A multi-agent encryption event-triggered control method and system, a terminal and a medium

Abstract

The application relates to the technical field of intelligent control, and discloses a multi-agent encrypted event triggering control method, a multi-agent encrypted event triggering control system, a terminal and a medium. The multi-agent encrypted event triggering control method comprises the following steps: constructing a lightweight model corresponding to each nonlinear multi-agent; acquiring state information corresponding to each nonlinear multi-agent; inputting first state information into a first model to obtain first triggering element information; respectively encrypting the first state information and the first triggering element information to obtain a first encrypted data packet; and obtaining second control information of a second agent according to the first encrypted data packet and second state information, so that the second agent performs collaborative control with the first agent according to the second control information. The application can improve the communication efficiency of the nonlinear multi-agent and enhance the security.

Description

Technical Field

[0001] This application relates to the field of intelligent control technology, and in particular to a multi-agent encrypted event triggering control method, system, terminal and medium. Background Technology

[0002] With the widespread application of distributed control and multi-agent systems (MASs) in smart grids, autonomous vehicle fleets, smart manufacturing, collaborative robots, and other scenarios, efficient communication and secure collaborative control among multiple agents have gradually become key issues.

[0003] Traditional time-triggered control (TTC) relies on a fixed sampling period for state updates. While easy to implement, it suffers from frequent communication and excessive resource consumption. Event-triggered control (ETC) has gradually emerged as a more efficient paradigm, significantly reducing communication burden by exchanging data only when trigger conditions are met. However, in nonlinear multi-agent systems, trigger conditions struggle to adapt flexibly to complex dynamics, making it difficult to balance performance and communication efficiency. Furthermore, information security issues are becoming increasingly prominent. In distributed environments, state and trigger information between agents can be vulnerable to attacks such as eavesdropping, replay attacks, tampering, or denial-of-service (DoS) attacks, seriously threatening system stability and reliability. Existing methods often employ traditional encryption algorithms, but these methods either incur excessive computational overhead, hindering real-time control, or prioritize data protection over the security of the trigger signals themselves, thus failing to fully defend against attacks such as traffic analysis.

[0004] Therefore, existing technologies still need to be improved and developed. Summary of the Invention

[0005] The main purpose of this application is to provide a multi-agent encrypted event-triggered control method, system, terminal and medium, which aims to solve the problem that the event-triggered control between nonlinear multi-agents in the prior art is difficult to flexibly adapt to different dynamics, resulting in the inability to balance communication efficiency and information security.

[0006] The first aspect of this application provides a multi-agent encryption event triggering control method, which includes the following steps:

[0007] Construct lightweight models corresponding to each of the nonlinear multi-agents, wherein the nonlinear multi-agents include a first agent and a second agent, and the lightweight model includes a first model corresponding to the first agent;

[0008] Obtain the state information corresponding to each of the nonlinear multi-agents, wherein the state information includes the first state information of the first agent and the second state information of the second agent;

[0009] The first state information is input into the first model to obtain the first trigger meta-information;

[0010] The first status information and the first trigger element information are encrypted respectively to obtain the first encrypted data packet;

[0011] Based on the first encrypted data packet and the second status information, the second control information of the second intelligent agent is obtained, so that the second intelligent agent can perform collaborative control with the first intelligent agent according to the second control information.

[0012] Optionally, in one embodiment of this application, the lightweight model further includes a second model corresponding to the second agent;

[0013] The multi-agent encryption event triggering control method further includes:

[0014] The second state information is input into the second model to obtain the second trigger meta-information;

[0015] The second status information and the second trigger element information are encrypted respectively to obtain a second encrypted data packet;

[0016] Based on the second encrypted data packet and the first status information, the first control information of the first intelligent agent is obtained, so that the first intelligent agent can perform collaborative control with the second intelligent agent according to the first control information.

[0017] Optionally, in one embodiment of this application, the first state information includes a first state vector, a first tracking error, and first neighbor summary information, and the first trigger meta-information includes a first trigger score and a first trigger time;

[0018] The step of inputting the first state information into the first model to obtain the first trigger meta-information specifically includes:

[0019] The first state vector, the first tracking error, and the first neighbor summary information are input into the first model to obtain the first trigger score;

[0020] If the first trigger score meets the triggering condition, then the first trigger moment is generated;

[0021] If the first trigger score does not meet the triggering conditions, then the first trigger moment will not be generated.

[0022] Optionally, in one embodiment of this application, the first state vector includes the state vector matrix of the first agent, and the first neighbor summary information includes the parameter matrix of the communication graph of the first agent;

[0023] The expression for the first model is:

[0024] ;

[0025] in, The first trigger score; Representative parameters of the large model calculation process; Let be the state vector matrix of the first agent. The first tracking error of the first intelligent agent; , This is the parameter matrix of the communication graph of the first agent.

[0026] Optionally, in one embodiment of this application, the first encrypted data packet includes encrypted first state information and masked first trigger metadata;

[0027] The step of encrypting the first state information and the first trigger element information to obtain the first encrypted data packet specifically includes:

[0028] The first state information is encrypted using a symmetric encryption method to obtain the encrypted first state information;

[0029] The first trigger element information is then subjected to secondary protection encryption to obtain the masked first trigger element information.

[0030] Optionally, in one embodiment of this application, the second control information includes a second control law;

[0031] The step of obtaining the second control information of the second intelligent agent based on the first encrypted data packet and the second state information specifically includes:

[0032] The encrypted first state information is decrypted using a symmetric key to obtain the decrypted first state information, and the masked first trigger element information is reverse-processed to obtain the recovered first trigger element information.

[0033] Based on the second state information, state estimation is performed to obtain the second state estimation data of the second agent;

[0034] The decrypted first state information, the recovered first trigger meta-information, and the second state estimation data are weighted and fused to obtain fused data.

[0035] Based on the fused data and the recovered first trigger meta-information, a second control law for the second agent is generated.

[0036] Optionally, in one embodiment of this application, the step of obtaining the second control information of the second intelligent agent based on the first encrypted data packet and the second state data, so that the second intelligent agent can perform collaborative control with the first intelligent agent according to the second control information, further includes:

[0037] Acquire local data of nonlinear multi-agent systems and generate update parameters based on the local data;

[0038] The global model is updated based on the updated parameters to obtain the updated global model parameters;

[0039] The first model and the second model are updated based on the updated global model parameters to obtain the updated first model and the second model.

[0040] A second aspect of this application also provides a multi-agent encryption event triggering control system, wherein the multi-agent encryption event triggering control system is applied to the multi-agent encryption event triggering control method described in any of the above solutions; the multi-agent encryption event triggering control system includes:

[0041] A lightweight model building module is used to build lightweight models corresponding to each of the nonlinear multi-agents, wherein the nonlinear multi-agents include a first agent and a second agent, and the lightweight model includes a first model corresponding to the first agent.

[0042] A state information acquisition module is used to acquire the state information corresponding to each of the nonlinear multi-agents, wherein the state information includes the first state information of the first agent and the second state information of the second agent.

[0043] An edge reasoning module is used to input the first state information into the first model to obtain the first trigger meta-information;

[0044] The encryption and communication module is used to encrypt the first status information and the first trigger element information respectively to obtain the first encrypted data packet;

[0045] The control execution and observation module is used to obtain the second control information of the second intelligent agent based on the first encrypted data packet and the second state information, so that the second intelligent agent can perform collaborative control with the first intelligent agent based on the second control information.

[0046] A third aspect of this application also provides a terminal, wherein the terminal includes: a memory, a processor, and a multi-agent encryption event triggering control program stored in the memory and executable on the processor, wherein when the multi-agent encryption event triggering control program is executed by the processor, it implements the steps of the multi-agent encryption event triggering control method as described above.

[0047] A fourth aspect of this application also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a multi-agent encryption event triggering control program, which, when executed by a processor, implements the steps of the multi-agent encryption event triggering control method as described above.

[0048] Beneficial effects: This application provides a multi-agent encrypted event triggering control method, system, terminal and medium. This application generates corresponding trigger meta-information by deploying a lightweight model on each agent, and encrypts it to communicate with other agents, so that each agent executes control commands to achieve collaborative control. In this way, this application can improve the communication efficiency of nonlinear multi-agents (reduce unnecessary communication), enhance security (defend against eavesdropping, replay, tampering and DoS attacks), and ensure robustness (maintain system stability under nonlinear interference and malicious attacks). Attached Figure Description

[0049] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0050] Figure 1 This is the overall architecture diagram of the multi-agent encrypted event triggering control system of this application;

[0051] Figure 2 A flowchart illustrating a preferred embodiment of the multi-agent encryption event triggering control method of this application;

[0052] Figure 3 This is an experimental result diagram of a preferred embodiment of the multi-agent encryption event triggering control method of this application;

[0053] Figure 4 This is a graph showing the agent consistency error in a preferred embodiment of the multi-agent encryption event triggering control method of this application.

[0054] Figure 5 This is a diagram showing the agent position curves in a preferred embodiment of the multi-agent encryption event triggering control method of this application.

[0055] Figure 6 This is a structural diagram of a preferred embodiment of the multi-agent encrypted event triggering control system of this application;

[0056] Figure 7 This is a structural diagram of a preferred embodiment of the terminal of this application.

[0057] Explanation of reference numerals in the attached figures:

[0058] 100. Lightweight Model Building Module; 200. State Information Acquisition Module; 300. Edge Reasoning Module; 400. Encryption and Communication Module; 500. Control Execution and Observation Module. Detailed Implementation

[0059] To make the objectives, technical solutions, and effects of this application clearer and more explicit, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. The described embodiments are only possible technical implementations of this application and not all possible implementations. Based on the embodiments in this application, those skilled in the art can obtain other embodiments without creative effort, and these embodiments are also within the protection scope of this application.

[0060] First, let's introduce the terms used in the embodiments of this application:

[0061] Large Language Model (LLM) refers to a large-scale artificial intelligence model based on the Transformer architecture, which has strong context modeling and uncertainty estimation capabilities. In this solution, a lightweight version is generated through knowledge distillation / pruning / quantization and deployed on edge nodes.

[0062] Federated learning is a distributed machine learning framework that allows multiple agents to train models locally and securely aggregate parameters while protecting data privacy.

[0063] Secure Aggregation, combined with differential privacy technology, protects the model parameters uploaded by each agent from being leaked or reverse-analyzed during federated learning.

[0064] Differential privacy enhances privacy protection by adding noise to the data to ensure that individual data cannot be identified in statistical analysis.

[0065] A replay attack is an attack method in which an attacker intercepts and repeatedly sends legitimate communication data to deceive the system or tamper with control commands.

[0066] Traffic analysis is the process of analyzing the frequency, patterns, and other characteristics of network communication to infer system behavior or attack methods for sensitive information.

[0067] Trigger Score is a dynamic evaluation value generated by the large model based on state error, attention distribution, and uncertainty estimation, used for adaptive triggering decisions.

[0068] Dynamic Threshold is a trigger condition threshold that is adjusted in real time according to the system status, balancing communication efficiency and control performance.

[0069] Lightweight models are large models compressed using techniques such as distillation, pruning, and quantization, making them suitable for real-time inference on edge devices.

[0070] Attention mechanism, a core component of deep learning, focuses on key information by assigning different weights, thereby improving the flexibility of model decision-making.

[0071] Uncertainty estimation quantifies the reliability of model predictions and is used to detect anomalous behavior or attacks.

[0072] Robust control is a control strategy that maintains system stability even in the presence of interference or attacks.

[0073] Knowledge distillation is a technique that transfers knowledge from large models to smaller models, preserving core capabilities while reducing computational overhead.

[0074] The Secure Aggregation Center is the central node in federated learning responsible for securely aggregating the parameters of each agent, and it uses differential privacy technology to prevent data leakage.

[0075] DoS, Denial of Service.

[0076] EKF stands for Extended Kalman Filter.

[0077] UKF stands for Unscented Kalman Filter.

[0078] Because large-scale artificial intelligence models demonstrate powerful context modeling and uncertainty estimation capabilities in sequence modeling, this application introduces them into the control field to achieve smarter and more flexible event-triggered decision-making. It combines large-scale models with multi-agent event-triggered control while simultaneously meeting the requirements of low communication overhead and high security. This application can simultaneously improve the system's communication efficiency, security, and robustness.

[0079] To address the problem that event-triggered control in nonlinear multi-agent communication is difficult to adapt flexibly to different dynamics, resulting in an inability to balance communication efficiency and information security, this application uses a lightweight model deployed on each agent to generate corresponding trigger metadata, which is then encrypted and used to communicate with other agents. This enables each agent to execute control commands and achieve collaborative control. Consequently, this application can improve the communication efficiency of nonlinear multi-agent communication (reducing unnecessary communication), enhance security (defending against eavesdropping, replay, tampering, and DoS attacks), and ensure robustness (maintaining system stability under nonlinear interference and malicious attacks).

[0080] See Figure 1 First, the system architecture of the embodiments of this application will be introduced.

[0081] Cloud-based training and distillation module: Train large models (such as the Transformer architecture) using global data in the cloud, and generate lightweight models through knowledge distillation, pruning and quantization, and distribute them to various edge agents.

[0082] Edge inference module: Deployed on each agent node, it takes as input the current state, tracking error, and neighbor summary information, and outputs a trigger score. The edge model is a lightweight large model that uses attention mechanisms and uncertainty estimation to intelligently determine triggers.

[0083] Encryption and Communication Module: When the triggering conditions are met, the agent encrypts the state information and triggering metadata and sends them to the neighboring nodes. The state data is encrypted using symmetric encryption, and the triggering metadata hides the actual triggering pattern through masking or random delay.

[0084] Control execution and observer module: After the receiving end decrypts the information, it is input into the local observer. The observer integrates the neighbor data and generates the control law, thereby realizing multi-agent cooperative control.

[0085] Federated learning and secure aggregation module: Each agent periodically or event-driven uploads model update parameters, which are then merged in the cloud or distributed aggregation center through secure aggregation, and then distributed to the edge through distillation to achieve adaptive optimization of the model.

[0086] This application, for the first time, introduces the attention mechanism and uncertainty estimation of large models into event-triggered control, enhancing the flexibility and intelligence of triggering decisions. It proposes a two-layer encryption strategy combining event trigger signal encryption and state data encryption to effectively defend against traffic analysis and information leakage. Furthermore, it integrates federated learning and secure aggregation, enabling the multi-agent system to adaptively evolve in a distributed environment and possess strong privacy protection capabilities. Therefore, this application not only significantly reduces the communication overhead of multi-agent systems but also maintains the system's security and robust operation under nonlinear interference and malicious attacks, demonstrating high theoretical value and application prospects.

[0087] The technical solutions of this application will be described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0088] The multi-agent encryption event triggering control method described in the preferred embodiment of this application, such as... Figure 2 As shown, the multi-agent encryption event triggering control method includes the following steps:

[0089] In step S101, lightweight models corresponding to each nonlinear multi-agent are constructed, wherein the nonlinear multi-agent includes a first agent and a second agent, and the lightweight model includes a first model corresponding to the first agent.

[0090] It should be noted that this application utilizes a lightweight large model to model the state, error, and neighbor information of the agent, and outputs a trigger score, thereby achieving adaptive triggering decisions based on prediction error, attention distribution, and uncertainty estimation, overcoming the limitations of traditional fixed threshold methods. Lightweight symmetric encryption is applied to the agent state data, and event trigger signals (trigger time and trigger score) are encrypted or randomly masked to prevent attackers from inferring control strategies through traffic analysis. The edge large model parameters are updated periodically or event-driven through a federated learning mechanism, improving model adaptability while protecting the privacy and communication security of each agent during training by combining secure aggregation and differential privacy technologies. The uncertainty detection capability of the large model is used to identify potential anomalies or attack behaviors, and the triggering strategy is automatically adjusted or a backup control law is switched when necessary, ensuring the system remains stable even under attack.

[0091] Specifically, based on the Transformer architecture (such as self-attention and multi-head attention mechanisms), self-supervised pre-training is performed using large-scale unlabeled data to form a general feature representation. After pre-training, supervised fine-tuning is used to adapt it to specific tasks. Knowledge distillation is performed to combine the "soft labels" of the large model with the real labels to train a lightweight model. Model pruning is then performed, evaluating parameter importance based on weight magnitude, gradient, or activation frequency, using an iterative cycle of training-pruning-fine-tuning. The cloud-based training, distillation, pruning, and quantization processes form a closed loop, ensuring that the model outperforms related technologies in terms of communication efficiency, security, and robustness. This application generates a lightweight model adapted to edge devices through the synergistic effect of pre-training, distillation, pruning, and quantization, enabling efficient, secure, and stable collaborative control of multi-agent systems.

[0092] In step S102, the state information corresponding to each of the nonlinear multi-agents is obtained, wherein the state information includes the first state information of the first agent and the second state information of the second agent.

[0093] Specifically, the state information includes: the agent's real-time dynamic state, such as physical quantities like position, velocity, acceleration, and attitude; tracking error (such as the deviation between the actual state and the desired state), used to measure the system's control effectiveness; and neighbor summary information, i.e., state summaries of neighboring agents, used for distributed cooperative control. Its sources are the agent's own sensor data, the current state fed back by the control module, and neighbor state summaries received through the communication module. In the event-triggered decision-making process, the state information serves as input to the lightweight model.

[0094] In step S103, the first state information is input into the first model to obtain the first trigger element information.

[0095] In one possible implementation, the first state information includes a first state vector, a first tracking error, and first neighbor summary information, and the first trigger meta-information includes a first trigger score and a first trigger time. The first state vector, the first tracking error, and the first neighbor summary information are input into the first model to obtain a first trigger score; if the first trigger score meets the trigger condition, a first trigger time is generated; if the first trigger score does not meet the trigger condition, a first trigger time is not generated.

[0096] Specifically, the current state is the agent's real-time state vector (such as position, velocity, and acceleration), collected by sensors or control modules and normalized to ensure numerical stability; the tracking error is the deviation between the actual state and the desired state (such as position deviation in consistency control), used to quantify the control effect and drive the triggering of score calculation; the neighbor summary information is the state summary of neighboring agents. The input data is integrated into a unified feature vector through feature concatenation or attention-weighted fusion, and then input into the lightweight large model.

[0097] Furthermore, during the trigger determination process, the edge agent inputs the state vector, error, and neighbor summary into the lightweight large model. The model uses an attention mechanism to analyze the importance distribution of the input information and calculates the trigger score based on the prediction error and uncertainty.

[0098] ;

[0099] in, To trigger the rating, Represents parameters in the large model calculation process. For a multi-agent state matrix, For multi-agent tracking error, , This is the parameter matrix of the multi-agent communication graph. If the trigger score is greater than the dynamic threshold, a trigger signal is generated and the encrypted communication process begins; otherwise, silence is maintained to reduce communication overhead.

[0100] ;

[0101] in, To be at the moment the event is triggered Then, when the trigger score reaches the trigger threshold The time when the next event will be triggered.

[0102] Understandably, the threshold is dynamically adjusted through an adaptive algorithm; the output trigger score (numerical result, such as 0.8) and trigger signal (binary flag, such as 1 for triggering, 0 for silence) are used by the encryption and communication module to determine whether to encrypt and transmit state information and trigger metadata; the trigger signal is used to control the execution and observer module to initiate control law generation. When the trigger signal is 1, the encryption and communication module encrypts the state information (symmetric encryption) and trigger metadata (mask / random delay) and sends it to neighboring nodes to achieve secure communication. After the receiving end decrypts the information, the local observer fuses neighbor data (such as neighbor state summaries) with the local state to generate a control law, driving the agent to adjust its state and achieving multi-agent cooperative control.

[0103] In step S104, the first status information and the first trigger element information are encrypted respectively to obtain the first encrypted data packet.

[0104] In one possible implementation, the first encrypted data packet includes encrypted first state information and masked first trigger metadata. The first state information is encrypted using a symmetric encryption method to obtain encrypted first state information; the first trigger metadata is then subjected to secondary protection encryption to obtain masked first trigger metadata.

[0105] It should be noted that a lightweight symmetric encryption algorithm is used for the agent state data to ensure real-time performance; the trigger signal is protected by a second layer of protection, using methods such as random masking and spurious trigger insertion to prevent attackers from inferring system patterns through traffic analysis; and during the model parameter update phase, differential privacy and security aggregation technology is used to ensure that the data privacy of each agent is not leaked during model training.

[0106] Understandably, lightweight symmetric encryption algorithms are employed due to their high computational efficiency, good real-time performance, and suitability for the computing power limitations of edge devices. The actual triggering patterns are concealed through random masking (e.g., adding random noise), spurious trigger insertion (simulating trigger events to confuse the actual patterns), or random delays (e.g., 0-100ms random delays). Neighbor node identification and communication path planning are performed based on the multi-agent communication graph parameter matrix. , The system determines the set of neighboring nodes, ensuring that data is transmitted only to authorized neighbors. It dynamically selects low-latency, high-bandwidth communication links, prioritizing the transmission of critical data based on quality of service policies, ensuring timely delivery of encrypted data packets. Encrypted status information and trigger metadata are transmitted to neighboring nodes through a secure channel to prevent man-in-the-middle attacks.

[0107] In step S105, the second control information of the second agent is obtained based on the first encrypted data packet and the second status information, so that the second agent can perform collaborative control with the first agent based on the second control information.

[0108] In one possible implementation, the second control information includes a second control law. The encrypted first state information is decrypted using a symmetric key to obtain decrypted first state information. The masked first trigger meta-information is then reverse-processed to obtain recovered first trigger meta-information. State estimation is performed based on the second state information to obtain second state estimation data for the second agent. The decrypted first state information, the recovered first trigger meta-information, and the second state estimation data are then weighted and fused to obtain fused data. Finally, the second control law for the second agent is generated based on the fused data and the recovered first trigger meta-information.

[0109] Understandably, the receiving end first decrypts the state information (symmetric encryption and decryption), then parses the trigger metadata (removing masks / delays), and inputs it into the local observer. The observer merges the decrypted data from neighbors (such as neighbor state summaries) with the local state to generate a control law, which drives the agent to adjust its state, thus achieving multi-agent cooperative control.

[0110] It should be noted that the model uses uncertainty estimation and attention distribution to detect potential abnormal patterns. Once a DoS attack, replay attack, or data tampering is detected, the system will immediately issue an alarm. The control module will automatically switch to a robust backup control law according to the level of abnormality, ensuring that the multi-agent system can maintain basic stability and consistency even in malicious environments.

[0111] It should be noted that local observer state estimation uses a dynamic system model and filtering algorithms to perform real-time and accurate estimation and updating of the agent's current state (such as position, velocity, acceleration, etc.). Specifically, the state vector (such as position and velocity), tracking error (the deviation between the actual state and the desired state), and neighbor summary information (the mean and variance of the states of neighboring nodes, etc.) are the observer's input data, used to correct the accuracy of the state estimation. State estimation is achieved using filtering algorithms and model-driven approaches. An extended Kalman filter (EKF) or an unscented Kalman filter (UKF) is used, combined with a system dynamics model (such as nonlinear differential equations) and sensor noise characteristics, to perform real-time state estimation. For example, in the prediction step, based on the state estimate from the previous time step and the control input, the current state is predicted using the state transition equation; in the update step, combined with the current sensor measurements (such as position and velocity), the residual is calculated using the observation equation, and the predicted value is corrected to obtain the final state estimate. The uncertainty (such as variance) of the state estimate is estimated using Bayesian deep learning or Monte Carlo methods, providing a confidence index for triggering score calculation and control law generation.

[0112] Specifically, in the decryption and reverse processing, the encrypted first state information is decrypted using a symmetric key to obtain the decrypted first state information (such as the neighbor node state matrix and tracking error); the masked first trigger metadata (trigger score and trigger time) is reverse-processed (such as removing the random mask and correcting the delay) to restore the original trigger metadata. In the local state estimation and data fusion process, local state estimation is performed first. The second agent performs local state estimation based on the second state information (its own state vector, tracking error, and neighbor summary) to generate second state estimation data (such as estimated values ​​of position and velocity); then, neighbor state fusion is performed, where the decrypted first state information (neighbor state data) and the second state estimation data are weighted and fused to form fused data. For example, neighbor state data (such as position and velocity) is used as external input to correct the accuracy of local state estimation, and the fusion weights can be dynamically adjusted according to the trigger score (such as increasing the neighbor state weight when the score is high to enhance the response to highly dynamic environments); then, trigger metadata fusion is performed, where the restored trigger score and trigger time are incorporated into the fused data to form ternary fused data (fused data of neighbor state, local state estimation, and trigger metadata). During the global consistency error calculation process, based on the ternary fusion data and the state data of all agents, the global consistency error (such as the sum of position deviations and velocity differences) is calculated through a consistency protocol (such as Laplace matrix aggregation). The trigger time is used to coordinate the synchronous control of multiple agents (such as ensuring that all agents execute the control law at the same time point). During the control law generation and anomaly detection process, a second control law for the second agent is generated based on the ternary fusion data, global consistency error, and trigger meta-information. For example, the trigger score directly adjusts the aggressiveness of the control law, and the trigger time ensures that the control law is triggered at the correct time. Anomaly detection and robust control are performed in the above process. If the trigger meta-information shows an abnormal pattern (such as frequent triggering or abnormal timing), an alarm is triggered and the system switches to a robust backup control law to resist DoS attacks or data tampering.

[0113] It is worth noting that the decrypted first state information is the real-time state data of neighboring nodes, which is an important input for the local observer's state estimation. Its fusion with the local state estimation data ensures the accuracy and real-time performance of the state estimation, forming the basis for control law generation. Trigger meta-information has a synergistic effect; trigger scores and trigger times not only drive communication and control law generation but also optimize the system's response performance and stability in dynamic environments through dynamic weight adjustments and timing coordination. Therefore, this application ensures the synergistic effect of the decrypted neighbor state information, local state estimation data, and trigger meta-information, achieving efficient, safe, and stable operation of multi-agent cooperative control.

[0114] In one possible implementation, the lightweight model further includes a second model corresponding to the second agent. The second state information is input into the second model to obtain second trigger meta-information; the second state information and the second trigger meta-information are encrypted respectively to obtain a second encrypted data packet; based on the second encrypted data packet and the first state information, first control information for the first agent is obtained, so that the first agent can perform collaborative control with the second agent according to the first control information.

[0115] It should be noted that each agent independently generates its control law through a local observer. The receiving end (i.e., neighboring agents) decrypts the state information and trigger metadata, inputs the data into its local observer, and fuses local state (such as position, velocity, and tracking error) with neighbor data (such as state summaries and trigger information from neighboring nodes) to generate its own control law. The generation of the control law depends on the global consensus objective of the multi-agent system.

[0116] This application achieves global consistency control and completes multi-agent collaborative control through bidirectional data interaction (the trigger meta-information and state information of the first intelligent agent are exchanged with the second intelligent agent, and vice versa).

[0117] In one possible implementation, local data of a nonlinear multi-agent system is acquired, and update parameters are generated based on the local data; the global large model is updated based on the update parameters to obtain updated global model parameters; the first model and the second model are updated based on the updated global model parameters to obtain updated first and second models.

[0118] Specifically, each agent trains a model using local data (such as state vectors, tracking errors, and neighbor summaries), generating update parameters (such as gradients and weight shifts). The training utilizes a lightweight large model with an edge inference module, combining attention mechanisms and uncertainty estimation to optimize the update strategy. Locally generated parameters undergo security processing (such as masking and encryption) before uploading. The agent uploads encrypted parameters, and the server performs aggregation (such as additive averaging) in the encrypted state, obtaining a global update after decryption. Noise is added during parameter uploading or aggregation to ensure that the presence or absence of a single data record has a quantifiable and minimal impact on the aggregation result. The server updates the global model parameters based on the secure aggregation result, combining weighted averaging or personalized aggregation strategies (such as horizontal / vertical federation) from federated learning to optimize model performance. The global model undergoes knowledge distillation, pruning, and quantization to generate a lightweight version, which is then distributed to the edge agents. Training is stopped based on a loss function threshold, the number of iterations, or model performance metrics (such as tracking error convergence speed). When the global consistency error converges to a preset range within a finite time, a convergence stop is triggered.

[0119] In simulation verification, this application employs a typical nonlinear multi-agent consensus control scenario for testing. Experimental results show that compared to time-triggered control, the number of communications can be reduced by 75%. (See [link to relevant documentation]). Figure 3 (2000×4-644-388-524-420) / (2000×4)=75.3%, significantly reducing bandwidth consumption; when subjected to eavesdropping and traffic analysis attacks, attackers cannot effectively infer the agent's state and triggering patterns; under DoS attack environments, the method of this application can still ensure that the system error converges to the preset range within a finite time. See [link to relevant documentation]. Figure 4 In the consensus error curve diagram, the consensus error of the four multi-agents is represented. The error converges to the bound of [-0.1, 0.1]. See [link to diagram]. Figure 5 The agent's position curve is the agent's consistent tracking position information; the error is calculated based on the position. Figure 4 and Figure 5 It can be seen that the control effect of the embodiment of this application is good and the system converges.

[0120] The method proposed in this application is superior to existing technologies in terms of communication efficiency, security, and robustness, and has high application value and promotion significance.

[0121] Next, referring to the accompanying drawings, a multi-agent encryption event triggering control system proposed according to the embodiments of this application is described, which is applied to the multi-agent encryption event triggering control method described in any one of the above schemes.

[0122] Figure 6 This is a structural diagram of the multi-agent encrypted event triggering control system according to an embodiment of this application.

[0123] like Figure 6 As shown, the multi-agent encrypted event triggering control system includes: a lightweight model construction module 100, a state information acquisition module 200, an edge reasoning module 300, an encryption and communication module 400, and a control execution and observation module 500.

[0124] Specifically, the lightweight model construction module 100 is used to construct lightweight models corresponding to each of the nonlinear multi-agents, wherein the nonlinear multi-agents include a first agent and a second agent, and the lightweight model includes a first model corresponding to the first agent.

[0125] The state information acquisition module 200 is used to acquire the state information corresponding to each of the nonlinear multi-agents, wherein the state information includes the first state information of the first agent and the second state information of the second agent.

[0126] The edge reasoning module 300 is used to input the first state information into the first model to obtain the first trigger meta-information;

[0127] The encryption and communication module 400 is used to encrypt the first status information and the first trigger element information respectively to obtain the first encrypted data packet;

[0128] The control execution and observation module 500 is used to obtain the second control information of the second intelligent agent based on the first encrypted data packet and the second status information, so that the second intelligent agent can perform collaborative control with the first intelligent agent based on the second control information.

[0129] Figure 7 A structural diagram of a terminal provided in an embodiment of this application. The terminal may include:

[0130] The memory 501, the processor 502, and the computer program stored on the memory 501 and capable of running on the processor 502.

[0131] When the processor 502 executes the program, it implements the multi-agent encryption event triggering control method provided in the above embodiments.

[0132] Furthermore, the terminal also includes:

[0133] Communication interface 503 is used for communication between memory 501 and processor 502.

[0134] The memory 501 is used to store computer programs that can run on the processor 502.

[0135] Memory 501 may include high-speed RAM memory, and may also include non-volatile memory, such as at least one disk storage device.

[0136] If the memory 501, processor 502, and communication interface 503 are implemented independently, then the communication interface 503, memory 501, and processor 502 can be interconnected via a bus to complete communication between them. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EIS) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of representation, Figure 7 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0137] Optionally, in a specific implementation, if the memory 501, processor 502, and communication interface 503 are integrated on a single chip, then the memory 501, processor 502, and communication interface 503 can communicate with each other through an internal interface.

[0138] Processor 502 may be a central processing unit (CPU), an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of this application.

[0139] This embodiment also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the multi-agent encrypted event triggering control method described above.

[0140] One embodiment of this application provides a computer program product, including a computer program that, when executed by a processor, implements the features described in this application. Figure 1 The multi-agent encryption event triggering control method provided in any of the corresponding embodiments.

[0141] In the description of this specification, the 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. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. 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 different embodiments or examples.

[0142] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "N" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0143] Any process or method described in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or N executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0144] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable storage medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable storage medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable storage media include: an electrical connection having one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Alternatively, the computer-readable storage medium could be paper or other suitable media on which the program can be printed, since the program can be obtained electronically by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in a computer memory.

[0145] It should be understood that the various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, the N steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0146] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

[0147] Furthermore, the functional units in the various embodiments of this application can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.

[0148] The storage medium mentioned above can be a read-only memory, a disk, or an optical disk, etc. Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of this application.

[0149] It should be understood that the application of this application is not limited to the examples above. Those skilled in the art can make improvements or modifications based on the above description, and all such improvements and modifications should fall within the protection scope of the appended claims.

[0150] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

Claims

1. A nonlinear multi-agent cryptographic event-triggered control method, characterized in that, The nonlinear multi-agent encryption event-triggered control method comprises the following steps: A lightweight model corresponding to each of the nonlinear multi-agents is constructed, wherein the nonlinear multi-agents comprise a first agent and a second agent, and the lightweight model comprises a first model corresponding to the first agent; State information corresponding to each of the nonlinear multi-agents is acquired, wherein the state information comprises first state information of the first agent and second state information of the second agent; The first state information is input into the first model to obtain first trigger element information; The first state information and the first trigger element information are respectively encrypted to obtain a first encrypted data packet; Second control information of the second agent is obtained according to the first encrypted data packet and the second state information, so that the second agent cooperates with the first agent according to the second control information; The lightweight model further comprises a second model corresponding to the second agent; The nonlinear multi-agent encryption event-triggered control method further comprises the following steps: The second state information is input into the second model to obtain second trigger element information; The second state information and the second trigger element information are respectively encrypted to obtain a second encrypted data packet; First control information of the first agent is obtained according to the second encrypted data packet and the first state information, so that the first agent cooperates with the second agent according to the first control information; The first encrypted data packet comprises encrypted first state information and masked first trigger element information; The first state information and the first trigger element information are respectively encrypted to obtain a first encrypted data packet, specifically comprising the following steps: The first state information is encrypted by using a symmetric encryption method to obtain encrypted first state information; The first trigger element information is encrypted for secondary protection to obtain masked first trigger element information; The second control information comprises a second control law; The second control information of the second agent is obtained according to the first encrypted data packet and the second state information, specifically comprising the following steps: The encrypted first state information is decrypted by using a symmetric key to obtain decrypted first state information, and the masked first trigger element information is inversely processed to obtain restored first trigger element information; Second state estimation data of the second agent is obtained according to state estimation of the second state information; The decrypted first state information, the restored first trigger element information and the second state estimation data are weighted and fused to obtain fused data; The second control law of the second agent is generated according to the fused data and the restored first trigger element information.

2. The nonlinear multi-agent cryptographic event-triggered control method according to claim 1, wherein, The first state information comprises a first state vector, a first tracking error and first neighbor summary information, and the first trigger element information comprises a first trigger score and a first trigger time; The first state information is input into the first model to obtain first trigger element information, specifically comprising the following steps: inputting the first state vector, the first tracking error and the first neighbor summary information into the first model to obtain a first trigger score; if the first trigger score meets a trigger condition, generating a first trigger time; if the first trigger score does not meet the trigger condition, not generating the first trigger time.

3. The nonlinear multi-agent cryptographic event-triggered control method according to claim 2, wherein, The first state vector includes a state vector matrix of the first agent, and the first neighbor summary information includes a parameter matrix of a first agent communication graph. An expression of the first model is: ； wherein, is a first trigger score; represents a large model computation process parameter; is a state vector matrix of the first agent, is a first tracking error of the first agent; , is a parameter matrix of the first agent communication graph.

4. The nonlinear multi-agent cryptographic event-triggered control method of claim 1, wherein, The second agent is controlled according to the second control information, and the first agent is controlled according to the first control information. obtaining local data of the nonlinear multi-agent, and generating an update parameter according to the local data; updating a global model according to the update parameter to obtain an updated global model parameter; updating the first model and the second model according to the updated global model parameter to obtain updated first and second models.

5. A nonlinear multi-agent cryptographic event-triggered control system, characterized in that, The nonlinear multi-agent encryption event-triggered control system is used to implement the nonlinear multi-agent encryption event-triggered control method of any one of claims 1-4. The nonlinear multi-agent encryption event-triggered control system includes: a lightweight model construction module configured to construct a lightweight model corresponding to each of the nonlinear multi-agents, wherein the nonlinear multi-agents include a first agent and a second agent, and the lightweight model includes a first model corresponding to the first agent; a state information acquisition module configured to acquire state information corresponding to each of the nonlinear multi-agents, wherein the state information includes first state information of the first agent and second state information of the second agent; an edge inference module configured to input the first state information into the first model to obtain first trigger information; an encryption and communication module configured to encrypt the first state information and the first trigger information respectively to obtain a first encrypted data packet; a control execution and observation module configured to obtain second control information of the second agent according to the first encrypted data packet and the second state information, so that the second agent performs collaborative control with the first agent according to the second control information.

6. A terminal, characterized by comprising: The terminal includes a memory, a processor, and a nonlinear multi-agent encryption event-triggered control program stored on the memory and executable on the processor, and the nonlinear multi-agent encryption event-triggered control program implements the steps of the nonlinear multi-agent encryption event-triggered control method of any one of claims 1-4 when executed by the processor.

7. A computer readable storage medium characterized in that, The computer-readable storage medium stores a nonlinear multi-agent encryption event-triggered control program, and the nonlinear multi-agent encryption event-triggered control program implements the steps of the nonlinear multi-agent encryption event-triggered control method of any one of claims 1-4 when executed by the processor.

Images