review data archiving of the entire simulation process.

[0014] Furthermore, in step S2, the hierarchical master-slave clock synchronization also includes a dynamic master-slave switching mechanism: when the global master clock source node fails or the link is interrupted, a new global master clock source is automatically elected from the preset backup core nodes to complete the seamless switching of clock synchronization services; when the regional slave clock source node fails, its subordinate terminal simulation nodes are automatically switched to the adjacent normally operating regional slave clock source to ensure the continuity of clock synchronization services.

[0015] One or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages compared with the prior art: A. This invention adopts a hierarchical simulation architecture and a hierarchical master-slave clock synchronization mechanism, which is adapted to the characteristics of long-distance transmission and cross-regional distribution of wide area network links. Through a two-level synchronization architecture of first-level global synchronization + second-level regional synchronization, combined with dynamic prediction of link delay and feedforward compensation, it corrects the asymmetric deviation of uplink and downlink, realizes the sub-millisecond level clock reference unification of all simulation nodes, greatly improves the clock synchronization accuracy in wide area network environment, and avoids simulation timing disorder problems from the root.

[0016] B. This invention employs a dual-loop synchronization control mechanism consisting of an inner clock synchronization loop and an outer simulation state synchronization loop. It combines a hybrid synchronization mode that integrates time step-driven and event-triggered synchronization. This ensures the timing consistency of the global simulation through fixed-step forced synchronization, while also taking into account the real-time nature of adversarial actions through event-triggered emergency synchronization. Simultaneously, it adopts an incremental data synchronization mechanism that synchronizes only the changing incremental data. While ensuring the consistency of the simulation state, this significantly reduces the WAN bandwidth usage, making it suitable for application scenarios with limited WAN bandwidth.

[0017] C. This invention adopts a distributed execution architecture that decouples collaborative decision-making from local inference. It completes collaborative decision-making among multiple agents within a region through regional edge nodes, and terminal nodes complete autonomous inference and calculation of local agents. This significantly reduces the frequency of cross-WAN interaction and data transmission, effectively avoids the impact of WAN transmission latency on collaborative decision-making efficiency, and can support large-scale, wide-area distributed multi-agent collaborative adversarial inference.

[0018] D. This invention establishes a comprehensive fault-tolerant synchronization and conflict resolution mechanism. It achieves rapid resynchronization after link interruption through periodic state snapshots, ensures the continuity of synchronization services through dynamic primary / backup switching and redundant path switching, and achieves real-time correction of state conflicts through weighted fusion and consistency verification of multi-source situational data. It can effectively cope with various abnormal scenarios such as wide area network link fluctuations, interruptions, and node failures, ensure the continuous operation of simulation and simulation, and maintain the consistency and credibility of the global adversarial situation.

[0019] Other advantages, objectives and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination or study, or may be learned from the practice of the invention. Attached Figure Description

[0020] Figure 1 This invention presents a layered architecture for wide area network (WAN) multi-agent cooperative adversarial simulation based on WAN synchronization. Figure 2 This is a flowchart illustrating the distributed cooperative adversarial simulation execution process of the multi-agent cooperative adversarial simulation method based on wide area network synchronization according to the present invention. Figure 3 This is a schematic diagram of the fault-tolerant synchronization and conflict resolution mechanism of the multi-agent cooperative adversarial simulation method based on wide area network synchronization in this invention. Figure 4 This is a timing diagram of the global simulation fence synchronization and timeout fault tolerance mechanism of the multi-agent cooperative adversarial simulation method based on wide area network synchronization in this invention. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] It should be noted that the terms "vertical," "horizontal," "up," "down," "left," "right," and similar expressions used in this article are for illustrative purposes only and do not represent the only possible implementation.

[0023] 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 limit the invention; the term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0024] The following describes specific embodiments and appendices. Figure 1-4 The present invention will be described in further detail below.

[0025] This embodiment provides a multi-agent cooperative adversarial simulation method based on wide area network synchronization, applied to a cross-regional military unmanned swarm cooperative adversarial simulation scenario. The wide area network covers five geographical regions in China, with one edge computing node deployed in each geographical region. Each edge computing node has 20 subordinate terminal simulation nodes, and each terminal simulation node carries one instance of an unmanned swarm adversarial intelligent agent, totaling 100 simulated intelligent agents, divided into two adversarial camps: the red team and the blue team. The specific implementation steps are as follows: First, step S1 is executed to construct a multi-agent collaborative adversarial simulation layered architecture in a wide area network (WAN) environment. This layered architecture, from top to bottom, includes a global simulation control layer, a regional collaborative decision-making layer, and a local agent simulation and inference layer. The global simulation control layer is deployed on the core backbone nodes of the WAN, configured with two core servers that serve as backups for each other, and is responsible for the functions of global master clock source, global simulation control, and global situation verification. The regional collaborative decision-making layer is deployed on edge computing nodes in five geographical partitions, with each edge computing node corresponding to one geographical partition, and is responsible for the functions of regional clock source, regional collaborative decision-making, and regional state aggregation. The local agent simulation and inference layer is deployed on 100 terminal simulation nodes, with each terminal simulation node corresponding to one unmanned cluster simulation agent instance, and is responsible for the functions of agent local behavior inference, environmental interaction calculation, and local situation awareness.

[0026] Then, step S2 is executed to perform WAN hierarchical master-slave clock synchronization. The main core server of the global simulation control layer serves as the global master clock source, and the five regional edge computing nodes serve as regional slave clock sources. First, a first-level clock synchronization is performed. The five regional slave clock sources complete the link transmission delay measurement through a bidirectional timestamp exchange mechanism with the global master clock source via the WAN. An improved unscented Kalman filter algorithm is used to continuously and dynamically predict the delay jitter and packet loss rate of the WAN link. Based on the prediction results, feedforward delay compensation is performed on the clock synchronization signal. At the same time, deviation correction is performed for the asymmetric transmission delay of the uplink and downlink of the WAN, completing high-precision synchronization between the regional slave clock sources and the global master clock source, with the synchronization error controlled within 0.5ms. The first-level synchronization is complete. After completion, a two-level clock synchronization is implemented. Each region's slave clock source provides clock synchronization services to its subordinate terminal simulation nodes, using the same latency measurement and compensation mechanism to complete the synchronization between the terminal simulation nodes and their respective region's slave clock source. Ultimately, sub-millisecond-level clock reference unification is achieved for all simulation nodes, with the clock synchronization error of all nodes controlled within 1ms. At the same time, a dynamic master-slave switching mechanism is set up. When the master core server of the global master clock source fails or the link is interrupted, a new global master clock source is automatically elected from the backup core server to complete the seamless switching of clock synchronization services. When a slave clock source node in a certain region fails, its subordinate terminal simulation nodes are automatically switched to the adjacent normally operating regional slave clock source to ensure the continuity of clock synchronization services.

[0027] Next, step S3 is executed to establish a dual-loop synchronization control mechanism for multi-agent simulation. The inner loop is a clock synchronization loop, which, based on the hierarchical master-slave clock synchronization in step S2, continuously performs full-node clock synchronization verification and correction with a period of 10ms to maintain the consistency of the clock reference of all simulation nodes. The outer loop is a simulation state synchronization loop, which adopts a hybrid synchronization mode combining time step-driven and event-triggered methods. A globally unified minimum simulation step size of 50ms is set as the reference synchronization period. After each reference synchronization period, a forced synchronization of the critical state of all nodes is triggered. At the same time, an event trigger threshold is set. When the adversarial behavior instructions, local situational changes, and environmental interaction results of the simulated agent exceed the preset event trigger threshold, incremental emergency synchronization is immediately triggered without waiting for the reference synchronization period to arrive. The simulation state synchronization loop adopts an incremental data synchronization mechanism, which only synchronizes the incremental simulation state data, adversarial behavior instructions, and situational feature data that have changed within the current period, without synchronizing the full simulation state data, which greatly reduces the wide area network bandwidth usage. At the same time, a globally unified clock stamp is attached to all synchronized data, and the timing sorting and disorder correction of synchronized data are completed based on the clock stamp to ensure the correctness of data timing.

[0028] Next, step S4 is executed, performing a distributed cooperative adversarial simulation. This employs an execution architecture that decouples cooperative decision-making from local simulation. Each terminal simulation node, based on a unified local clock reference, autonomously completes the kinematic calculations, autonomous behavior decisions, virtual environment collision detection, and local battlefield situational awareness of its corresponding unmanned swarm intelligence agent, without waiting for instructions from upper-level nodes. The regional cooperative decision-making layer, based on the situational data of each node within the synchronized region, generates and distributes cooperative adversarial target instructions, task constraints, and cooperative rules for the red / blue team intelligence agents within its jurisdiction, without interfering with the specific behavior calculations of the simulated intelligence agents. The global simulation control layer completes... The system includes unified verification of the global adversarial situation, global step size control of the simulation process, and global execution of adversarial rules. It also features a global simulation fence synchronization mechanism, which releases the global simulation step size advancement command only after all edge computing nodes in the current baseline synchronization cycle have completed the reporting and verification of key states, thus preventing simulation timing errors. Additionally, a timeout tolerance mechanism is implemented, interpolating and completing the simulation process for unresponsive nodes that exceed a preset timeout threshold, using valid state data from the previous cycle to avoid blocking the continuous advancement of the global simulation process. During simulation, the global simulation control layer synchronizes the simulation process data of all nodes, completing the determination of victory or defeat for both sides and archiving the post-simulation data.

[0029] Finally, step S5 is executed to perform fault-tolerant synchronization and conflict resolution for WAN link fluctuations. First, a node periodic state snapshot mechanism is implemented. After each baseline synchronization cycle, each terminal simulation node generates a local simulation state snapshot and synchronizes it to the storage of its respective regional edge computing node. The regional edge computing node then aggregates the snapshots of its subordinate nodes and synchronizes them to the global simulation control layer for archiving. When the WAN link is interrupted and restored, the interrupted node completes local state rollback and resynchronization based on the most recent globally consistent state snapshot to avoid discontinuities in the simulation state. At the same time, a dynamic redundant synchronization path switching mechanism is set up. When the primary synchronization link experiences high packet loss, high latency, or interruption, it automatically switches to a preset redundant synchronization path. The system ensures stable transmission of synchronized data. To address conflicts in situational data reported by multiple nodes, it performs consistency verification and conflict resolution of multi-source situational data. The global simulation control layer and the regional collaborative decision-making layer verify the consistency of situational data reported by each node. When conflicts occur in situational data from different nodes, a weighted fusion is performed based on the clock stamp priority, node data reliability weight, and data acquisition accuracy level to generate globally consistent adversarial situational baseline data. Simultaneously, the corrected situational baseline data is synchronized to all relevant nodes to complete the consistency correction of the node's local state, ensuring the continuity, timing accuracy, and situational reliability of multi-agent collaborative adversarial simulation in a wide area network environment.

[0030] Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Anyone skilled in the art can make various modifications and alterations without departing from the spirit and scope of the present invention. Therefore, the scope of protection of the present invention should be determined by the claims.

Claims

1. A multi-agent cooperative adversarial simulation method based on wide area network synchronization, characterized in that, Includes the following steps: S1 constructs a multi-agent collaborative adversarial simulation layered architecture in a wide area network environment. The layered architecture includes, from top to bottom, a global simulation control layer, a regional collaborative decision-making layer, and a local agent simulation and inference layer. The global simulation control layer is deployed on the core backbone node of the wide area network, the regional collaborative decision-making layer is deployed on the edge computing nodes of each geographical partition, and the local intelligent agent simulation inference layer is deployed on each terminal simulation node. Each terminal simulation node carries at least one simulation intelligent agent instance. S2 performs WAN hierarchical master-slave clock synchronization, with the core node of the global simulation control layer as the global master clock source and the edge computing nodes of each region as the regional slave clock sources. After the regional slave clock sources complete the first-level clock synchronization with the global master clock source, they provide the second-level clock synchronization service for their subordinate terminal simulation nodes. Bidirectional transmission delay compensation is completed based on the dynamic prediction results of WAN link delay, and sub-millisecond level clock reference is unified for all simulation nodes; S3 establishes a dual-loop synchronization control mechanism for multi-agent simulation, with the inner loop being a clock synchronization loop. Based on the hierarchical master-slave clock synchronization in step S2, the clock reference consistency of all simulation nodes is continuously maintained. The outer ring is a simulation state synchronization ring, which adopts a hybrid synchronization mode that combines time step-driven and event-triggered methods to complete the wide area network distributed synchronization of multi-agent simulation state, adversarial behavior data and global adversarial situation data. S4 performs distributed collaborative adversarial simulation and inference. Each terminal simulation node completes the local behavior inference, environmental interaction physics calculation and local situational awareness of the corresponding simulated intelligent agent based on the unified local clock reference. Based on synchronized regional situational data, the regional collaborative decision-making layer generates and issues collaborative confrontation decision-making instructions for the multiple agents under its jurisdiction; the global simulation and control layer completes the unified verification of the global confrontation situation, the global step size control of the simulation process, and the global execution of the confrontation rules. S5 performs fault-tolerant synchronization and conflict resolution for wide area network link fluctuations. Through node periodic state snapshot mechanism, dynamic redundant synchronization path switching and multi-source situational data consistency verification, it completes real-time correction of simulation timing disorder and state conflict, ensuring the continuity, timing correctness and situational credibility of multi-agent cooperative adversarial simulation in wide area network environment.

2. The multi-agent cooperative adversarial simulation method based on wide area network synchronization according to claim 1, characterized in that, In step S2, the hierarchical master-slave clock synchronization specifically includes: using a bidirectional timestamp exchange mechanism to complete the link transmission delay measurement between master and slave nodes; using an improved unscented Kalman filter algorithm to continuously and dynamically predict the delay jitter and packet loss rate of the wide area network link; and using feedforward delay compensation for the clock synchronization signal based on the prediction results. At the same time, deviation correction is performed for the asymmetric transmission delay of the uplink and downlink of the wide area network to control the clock synchronization error of the fully simulated node within 1ms.

3. The multi-agent cooperative adversarial simulation method based on wide area network synchronization according to claim 1, characterized in that, In step S3, the hybrid synchronization mode combining time-step driven and event-triggered methods specifically refers to: Set a globally uniform minimum simulation step size as the baseline synchronization period, and trigger a full node critical state forced synchronization once after each baseline synchronization period ends. At the same time, an event trigger threshold is set. When the adversarial behavior command, local situational change, or environmental interaction result of the simulated intelligent agent exceeds the preset event trigger threshold, incremental emergency synchronization is immediately triggered without waiting for the baseline synchronization period to arrive.

4. The multi-agent cooperative adversarial simulation method based on wide area network synchronization according to claim 3, characterized in that, In step S3, the simulation state synchronization loop adopts an incremental data synchronization mechanism, which only synchronizes the incremental simulation state data, adversarial behavior instructions and situational characteristic data that have changed in the current cycle, and does not synchronize the full simulation state data; at the same time, a globally unified clock stamp is attached to all synchronized data, and the timing sorting and disorder correction of synchronized data are completed based on the clock stamp.

5. The multi-agent cooperative adversarial simulation method based on wide area network synchronization according to claim 1, characterized in that, In step S4, an execution architecture that decouples collaborative decision-making from local inference is adopted: the regional collaborative decision-making layer only issues the target instructions, task constraints and collaborative rules for collaborative confrontation to the terminal simulation nodes under its jurisdiction, without interfering with the specific behavior calculation of the simulated intelligent agent; after receiving the decision instructions, the terminal simulation nodes autonomously complete the kinematic calculation, behavior decision-making, environmental collision detection and local situational awareness of the intelligent agent locally, and only synchronize the key behavior results and situational data to the upper-level nodes through the wide area network.

6. The multi-agent cooperative adversarial simulation method based on wide area network synchronization according to claim 1, characterized in that, In step S4, the global simulation control layer sets up a global simulation fence synchronization mechanism. Only after all regional edge computing nodes have completed the reporting and verification of key states in the current benchmark synchronization cycle will the global simulation step size advancement command be released to avoid simulation timing errors. At the same time, a timeout fault tolerance mechanism is set up. For unresponsive nodes that exceed the preset timeout threshold, the valid state data of the previous cycle of the node is used for interpolation to complete the process, so as not to block the continuous advancement of the global simulation process.

7. The multi-agent cooperative adversarial simulation method based on wide area network synchronization according to claim 1, characterized in that, In step S5, the node periodic state snapshot mechanism is as follows: after each benchmark synchronization cycle ends, each terminal simulation node generates a local simulation state snapshot and synchronizes it to the storage of the edge computing node in its region. The edge computing node summarizes the snapshots of the nodes under its jurisdiction and synchronizes them to the global simulation control layer for archiving. When the WAN link is restored after an interruption, the interrupted node completes local state rollback and resynchronization based on the most recent globally consistent state snapshot, thus avoiding gaps in the simulation state.

8. The multi-agent cooperative adversarial simulation method based on wide area network synchronization according to claim 1, characterized in that, In step S5, the consistency verification and conflict resolution of the multi-source situational data specifically involves: the global simulation control layer and the regional collaborative decision-making layer performing consistency verification on the situational data reported by each node; when there is a conflict in the situational data of different nodes, the data is weighted and fused based on the clock stamp priority, node data credibility weight, and data acquisition accuracy level to generate globally consistent adversarial situational benchmark data. At the same time, the corrected situation baseline data is synchronized to all relevant nodes to complete the consistency correction of the local state of the nodes.

9. The multi-agent cooperative adversarial simulation method based on wide area network synchronization according to claim 1, characterized in that, The simulated intelligent agents include red team adversarial intelligent agents, blue team adversarial intelligent agents, and neutral environment intelligent agents. Each simulated intelligent agent belongs to a terminal simulation node in a different geographical region within the wide area network. The intelligent agents can complete cross-node and cross-regional collaborative interaction and trigger adversarial behavior through the wide area network. The global simulation control layer synchronizes the simulation process data of all nodes to complete the determination of the winner between the adversaries and the review data archiving of the entire simulation process.

10. The multi-agent cooperative adversarial simulation method based on wide area network synchronization according to claim 1, characterized in that, In step S2, the hierarchical master-slave clock synchronization also includes a dynamic master-slave switching mechanism: when the global master clock source node fails or the link is interrupted, a new global master clock source is automatically elected from the preset backup core nodes to complete the seamless switching of clock synchronization services. When a regional clock source node fails, its subordinate terminal simulation nodes are automatically switched to an adjacent, normally functioning regional clock source to ensure the continuity of clock synchronization services.

Images