Heterogeneous multi-agent system control method under DoS attack based on event triggering strategy
By using a Dynamic Threshold Event Triggered Preset Performance Control (DT-ETPPC) architecture, combined with radial basis function neural networks and Lyapunov theory, the cooperative control problem of heterogeneous multi-agent systems under DoS attacks is solved, achieving efficient communication and performance assurance in the attack environment.
Patent Information
- Application Number
- CN202510979659.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-16
- Publication Date
- 2026-02-03
AI Technical Summary
Existing ETC solutions are limited in performance and lack security defense mechanisms in heterogeneous multi-agent systems, making it difficult to meet the performance requirements of complex task scenarios under DoS attacks.
A dynamic threshold event-triggered preset performance control (DT-ETPPC) architecture is adopted, and a counter-attack controller is designed by combining radial basis function neural networks and Lyapunov theory. The communication frequency and preset performance constraints are optimized through an adaptive dynamic threshold strategy to ensure that the system meets the performance indicators under DoS attack.
Effectively respond to DoS attacks, reduce communication resource consumption, improve system stability and recovery capabilities, and ensure that the system meets task performance requirements under attack conditions.
Smart Images

Figure CN121454993A_ABST
Abstract
Description
Technical Field
[0001] This invention addresses the cooperative control problem of heterogeneous multi-agent systems under denial-of-service (DoS) attacks, proposing a control method for heterogeneous multi-agent systems under DoS attacks based on an event-triggered strategy. Background Technology
[0002] With the deep integration of information technology and control theory, heterogeneous multi-agent systems have demonstrated remarkable application value in key areas such as industrial automation, intelligent transportation, drone swarms, and distributed sensing networks. The core advantages of such systems lie in their distributed architecture and collaborative decision-making capabilities, characteristics that highly depend on reliable communication mechanisms between agents. However, in open network environments, denial-of-service (DoS) attacks, as a highly covert and destructive network threat, severely impair the collaborative performance and dynamic stability of systems by specifically interfering with or blocking communication channels. This security challenge is becoming increasingly prominent as the application scope of networked control systems expands, making it an urgent problem to ensure the reliable operation of equipment under attack environments.
[0003] Traditional cooperative control methods are mostly based on continuous-time or fixed-sampling control strategies. These methods achieve good control performance under normal communication conditions, but their performance deteriorates significantly when facing DoS attacks because they cannot effectively handle communication interruptions and information loss. Event-triggered control (ETC) offers a new perspective on combating DoS attacks due to its efficient use of communication resources. Compared to traditional time-triggered mechanisms, ETC only triggers communication and control updates when the system state meets specific conditions, thus significantly improving the efficiency of communication resource utilization. However, most existing event-triggered cooperative control methods do not consider preset performance constraints, making it difficult to meet the stringent performance requirements of complex task scenarios. For example, in intelligent transportation systems, vehicle platooning needs to maintain formation while ensuring that the safe distance and speed between vehicles meet specific performance indicators to avoid collisions. Summary of the Invention
[0004] To address the aforementioned challenges of limited performance in heterogeneous environments and the lack of security defense mechanisms in PPC methods, this invention proposes an innovative Dynamic Threshold Event-Triggered Prescribed Performance Control (DT-ETPPC) architecture. This architecture achieves, for the first time, a deep integration of event-triggered mechanisms and preset performance constraints. By introducing an adaptive dynamic threshold strategy, it not only effectively counters malicious interference from DoS attacks on communication links but also ensures that heterogeneous nodes still meet predetermined performance indicators under attack scenarios.
[0005] This invention provides a method for controlling a heterogeneous multi-agent system under a DoS attack based on an event-triggered strategy, comprising: Construct a multi-agent system, use radial basis function neural networks to approximate unknown nonlinear functions in the multi-agent system, and constrain system errors through preset performance control; A dynamic event triggering mechanism is set up that can be triggered when a multi-agent system is subjected to a DoS attack; Based on the performance function of the preset performance control and the control signal when each agent in the multi-agent system suffers an invalid DoS attack, the dynamic threshold in the dynamic event triggering mechanism is set. When the control deviation between the control signal and the ideal signal exceeds the set dynamic threshold and is not during a DoS attack, the system triggering time for the dynamic event triggering mechanism is determined and the control signal is updated. An anti-attack controller is designed based on the aforementioned dynamic event triggering mechanism and Lyapunov theory.
[0006] In a preferred embodiment, the dynamic threshold is set as follows:
[0007]
[0008] in, Control signals indicating an invalid DoS attack. This represents the performance function for preset performance control. For dynamic attenuation parameters, The threshold is set as the lower limit to avoid the Zeno phenomenon, where i represents the i-th agent.
[0009] In a preferred embodiment, the determination of the system trigger time satisfies the following conditions:
[0010] in, Indicates control deviation. This is the DoS attack zone. For the first Duration of the attack.
[0011] As a preferred embodiment, the update control signal is represented as follows: From the conditions satisfied at the system trigger time and Export:
[0012] in , ; Further obtain
[0013] That is hour, , ,have To obtain updated control signals: .
[0014] As a preferred embodiment, the multi-agent system consists of N agents, and the dynamics of each agent are described as follows:
[0015] in, This represents the state vector of the follower agent. To control the input, It is an output variable. For an unknown continuous function, For time-varying unknown perturbations, satisfy , Where is an unknown positive constant; The leader's dynamic equations satisfy the following form:
[0016] in and These are the leader's state and output; Suppose the network topology between a leader and N followers is defined by a directed graph. To describe the information flow between N intelligent agents, where It is a set of nodes. It is a set of edges, and It is an adjacency matrix; if the node Can receive from nodes Information, then ,otherwise, ; Let represent the interior degree matrix, where Laplace matrix If a leader with a flag of 0 exists, it can pass information to... Then we get an enhanced graph. If the intelligent agent Information can be obtained from leaders. ,otherwise .
[0017] As a preferred embodiment, a radial basis function neural network is used to approximate the unknown nonlinear function in a multi-agent system as follows:
[0018] in, It is a smooth nonlinear function. Represents the input vector. For the basis function vector, The number of nodes in the RBFNN; It is an approximation error. As a boundary; It is an ideal weighted vector, denoted as
[0019] in, For weighted vectors, For The estimated value, then This represents the weighting error.
[0020] In a preferred embodiment, the preset performance control uses a local neighborhood to include the error. To satisfy the following inequalities:
[0021] in, and These are two positive design parameters. It is a defined performance function, where Let be the initial boundary and the steady-state boundary of the error, respectively. Let α control the error convergence rate and be greater than 0. Design appropriate parameters to make the following inequality hold. Constrained tracking errors can be converted into equivalent unconstrained signals to achieve the desired control performance.
[0022] Compared to existing technologies, this invention proposes an event-triggered preset performance constraint control method, aiming to provide an effective collaborative control strategy for heterogeneous multi-agent systems under DoS attacks. This method reduces communication frequency and mitigates the impact of DoS attacks on system communication by introducing an event-triggered mechanism; simultaneously, combined with preset performance constraints, it ensures that the system still meets the performance requirements of the task under attack conditions. This provides a new approach and method for solving network security problems in practical applications, and is expected to promote the widespread application and development of heterogeneous multi-agent systems in complex environments. Attached Figure Description
[0023] Figure 1 This is a schematic diagram of the method flow of the present invention; Figure 2 This is a schematic diagram of the communication topology in this invention; Figure 3 The output and leader of the five intelligent agents in this invention. A schematic diagram of the trajectory; Figure 4 This is a schematic diagram of the trajectories of each intelligent agent in the x and y planes in this invention; Figure 5 This is a tracking error diagram from a comparative experiment of the unmanned vehicle of this invention; Figure 6 This is a comparison experiment tracking error diagram of the UAV of this invention; Figure 7 This is a comparison chart of consistency errors of the unmanned vehicles of this invention; Figure 8 This is a schematic diagram of the consistency error comparison of the UAV of the present invention; Figure 9 This is the control input diagram for the intelligent agent algorithm of this invention; Figure 10 This is a schematic diagram illustrating the event triggering interval of the intelligent agent in the x-direction according to the present invention; Figure 11 This is a schematic diagram of the event triggering interval of the intelligent agent in the y-direction according to the present invention; Detailed Implementation
[0024] The technical solutions in 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.
[0025] This invention abandons the traditional paradigm of periodic sampling and constructs an adaptive dynamic triggering threshold strategy based on system state and performance requirements. This strategy significantly reduces communication resource consumption while optimizing information transmission efficiency and ensuring that the system maintains a high degree of responsiveness during critical state transitions.
[0026] By introducing a Preset Performance Function (PPF), a dual-constraint triggering strategy that integrates preset performance and system state is provided. This mechanism breaks through the bottleneck of traditional triggers where "triggering frequency and control precision cannot be achieved simultaneously." Through an exponential decay mechanism, it fundamentally reconstructs the resource allocation and performance guarantee mechanism in distributed systems.
[0027] By using the RBF network to identify the uncertain dynamics of the system in real time, an adaptive control strategy was designed to mitigate the impact of DoS attacks on the communication network, improve the robustness and recovery capability of the system, and thus enhance the stability of the system.
[0028] The proposed control scheme was experimentally verified on a heterogeneous multi-agent physical experimental platform, injecting DoS attack sequences in a collaborative task scenario. Comparative results show that, compared to fixed threshold event triggering and methods without preset dynamic thresholds, the controller of this invention achieves significant advantages in three key indicators: control accuracy, communication load, and post-attack recovery time.
[0029] Based on the above, the solution of this invention aims to solve the problem of collaborative control in heterogeneous multi-agent systems under DoS attacks, providing an innovative and comprehensive solution for the stable operation of such systems in complex network environments. The invention mainly consists of the following parts: The steps for performing state description and topology modeling of heterogeneous multi-agent systems, and establishing dynamic models of the agents, include: 1. Consider a class of high-order nonlinear multi-agent systems. This system consists of N agents, and the dynamics of each agent are described as follows: (1) in, This represents the state vector of the follower agent. To control the input, It is an output variable. For an unknown continuous function, For time-varying unknown perturbations, satisfy , Where is an unknown positive constant.
[0030] The leader's dynamic equations satisfy the following form: (2) in and These are the leader's status and output, respectively.
[0031] Suppose the network topology between a leader and N followers is defined by a directed graph. To describe the information flow between N intelligent agents, where It is a set of nodes. It is a set of edges, and It is an adjacency matrix. If the node Can receive from nodes Information, then ,otherwise, . Let represent the interior degree matrix, where .
[0032] The Laplace matrix is If a leader with a flag of 0 exists, it can pass information to... Then we get an enhanced graph. If the intelligent agent Information can be obtained from leaders. ,otherwise .
[0033] 2. The steps for approximating the unknown nonlinear function in the system using a radial basis function neural network include: Radial Basis Function Neural Networks (RBFNNs) can be used to approximate unknown functions. It can be described in the following form: (3) in, It is a smooth nonlinear function. Represents the input vector. For the basis function vector, This represents the number of nodes in the RBFNN. It is an approximation error. To take as a boundary. It is an ideal weighted vector, denoted as (4) (5) in, For weighted vectors, For The estimated value, then This represents the weighting error.
[0034] 3. The steps for constraining errors through preset performance functions include: The core objective of Prescribed Performance Control (PPC) is to explicitly constrain the transient and steady-state behavior of the system by designing a time-varying performance function, thereby avoiding undesirable dynamic responses such as overshoot and oscillation. To obtain the desired specified performance, it is necessary to use a local neighborhood containing the error. To satisfy the following inequalities: (6) in, and These are two positive design parameters. It is a defined performance function, where Let be the initial and steady-state boundaries of the error, respectively. α controls the error convergence rate and is greater than 0. During the control design process, we will design appropriate parameters to make the following inequality hold. .
[0035] The constrained tracking error can be converted into an equivalent unconstrained signal to achieve the control performance in (6), defined as follows: (7) Where γ is the transformed signal, Γ(γ) is a strictly increasing smooth function, and (8) That is, get (9) Differentiating it has (10) in An error transformation function needs to be constructed. (11) We can get it, as long as If it is bounded, then the proposed performance requirements are met.
[0036] 4. DoS attack model; In practical applications, sensors measure signals and transmit them to a controller via a network. If an adversary attacks the communication channel, the controller may receive incorrect messages or fail to receive messages at all. Therefore, we consider intermittent DoS attacks.
[0037] The sequence and duration of a DoS attack are defined as follows: , and The i-th attack is For any interval Divide it into attacked zones and the interval for normal information transmission Two parts, of which It is a set of intervals in a DoS state. It is a set of time intervals that are not subject to DoS attacks.
[0038] Assumption 1 For any There exists a nonnegative constant. and a positive number , making in Representative interval The frequency of internal DoS attacks.
[0039] Assumption 2 for any There are two positive numbers. and , making satisfy ,in Representative interval Duration of an internal DoS attack.
[0040] 5. Design of dynamic event triggering mechanism; This invention proposes a novel event-triggered control mechanism that achieves a balance between efficient utilization of network resources and system stability through a dynamic relative threshold strategy. The specific definitions are as follows: The control signal is represented as: (12) in, This represents the control input when subjected to an invalid DoS attack. Considering the balance between system communication efficiency and control stability, a Dynamic Event Trigger (DET) mechanism is proposed. Its core theory lies in minimizing data transmission by adjusting the trigger threshold in real time, transmitting control commands only when the system deviates from the target state. The dynamic threshold is set as follows: (13) in, For dynamic attenuation parameters, For the relevant terms of the preset performance function mentioned above, the time-varying characteristics of the weight parameters are characterized by the following exponential decay model: (14) in, , This is a traditional exponentially decaying term, used to suppress the triggering frequency during the steady-state phase of the system. This feature dynamically synchronizes the global decay rate with a preset performance function, thereby optimizing transient response characteristics. A lower threshold is used to avoid Zeno's phenomenon. The determination of the system trigger time satisfies the following mathematical conditions: (15) in This determines when the control signal u is updated. This is the DoS attack zone. For the first Duration of the attack.
[0041] From the above (15) and Exportable: (16) in , .
[0042] Therefore, we can obtain (17) That is hour, , ,have .
[0043] get (18) 6. Design the controller using event-triggered adaptive control theory to ensure that the system error is within the preset range; Step 1: Follower Multi-Agent The neighbor consistency error is expressed as (19) in, This represents the weight gain from leader to follower. To further eliminate the influence of initial conditions, we introduce... Based on the above text From the definition, we can calculate: (20) In controller design (twenty one) The error of the virtual control law is defined as: (twenty two) in To avoid the "complexity explosion" problem caused by continuous differentiation, a first-order low-pass filter is used to approximate the virtual control law. The output of the following filter is represented: (twenty three) Define the filtering error as (twenty four) (25) in , , An RBF neural network is used to approximate the unknown term.
[0044] (26) in ,definition For The estimated value, then Choose the following Lyapunov functions: (27) right Taking the derivative, we get: (28) in, This represents a design parameter. Using Young's inequality, we further derived... (29) Substituting (29) into (28), we have (30) Therefore, virtual controller Designed as (31) in, For the parameters to be designed, the adaptive update law is: (32) It is a positive parameter. Substituting the above equation into (30) yields... (33) Step m :definition (34) From (22), we can obtain (35) From (24) we can obtain (36) in, It is a continuous function, in a compact set Above, there is an upper realm. Then, the derivative of the m-th Lyapunov function is calculated as follows: (37) Using Young's inequality, we can obtain (38) (39) (40) Substituting the above equation into (37) yields the following result: (41) The virtual controller for step m is designed as (42) The update law is (43) Substituting (38)-(43) into (37), we can obtain (44) Step n: Let (45) Its derivative is calculated to be: (46) Similarly, using Young's inequality, we have (47) (48) (49) exist Formula (12) thus transforms formula (46) into (50) Adaptive controller as follows (51) in , virtual signals Adaptive Law of Parameters for (52) (53) Then, by using (50)-(53), we can obtain (54) 7. Analyze the stability of the control system to ensure that the system converges within a fixed time, as follows: This section provides a stability proof for multi-agent systems, considering the final Lyapunov function of the multi-agent system as follows: (55) According to equation (54), for Differentiating gives (56) According to note 2, ,Right now Established. According to... get: (57) Therefore there is (58) in (59) Substituting (58) into (56) yields (60) Choose appropriate parameters to make , .
[0045] definition . (61) get (62) in In a distributed network environment, denial-of-service (DoS) attacks significantly disrupt network communication, blocking the normal transmission path of data packets, while information can be transmitted and updated as expected in a non-attack state. This invention establishes the following operating mechanism: system state updates are performed only when the network is secure and the triggering conditions are met; when a DoS attack is detected, even if the triggering conditions are met, control operations will not be executed, and control parameters will only be updated according to a predetermined period.
[0046] When the parameters satisfy the condition set The event-triggered mechanism maintains security or only results in an invalid DoS attack. When the parameters meet the set of conditions... At that time, the event triggering mechanism experienced an effective DoS attack. This invention will analyze in depth the stability characteristics of the multi-agent system under the above two parameter configurations.
[0047] Case 1:
[0048] (63) in ,and and They represent and , The length of the disjoint union.
[0049] when ,have (64) in This stems from the complementary relationship between attack and non-attack periods; we have: (65) Case 2:
[0050] (66) Given (64) and Assumption 1, we can conclude that (67) It can be obtained (68) in .
[0051] From formulas (65) and (68) and the conditions of assumption 1, the parameters can be verified. , , and It is strictly positive. We consider case 1, where as time approaches infinity, the term... It decays exponentially toward zero, thus becoming bounded. approaching Case 2 is derived from (68). In this case, it can be deduced that as time approaches infinity, Approaching zero, leading to approaching This established the stability of the closed-loop system and provided the proven conclusion.
[0052] It has now been proven that the event triggering mechanism can effectively prevent Zeno behavior from occurring. This can be obtained from formula (54) and the dynamic threshold function. It is bounded, and by taking its derivative, we can deduce that the derivative is a bounded value and is less than 1 / 2. And there are , And there are We can conclude that when t approaches The threshold is close to It is greater than 0. Therefore, there exists a minimum trigger interval. This effectively prevented the occurrence of Zeno's behavior.
[0053] In one specific embodiment, the present invention underwent comprehensive experimental verification based on a robot experimental platform to demonstrate the effectiveness and robustness of the proposed distributed adaptive control algorithm based on Lyapunov stability theory.
[0054] The experimental platform integrates a heterogeneous multi-agent system, comprising a four-wheeled mobile robot, two differential-drive robots, two quadcopter drones, a high-precision motion capture system, and a ground-based central computing workstation. The system is equipped with eight infrared cameras operating at 240Hz. This high-precision positioning system acquires position data in real time via a high-speed data link and transmits it to the workstation for processing, providing essential benchmark data for evaluation and meeting the experiment's stringent requirements for high-precision, low-latency reference data.
[0055] The system adopts a distributed heterogeneous computing architecture with two processing layers: (1) A central workstation (Intel i9-12900K, 64GB RAM) is responsible for complex trajectory planning, multi-agent collaborative optimization, and global decision-making, and performs multi-sensor data fusion, transmitting the data to each heterogeneous agent via a wireless network. (2) A microcontroller is responsible for low-level drive control, sensor data acquisition, and real-time response, ensuring fast closed-loop control. This hierarchical processing achieves scalable computational load distribution under different network conditions while maintaining deterministic control performance.
[0056] For a given topology graph, the communication structure of the system is determined by the adjacency matrix. Mathematically, although the aerial intelligent agent operates with six degrees of freedom in three-dimensional space, while the ground vehicle is restricted to planar motion, our analysis focuses primarily on the dynamics of the xy-plane. This dimensionality reduction is justified by two factors: (1) the formation control target is defined on the horizontal plane, and the vertical direction can be controlled independently. (2) due to the inherent platform asymmetry, the vertical dynamics of the UAV exhibit significantly faster convergence characteristics than the horizontal motion (time constant is about 2.7 times faster), without affecting the overall stability.
[0057] To verify the cooperative control performance of the Dynamic Event Triggering (DET) mechanism in a heterogeneous multi-agent system under a DoS attack, this study focuses on a comprehensive evaluation of trajectory tracking accuracy, communication resource utilization efficiency, and anti-interference capability. The experiment sets the virtual leader generation radius to be... A circular reference trajectory with a period of 25 seconds is used, executed by QCar as the physical leader. Geometric formations are formed around the leader, with a motion formation structure of 2 meters in diameter. During system operation, we designed and injected four periodic DoS attacks of decreasing duration and intensity (attack periods t∈[5,6]∪[11,11.6]∪[15,15.4]∪[20,20.2] seconds) to comprehensively test the system's robustness and resilience under different attack intensities and durations.
[0058] The performance function used in the experiment is The initial performance boundary Steady-state accuracy requirements The attenuation rate is The dynamic event trigger threshold parameter is selected as follows: The parameters are selected as follows Unified selection ,in , For trigger coefficients set for different intelligent agent control settings, this parameter combination is analyzed using Lyapunov stability analysis to minimize communication load while ensuring system stability. The differentiator time constant and control gain matrix are defined. To effectively compensate for model uncertainties, nonlinear dynamic effects, and external disturbances in the system, a radial basis function (RBF) neural network model with 3 input nodes and 9 hidden layer nodes is designed. The network's output layer has a dimension of 2, directly corresponding to the compensation control input in the xy plane. (Experiments by changing the number of hidden layer nodes (from 5 to 15) show that 9 hidden layer nodes achieve the optimal balance between computational complexity and control accuracy.) The Gaussian kernel function and center point matrix are defined as follows:
[0059] By uniformly covering the key regions of the state space, the effectiveness of the compensator is ensured throughout the entire operating range. The initial weights and the weight adaptive learning algorithm employ a modified gradient descent method, with a weight adaptive learning rate of 0.01. The initial positions of the agent are as follows: , , , , .
[0060] To comprehensively evaluate the effectiveness of the proposed method, three sets of comparative experiments were designed: traditional fixed threshold event-triggered control (FTEC), fixed threshold δ=0.2, dynamic event-triggered control (DTETC), and performance-guaranteed adaptive dynamic event-triggered control (DT-ETPPC).
[0061] This invention utilizes a three-dimensional spatial trajectory diagram ( Figure 3 ) and xy-plane trajectory diagram ( Figure 4 The experiment fully demonstrated the system performance. The experiment adopted a standard communication protocol and met the DoS attack intensity setting in Hypothesis 2. When a DoS attack lasting 1.0s occurred at t=5.0s, the system showed the maximum trajectory deviation at t=6.2s. The recovery time after the fourth attack (t=20.0s) was shortened to 0.5s, which confirmed the learning and adaptation capabilities of the neural network compensator.
[0062] Figures 5-8The results comparing the system tracking error and consistency error of three control mechanisms under a DoS attack environment are presented. The data show that DT-ETPPC exhibits significant advantages after the attack: the system error convergence rate is improved by 22.3%, the maximum transient error is reduced by 37.8%, and the steady-state accuracy is improved by 26.5%. During the critical recovery phase from t=6.0s to t=7.5s, the error decay rate of DT-ETPPC is 2.1 times that of FTEC and 1.4 times that of DTETC. Figure 9 The time-varying characteristics of the control input u under the adaptive dynamic threshold were quantitatively analyzed, highlighting the optimization effect of the dynamic threshold event-triggered control algorithm on the control signal. Figure 10-11 The distribution of triggering times and durations of the agents were recorded. Experimental results show that the system applying the adaptive dynamic event triggering combined with the preset performance guarantee mechanism (DT-ETPPC) proposed in this invention reduces the communication load by 42.3% compared to the traditional fixed threshold triggering mechanism, improves the system response speed by 25.6% and reduces energy consumption by 21.4% compared to the existing dynamic threshold mechanism, while ensuring that the system tracking error always meets the preset convergence boundary. Compared to the traditional threshold triggering mechanism (FETC), DT-ETPPC reduces the average number of communications by 55.5%, shortens the recovery time by 57.1%, and reduces the average tracking error by 64.8%. Compared to DTEC, DT-ETPPC optimizes by 30.1%, 45.5%, and 48.1%, respectively. Energy efficiency analysis shows that DT-ETPPC reduces the total system energy consumption by 32.7% while maintaining a stricter preset performance boundary. Error convergence trajectory analysis of the five agents confirms that all agents meet the requirements of the preset exponential convergence rate, and the lower bound of the communication triggering interval remains above 98ms, effectively avoiding the occurrence of Zeno behavior. These quantitative indicators fully demonstrate that the proposed method achieves the theoretically optimal balance between resource efficiency and control performance.
Claims
1. A control method for a heterogeneous multi-agent system under DoS attacks based on an event-triggered strategy, characterized in that, include: Construct a multi-agent system, use radial basis function neural networks to approximate unknown nonlinear functions in the multi-agent system, and constrain system errors through preset performance control; A dynamic event triggering mechanism is set up that can be triggered when a multi-agent system is subjected to a DoS attack; Based on the performance function of the preset performance control and the control signal when each agent in the multi-agent system suffers an invalid DoS attack, the dynamic threshold in the dynamic event triggering mechanism is set. When the control deviation between the ideal signal and the control signal exceeds the set dynamic threshold and is not during a DoS attack, the system trigger moment for the dynamic event triggering mechanism is determined and the control signal is updated. An anti-attack controller is designed based on the aforementioned dynamic event triggering mechanism and Lyapunov theory.
2. The method for controlling a heterogeneous multi-agent system under a DoS attack based on an event-triggered strategy according to claim 1, characterized in that, The dynamic threshold is set as follows: e i (t)=n i (t)|u i (t)|+o i (t)≥Δ * Among them, u i (t) represents the control signal when subjected to an invalid DoS attack, o i (t) represents the performance function of the preset performance control, ν i (t) is the dynamic decay parameter, Δ * The threshold is set as the lower limit to avoid the Zeno phenomenon, where i represents the i-th agent.
3. The method for controlling a heterogeneous multi-agent system under a DoS attack based on an event-triggered strategy according to claim 2, characterized in that, The system trigger time is determined under the following conditions: in, Indicates control deviation, [h l+1 ,h l+1 +τ l+1 ] represents the DoS attack range, τ l The duration of the l-th attack.
4. The method for controlling a heterogeneous multi-agent system under a DoS attack based on an event-triggered strategy according to claim 3, characterized in that, The update control signal is represented as: From the conditions satisfied at the system trigger time and Export: Where k i (t)∈[-1,1], Further obtain That is, u i When (t)≥0, u i (t) < 0, therefore To obtain updated control signals:
5. The method for controlling a heterogeneous multi-agent system under a DoS attack based on an event-triggered strategy according to claim 1, characterized in that, A multi-agent system consists of N agents, and the dynamics of each agent are described as follows: in, u represents the state vector of the follower agent. i (t) represents the control input. It is an output variable. For an unknown continuous function, D i,k (t) represents a time-varying unknown disturbance that satisfies D i,k (t)≤D i * (t), D i * (t) where is an unknown positive constant; The leader's dynamic equations satisfy the following form: in and These are the leader's state and output; Suppose the network topology between a leader and N followers is defined by a directed graph. To describe the information flow between N intelligent agents, where It is a set of nodes. It is a set of edges, and It is an adjacency matrix; if node i can receive information from node j, then a ij >0, otherwise, a ij =0; D = diag{d1,d2,...,d N } represents the interior degree matrix, where Laplace matrix If a leader with a flag of 0 exists, it can pass information to... This results in an enhanced image. If agent i can obtain information from the leader, then b i >0, otherwise b i =0.
6. The method for controlling a heterogeneous multi-agent system under a DoS attack based on an event-triggered strategy according to claim 1, characterized in that, The unknown nonlinear function in a multi-agent system is approximated using a radial basis function neural network as follows: f(Z)=θ *T φ(Z)+σ(Z),Z∈Ω Z Where f(Z) is a smooth nonlinear function. Let φ(Z) represent the input vector, where φ(Z) = [φ1(Z), φ2(Z), ..., φ r (Z)] T ∈ r Let r be the basis function vector, and r be the number of nodes in the RBFNN; σ(Z) is the approximation error, σ(Z) ≤ σ * As a boundary; θ * =[θ1,θ2,...,θ r ] T ∈ r It is an ideal weighted vector, denoted as Where θ is the weighting vector. For the estimated value of θ, then This represents the weighting error.
7. The method for controlling a heterogeneous multi-agent system under a DoS attack based on an event-triggered strategy according to claim 1, characterized in that, The preset performance control uses the local neighborhood containing the error e(t) to satisfy the following inequality: Where, λ min and λ max These are two positive design parameters, o(t) = (o0 - o ∞ )e -αt +o ∞ This represents the preset performance function, where o0 > o ∞ >0 represent the initial and steady-state boundaries of the error, respectively. α controls the error convergence rate and is greater than 0. Appropriate parameters are designed to make the following inequality hold: -λ min o(0)<e(0)<λ max o(0); The constrained tracking error can be converted into an equivalent unconstrained signal to achieve the required control performance.
Citation Information
Patent Citations
Event trigger control method based on Nomoto nonlinear ship model
CN115685751A
Dynamic event triggering traction control method for complex network under hybrid attack
CN116614299A
Processor network security defense system and method
CN117692251A
Event trigger control method capable of ensuring transient performance of flexible joint robot
CN118305782A
Layered asynchronous dynamic event triggered time-varying output formation control method for multi-agent system under DoS attack
CN118945666A
Cited By
Fault-tolerant control method for restoring preset performance of disturbed multi-agent system
CN122219112A
Multi-agent security optimization tracking control method based on cross-layer coupling event triggering
CN122293714A
