Multi-agent distributed resilient cooperative control method and device based on attack reconstruction

Abstract

The application discloses a kind of multi-agent distributed flexible cooperative control method and device based on attack reconstruction, agent, program product.The method comprises: using the system model of current agent, the system actual output value of current agent obtained and the preset sliding mode state observer rule, the system model of sliding mode state observer is established, the system state of current agent is progressively estimated, and system state observation value is obtained;Attack signal is reconstructed using the system state observation value and system actual output value of current agent, and reconstruction signal is obtained;In response to reconstruction signal greater than preset attack signal detection threshold, the control parameter of current agent is determined using reconstruction signal and the distributed observation result of current agent;Current agent is controlled to perform tracking movement relative to leader using control parameter.In the scenario of network attack, each follower in agent cluster can realize consistent tracking of leader state.

Description

Technical Field

[0001] This invention belongs to the field of intelligent agent control technology, specifically relating to a multi-agent distributed elastic cooperative control method and device, intelligent agent, and program product based on attack reconstruction. Background Technology

[0002] The independent execution of missions by a single aircraft (e.g., a drone) is increasingly failing to achieve the desired combat effectiveness. However, employing multi-aircraft cooperative operations (i.e., multiple aircraft simultaneously tracking and attacking) can increase the load on enemy defense systems, making it difficult for them to effectively intercept and shoot down all friendly aircraft. This provides friendly aircraft with greater survivability and opportunities, forming a more powerful attack force and more precise combat capabilities. Ultimately, this improves the performance of multi-aircraft cooperative tracking and control, enhancing the security and reliability of the swarm system. This has led to accelerated research and development in aircraft swarm control in recent years. For example, in related technologies, Chinese patent CN116954087A, "A Neural Network Controller Design Method for Coordinating UAV Swarm Formation," proposes establishing a dynamic model of a single UAV and a system state-space equation. After obtaining the state vector at the corresponding moment using the UAV's own sensors, the network weight coefficients are updated in reverse according to the loss function, thereby inducing gradient descent of the control input. By continuously updating the control input and iterating through multiple strategies, the optimal control strategy at different moments can be obtained. This method does not rely on the construction of traditional loss functions and does not require a large amount of prior model parameter data, greatly improving the real-time performance and stability of UAV swarms.

[0003] Furthermore, the battlefield environment may harbor cyberattacks from the enemy (e.g., disinformation injection, malicious intrusion, unauthorized access, remote tampering, or communication interference against aircraft), which could weaken the coordinated tracking and combat capabilities of aircraft formations and even pose a serious threat to the safety of friendly personnel and equipment. Therefore, in addition to the aforementioned research and development on aircraft formation control, solutions to cyberattacks also need to be considered. However, currently disclosed multi-aircraft formation control schemes lack effective protection against cyberattacks. For example, Chinese patent CN116736889A, "A Method and Related Device for Aircraft Swarm Countermeasures Based on Dynamic Game Theory," proposes determining a first objective function for each defending aircraft according to the defending aircraft's task, and a second objective function for each intruding aircraft according to the intruder's task. The second objective function represents the relationship between the intruder's action control variables and its total task cost, resulting in a set of game-theoretic countermeasure functions for the aircraft swarm. Then, based on the initial flight states of each defending aircraft, each intruder's initial flight states, and the area information of the protected zone, the Nash equilibrium strategy set of the function set is solved to obtain the optimal action strategy for each defending aircraft and each intruder's aircraft. This allows each defending aircraft and each intruder to engage in real-time countermeasures according to their optimal action strategies. This improves the efficiency of interfering with intruders and increases the success rate of aircraft performing defensive tasks. It is evident that while the above-mentioned scheme of Chinese patent CN116736889A provides a method for aircraft swarm countermeasures based on dynamic game theory, it does not consider the issue of defense against network attacks.

[0004] The same problem applies to fields such as intelligent vehicles. Therefore, how to provide a multi-agent distributed resilient cooperative control scheme that can be reconstructed based on attacks has become an urgent problem to be solved. Summary of the Invention

[0005] This invention provides a multi-agent distributed elastic cooperative control method and related products based on attack reconstruction.

[0006] This invention provides the following technical solution: a multi-agent distributed resilient cooperative control method based on attack reconstruction, comprising: establishing a system model of a sliding mode state observer using the current agent's system model, the obtained actual system output value of the current agent, and preset sliding mode state observer rules; performing asymptotic estimation of the system state of the current agent to obtain system state observation values; wherein, the preset sliding mode state observer rules refer to the system matrix, observer gain, and sliding mode variables of the sliding mode state observer required by the established system model of the sliding mode state observer; wherein, the agent cluster includes a leader and multiple followers, the communication relationship between the leader and the multiple followers satisfies a preset communication topology, and at least one follower in the preset communication topology is an out-neighbor of the leader; wherein, the out-neighbor of the leader is a follower in the preset communication topology that can directly obtain the actual system output value of the leader. The system model includes a system matrix representing the current agent and information on control inputs suspected of being tampered with by an attack signal. The system state observations include a first system state observation and a second system state observation. The first system state observation is an estimate of states that cannot be directly measured within the system from a sliding mode state observer, while the second system state observation is an estimate of states that can be directly measured within the system from a sliding mode state observer. Using the system state observations of the current agent and the actual system output values, the attack signal is reconstructed to obtain a reconstructed signal. In response to the reconstructed signal exceeding a preset attack signal detection threshold, the control parameters of the current agent are determined using the reconstructed signal and the distributed observation results of the current agent. The control parameters include compensation for the attack signal. Using the control parameters, the current agent is controlled to perform tracking movements relative to the leader.

[0007] This invention provides the following technical solution: a multi-agent distributed elastic cooperative control device based on attack reconstruction, comprising: a sliding mode observation module, configured to establish a system model of a sliding mode state observer using the system model of the current agent, the obtained actual system output value of the current agent, and preset sliding mode state observer rules, and to perform asymptotic estimation of the system state of the current agent to obtain system state observation values; wherein, the preset sliding mode state observer rules refer to the system matrix, observer gain, and sliding mode variables of the sliding mode state observer required by the established system model of the sliding mode state observer; wherein, the agent cluster includes a leader and multiple followers, the communication relationship between the leader and the multiple followers satisfies a preset communication topology, and at least one follower in the preset communication topology is an out-neighbor of the leader; wherein, the out-neighbor of the leader is a follower in the preset communication topology that can directly obtain the actual system output value of the leader; the system model includes a representation of the... The system matrix of the current agent and information on control inputs suspected of being tampered with by an attack signal; wherein, the system state observation values ​​include a first system state observation value and a second system state observation value; wherein, the first system state observation value is an estimate of the unmeasurable state within the system in the sliding mode state observer, and the second system state observation value is an estimate of the directly measurable state within the system in the sliding mode state observer; an attack signal reconstruction module is configured to: reconstruct the attack signal using the system state observation values ​​of the current agent and the actual output value of the system to obtain a reconstructed signal; a control parameter determination module is configured to: determine the control parameters of the current agent using the reconstructed signal and the distributed observation results of the current agent in response to the reconstructed signal being greater than a preset attack signal detection threshold; wherein, the control parameters include compensation for the attack signal; a motion control execution module is configured to: control the current agent to perform tracking motion relative to the leader using the control parameters.

[0008] This invention provides the following technical solution: a multi-agent distributed resilient cooperative control device based on attack reconstruction, comprising: a memory and a processor, wherein the memory stores a program, and the processor runs the program to execute the aforementioned method.

[0009] The present invention provides the following technical solution: an intelligent agent, including the aforementioned multi-agent distributed elastic cooperative control device based on attack reconstruction.

[0010] The present invention provides the following technical solution: a program product, which executes the aforementioned multi-agent distributed elastic cooperative control method based on attack reconstruction when running on a processor.

[0011] As described above, this invention provides a multi-agent distributed resilient cooperative control method based on attack reconstruction. In scenarios dealing with network attacks, when it is determined that the agent cluster is under network attack (i.e., the reconstruction signal is greater than a preset attack signal detection threshold), it can quickly implement distributed cooperative control relative to the leader based on the reconstruction signal to mitigate the negative impact of the network attack. Furthermore, when it is determined that the agent cluster is not under network attack (i.e., the reconstruction signal is less than or equal to the preset attack signal detection threshold), the aforementioned "implementation of distributed cooperative control relative to the leader based on the reconstruction signal" can be omitted. Therefore, compared to existing solutions (e.g., using robust control techniques), the system control energy of the follower agents can be significantly reduced when it is determined that the agent cluster is not under network attack. Attached Figure Description

[0012] Figure 1 In the context of network attacks, this invention utilizes a multi-agent distributed elastic cooperative control method based on attack reconstruction to realize the overall control flowchart of the agent cluster motion.

[0013] Figure 2 This is an exemplary flowchart of the multi-agent distributed resilient cooperative control method based on attack reconstruction of the present invention;

[0014] Figure 3 This is an exemplary structural diagram of the preset communication topology described in this invention;

[0015] Figure 4 This is another exemplary flowchart of the multi-agent distributed resilient cooperative control method based on attack reconstruction of the present invention;

[0016] Figure 5 This is another exemplary flowchart of the multi-agent distributed elastic cooperative control method based on attack reconstruction of the present invention;

[0017] Figure 6 This is another exemplary flowchart of the multi-agent distributed elastic cooperative control method based on attack reconstruction of the present invention;

[0018] Figure 7 This is a waveform diagram of the leader's attack tilt angle in Embodiment 1 of the present invention;

[0019] Figure 8 This is a waveform diagram of the follower's consistency tracking error towards the leader in Embodiment 1 of the present invention;

[0020] Figure 9 This is a waveform diagram of the reconstructed attack signal by the tracker in Embodiment 1 of the present invention;

[0021] Figure 10This is a waveform diagram of the tracker's control input in Embodiment 1 of the present invention;

[0022] Figure 11 This is a schematic diagram of an example structure of the multi-agent distributed elastic cooperative control device based on attack reconstruction according to the present invention;

[0023] Figure 12 This is another exemplary structural diagram of the multi-agent distributed elastic cooperative control device based on attack reconstruction according to the present invention. Detailed Implementation

[0024] The present invention will be further described below with reference to specific embodiments, but the scope of protection of the present invention is not limited thereto.

[0025] Reference Figure 1 This paper outlines the inventive concept of the multi-agent distributed elastic cooperative control method based on attack reconstruction.

[0026] Figure 1 In the context of network attacks, this invention utilizes a multi-agent distributed elastic collaborative control method based on attack reconstruction to realize the overall control flowchart for the movement of an agent cluster.

[0027] Reference Figure 1 Intelligent agent clusters may include N There are 3 intelligent agents, among which the one numbered is... N The intelligent agents can act as leaders, while those numbered 1 to... N-1 Intelligent agents can act as followers 1 to 1 N-1 (Right now, Figure 1 Trackers 1 to N-1 Any follower in an agent cluster may include the following modules or units: sliding mode state observer ( Figure 1 In Chinese, these are abbreviated as sliding mode observer, distributed observer, controller, actuator, controlled object, and sensor.

[0028] The following is combined Figure 1 This paper outlines the control process of a follower implementing a distributed, cooperative tracking movement relative to a leader in a cyberattack scenario.

[0029] Step 1: The current agent (e.g., tracker 1 or tracker) can be utilized. N-1 The system model and the actual output values ​​of the system (e.g., Figure 1 As shown y 1 or y N-1The system model of the sliding mode state observer is established by using the preset sliding mode state observer rules to perform asymptotic estimation of the system state of the current agent and obtain the system state observation value. Then, the attack signal is reconstructed using the system state observation value of the current agent and the actual output value of the system to obtain the reconstructed signal.

[0030] Among them, the preset sliding mode state observer rules refer to the system matrix, observer gain and sliding mode variables of the sliding mode state observer required by the system model of the established sliding mode state observer; among them, the system state observation value can be determined by the system model of the sliding mode state observer, which represents the estimate of the current agent system state.

[0031] Step 2: Using sensors installed on the current agent itself, the actual output difference of the current agent relative to its neighboring agents can be directly measured (e.g., Figure 1 shown or Based on the actual output difference of the system and the system model of the distributed observer of the current agent, the tracking error observation value of the current agent is determined as the distributed observation result. The type of sensor is not limited in this invention. For example, it may include an inertial measurement unit (IMU), a lidar, etc. The inertial measurement unit can be used to measure the actual output value of the current agent itself; the lidar can be used to measure the aforementioned "difference between the actual output value of the current agent and its neighboring agents".

[0032] Step 3: In response to the reconstructed signal exceeding a preset attack signal detection threshold (i.e., indicating that the current agent is indeed under network attack, for example, under an actuator spoofing signal injection attack), the current agent's own controller can determine control parameters based on the reconstructed signal and distributed observation results (e.g., ...). Figure 1 shown u 1 or u N-1 The control parameters serve as control inputs, which may include compensation for attack signals to counteract their interference.

[0033] Step 4: The current intelligent agent's own actuators can correct or adjust the matched controlled object (e.g., position, attitude, relative speed, thrust and torque of the motor output shaft, yaw angle, pitch angle, roll angle, etc.) according to the control parameters, thereby realizing distributed cooperative tracking motion control relative to the leader.

[0034] Based on steps 1-4 above, distributed collaborative tracking of the leader relative to the follower in an intelligent agent cluster can be achieved, thus maintaining effective tracking in network attack scenarios.

[0035] It should be noted that in the preset communication topology, there is at least one follower. i For leaders N The leader's outgoing neighbors (here, the leader's outgoing neighbors are followers in the preset communication topology that can directly obtain the leader's actual system output value; that is, neighboring agents that can directly access the leader's actual system output value), thereby becoming followers i Followers may determine their relative system output difference based on the actual output value of their own system compared to that of the leader; and then... i The tracking motion relative to the leader can be achieved based on the actual output difference of the system.

[0036] Similarly, in a pre-defined communication topology, the follower... i For followers who are outside their neighbors, then it can be based on the followers. i The difference between the actual output of the system and its own can be used to indirectly track the movement of the relative leader; by analogy, effective tracking can be achieved in network attack scenarios, thereby providing a guarantee for completing the swarm movement task.

[0037] The following is combined Figures 2 to 12 The specific implementation details of the multi-agent distributed elastic cooperative control method based on attack reconstruction of the present invention are described below.

[0038] like Figure 2 The multi-agent distributed elastic cooperative control method based on attack reconstruction shown is executed by any follower in the agent cluster.

[0039] Figure 2 The control method shown can be implemented in a variety of ways in the follower. For example, from a programming perspective, it can be implemented through a computer program; or from a device perspective, it can be implemented through any hardware circuit, such as a processor carrying a computer program, or an application-specific integrated circuit or circuit board assembly that runs the control method.

[0040] Reference Figure 2 The control method specifically includes the following steps.

[0041] Step 101: Using the current agent's system model, the obtained actual system output value of the current agent, and the preset sliding mode state observer rules, establish the system model of the sliding mode state observer, perform asymptotic estimation of the system state of the current agent, and obtain the system state observation value.

[0042] The agent cluster may include a leader and multiple followers; the communication relationship between the leader and multiple followers satisfies a preset communication topology, and in the preset communication topology there exists at least one follower that is an out-neighbor of the leader.

[0043] In this context, the leader's outgoing neighbors are the followers within the preset communication topology that can directly obtain the leader's actual system output value. It should be noted that obtaining the leader's actual system output value serves as a tracking reference.

[0044] The system model includes a system matrix representing the current agent and information on control inputs that have been tampered with by suspected attack signals.

[0045] Among them, the preset sliding mode state observer rules refer to the system matrix, observer gain, and sliding mode variables of the sliding mode state observer required by the system model of the established sliding mode state observer.

[0046] For ease of understanding, the pre-defined communication topology can be described using a directed graph. In this graph, nodes represent agents, and the arrows connecting nodes indicate the data flow direction (it should be noted that the "out" and "in" in "incoming neighbor" and "outgoing neighbor" in this invention are based on this data flow direction; "out" can represent providing data outward relative to the current agent, and "in" can represent acquiring data inward relative to the current agent). In the directed graph, the farther a follower is from the leader, the later it obtains the tracking reference information (i.e., information including the leader's actual system output values). For example, referencing... Figure 3 Agent 5 can be the leader, and agents 1 through 4 can be followers. Among them, follower 1 can directly obtain tracking reference information from the leader, while followers 2 and 4 can only obtain tracking reference information from follower 1, and follower 3 can only obtain tracking reference information from follower 2.

[0047] The mathematical form for describing a predefined communication topology using directed graphs is as follows: Assume the number of agents is... N If there are 1, then it can be used to contain 1. N Each node (i.e., representing an agent) and l Directed graph with edges To describe the preset communication topology. It is a set of nodes. It is a set of edges between different intelligent agents. If This means that the intelligent agent j Capable of acquiring intelligent agents i Information, also referred to as intelligent agent i It is an intelligent agent j Into the neighbor, intelligent agent j It is an intelligent agent i The neighbors, in addition The situation will not be considered, therefore Intelligent agents j The in-neighbor set is defined as Connectivity matrix The elements in the array are defined as follows: If ,but ;if ,but .

[0048] It should be noted that the present invention does not limit the structure of the preset communication topology, and users can make adaptive adjustments according to the actual situation.

[0049] Furthermore, in the solution of this invention, it is assumed that the number is... N The intelligent agent is the leader, while those numbered 1 to... N- An intelligent agent of type 1 is N- One follower. N- None of the followers are allowed to receive input from the leader. (To prevent cyberattacks during communication), while the leader's input information meets... .in, This indicates that the leader's input L2 norm is less than a known constant. ; Known constants For example, it could be the output threshold of the controller.

[0050] As one alternative implementation method, for including N A cluster of agents, where any one agent... i The system model can be represented by the following calculation formulas (1) to (6):

[0051] (1)

[0052] (2)

[0053] (3)

[0054] (4)

[0055] (5)

[0056] (6)

[0057] in, N This represents the number of agents in the agent cluster, and is a positive integer. It is an intelligent agent i The actual output value of the system; It is an intelligent agent i The system state, and all system states cannot be directly measured by sensors; It is an intelligent agent i Input; A, B, C It is an intelligent agenti The system matrix; p, m as well as n Represent the real number space respectively Different dimensions; This indicates the control input that has been tampered with by an attack signal, where t is time; Represents intelligent agents i Control inputs (e.g., tilt angle); Indicates an attack signal; Represents the intensity matrix; Represents an unknown function; This represents a known nonnegative function.

[0058] Step 102: Using the current system state observation value of the intelligent agent and the actual output value of the system, the attack signal is reconstructed to obtain the reconstructed signal.

[0059] The "reconstruction of the attack signal" can be achieved in any available manner. For example, as an alternative implementation, an objective function for optimization can be established based on the system state observations and the actual system output values. Then, the reconstructed signal can be determined by adjusting the sliding mode variables of the system model of the sliding mode state observer until the objective function converges. Specific implementation details will be described below and will not be repeated here.

[0060] Additionally, as an optional implementation, refer to Figure 1 The actual output value of the current intelligent agent can be obtained by using sensors (e.g., inertial measurement units, IMUs) installed on the current intelligent agent itself.

[0061] Step 103: In response to the reconstructed signal being greater than a preset attack signal detection threshold, the control parameters of the current agent are determined using the reconstructed signal and the distributed observation results of the current agent.

[0062] The control parameters may include compensation for attack signals.

[0063] It should be noted that "the reconstructed signal is greater than the preset attack signal detection threshold" indicates that the current agent is indeed under network attack (e.g., under the attack of injecting false signals into the actuator).

[0064] The current agent is the executor of the current plan (i.e., any follower in the agent cluster).

[0065] The reconstructed signal being greater than the preset attack signal detection threshold means that the 2-norm of the reconstructed signal is greater than the preset attack signal detection threshold.

[0066] As an alternative implementation, the distributed observation results can be determined by the current agent's distributed observer based on relative output. Specific implementation details will be described below and will not be repeated here.

[0067] Step 104: Using the control parameters, control the current agent to perform tracking motion relative to the leader.

[0068] As described above, the multi-agent distributed resilient cooperative control method based on attack reconstruction provided by this invention can, in response to network attack scenarios, rapidly implement distributed cooperative control relative to the leader based on the reconstruction signal when it is determined that the agent cluster is under network attack (i.e., the reconstruction signal is greater than a preset attack signal detection threshold), thereby mitigating the negative impact of the network attack. Furthermore, when it is determined that the agent cluster is not under network attack (i.e., the reconstruction signal is less than or equal to the preset attack signal detection threshold), the aforementioned "implementation of distributed cooperative control relative to the leader based on the reconstruction signal" can be omitted. Therefore, compared to existing solutions (e.g., using robust control techniques), the system control energy of the follower agents can be significantly reduced when it is determined that the agent cluster is not under network attack.

[0069] As an optional implementation method, in Figure 2 Based on the embodiments, and referring to Figure 4 Step 101 can be achieved through the following steps 1011-1012.

[0070] Step 1011: Based on the preset sliding mode state observer rules, the control input suspected of being tampered with by the attack signal, and the actual system output value of the current agent, establish the system model of the sliding mode state observer; wherein, the preset sliding mode state observer rules refer to the system matrix, observer gain, and sliding mode variables of the sliding mode state observer required by the established system model of the sliding mode state observer.

[0071] Step 1012: Based on the system model of the sliding mode state observer, determine the system state observation value of the current agent.

[0072] The system matrix of the preset sliding mode state observer is designed as follows. Using linear transformation... System matrix of intelligent agents A, B, C The system matrix of the sliding mode state observer is obtained by processing, as shown in equations (7) to (9). , as well as .

[0073] Specifically, "matching" with the current agent means enabling the matrix to... The upper left section npThe order matrix is ​​a Hurwitz matrix. Above np The order-partitioned block matrix is ​​a zero matrix, and the matrix is... Depend on np 0-order matrices and np It consists of an identity matrix of order.

[0074] (7)

[0075] (8)

[0076] (9)

[0077] in, It is a Hurwitz matrix; , , They represent Submatrices after matrix partitioning; express Submatrices after matrix partitioning; express p An identity matrix of order 1; T This represents the transformation matrix.

[0078] In step 1011, the system matrix of the sliding mode state observer can be used. , as well as For the intelligent agent represented by the computational formulas (1) to (6) i The original system model is processed to obtain the intermediate system model as shown in calculation formulas (10) to (12). Then, based on the intermediate system model, the current intelligent agent is established. i The system model of the sliding mode state observer is shown in the calculation formulas (13) to (15).

[0079] (10)

[0080] (11)

[0081] (12)

[0082] in, Represents intelligent agents i The state of the intermediate system model; Represents intelligent agents i The state of the intermediate system model; express The derivative; express The derivative; and All represent intelligent agents i Control input; and Both indicate attack signals; Represents intelligent agents i The actual output status of the system.

[0083] (13)

[0084] (14)

[0085] (15)

[0086] in, Represents intelligent agents i The first state output by the sliding mode state observer, i.e., the agent's... i The first state observation of the system; Represents intelligent agents i The second state output by the sliding mode state observer, i.e., the agent's... i The system's second state observation value; wherein, the system's first state observation value is the estimate of the state within the system that cannot be directly measured in the sliding mode state observer, and the system's second state observation value is the estimate of the state within the system that can be directly measured in the sliding mode state observer; express The derivative; express The derivative; Represents intelligent agents i attack signal The reconstructed signal (i.e., the sliding mode variable); Represents intelligent agents i The output state of the sliding mode state observer; L This represents the observer gain of the sliding mode state observer.

[0087] As an optional implementation method, in Figure 2 Examples and Figure 4 Based on the implementation method, refer to Figure 5 Step 102 can be achieved through the following steps 1021-1022.

[0088] Step 1021: Establish the system second state estimation error based on the observed value of the system second state and the actual output value of the current agent. Step 1022: Using the system second state estimation error as the objective function and the convergence of the objective function as the optimization objective, adjust the sliding mode variables of the system model of the sliding mode state observer until the objective function converges, and use the sliding mode variables under the convergence condition as the reconstruction signal.

[0089] The first state estimation error of the system can be determined using the following formula (16). The second-state estimation error of the system can be determined using the following formula (17). .

[0090] (16)

[0091] (17)

[0092] in, Represents intelligent agents i The state of the intermediate system model; Represents intelligent agents i The first state output by the sliding mode state observer, i.e., the agent's... i The system's first state; Represents intelligent agents i The system's actual output status; Represents intelligent agents i The output state of the sliding mode state observer.

[0093] It should be noted that, , , , , , They're all about time. t The variable, therefore, according to time t Different values ​​of can determine the corresponding state value. For example, suppose that step 1021 is executed at the time . t 1 ,but At any moment t 1 The value of is the first state observation of the system. Similarly, the value can also be determined accordingly. , , , , At any moment t 1 The value of is used to complete the calculation of the above formulas (16) and (17).

[0094] In step 1022, the reconstructed signal can be determined by using the following calculation formulas (18) to (22) in combination with the aforementioned calculation formulas (10) to (17).

[0095] (18)

[0096] (19)

[0097] (20)

[0098] (twenty one)

[0099] (twenty two)

[0100] in, Indicates adaptive gain; Representation matrix The generalized inverse matrix; Represents any positive number; Indicates the system's first state estimation error The dynamic upper bound of the norm; express The initial value is greater than the system's first state estimation error. initial value ; Represents sliding mode variable Adaptive gain in; Indicating the intensity matrix The norm; express The estimate; express The adaptive rate; Represent a known nonnegative function; Representation matrix The smallest eigenvalue.

[0101] Specifically, the adjustment process can be summarized through the following steps a) to c):

[0102] Step a) Adjust the known nonnegative function any positive number and estimation error Dynamic upper bound of norm The adaptive rate, as shown in formula (21), and the adaptive gain, as shown in formula (19), are adjusted to indirectly adjust the sliding mode variable.

[0103] Step b) In response to the change of sliding mode variables, the first state observation value, the second state observation value, and the intermediate system state value can be determined based on the calculation formulas (10) to (15). The system second state estimation error is obtained by using the calculation formula (17).

[0104] Step c) Determine whether the estimation error of the second state of the system is less than the corresponding preset error threshold (i.e., convergence); if convergence is achieved, the sliding mode variable at this time is used as the reconstruction signal; if convergence is not achieved, iterate the aforementioned adjustment steps a) to c) until the error converges to determine the reconstruction signal.

[0105] As an optional implementation method, in Figure 2 Based on the embodiments, and referring to Figure 6 Step 103 may include the following steps:

[0106] Step 1031: Obtain the actual system output difference between the current agent and the neighboring agent.

[0107] Wherein, the incoming neighbors of the current agent are agents in the preset communication topology that can have their actual system output values ​​obtained by the current agent. (Refer to...) Figure 3 For example, for agents 2 and 4, agent 1 is their in-neighbor agent.

[0108] As an optional example, step 1031 can be implemented in any available manner. For example, the current agent's sensors can be used to measure the difference in the current agent's actual system output relative to the neighboring agent.

[0109] It should be noted that the type of sensor is not limited in this invention. In the example of step 1031, the sensor can be, for example, a lidar; that is, the lidar is used to measure the "difference between the current agent's actual system output and that of its neighboring agents".

[0110] The actual output difference of the system may include, but is not limited to, parameters such as relative distance and relative velocity. The working principle of the lidar includes: transmitting a detection signal (laser beam) to the target (here, the target is the incoming neighbor agent); then comparing the received signal reflected back from the target (target echo) with the transmitted signal; and after appropriate processing, obtaining relevant information about the target (e.g., relative distance, relative velocity, etc.), which serves as the actual output difference of the current agent relative to the incoming neighbor agent.

[0111] Step 1032: Based on the actual output difference of the system and the system model of the distributed observer of the current agent, determine the tracking error observation value of the current agent as the distributed observation result.

[0112] The system model of the distributed observer can be represented by the following calculation formulas (23) to (26):

[0113] (twenty three)

[0114] (twenty four)

[0115] (25)

[0116] (26)

[0117] in, A, B, C Represents intelligent agents i The system matrix; Represents intelligent agents i attack signal The reconstructed signal (i.e., the sliding mode variable); Represents intelligent agents i The tracking error observations; express n An identity matrix of order 1; express m An identity matrix of order 1; F This represents the feedback gain matrix.

[0118] In step 1032, by selecting the feedback gain matrix F Make GA-FC The stable matrix is ​​thus obtained; thereby, the intelligent agent is obtained. i Tracking error observations This indicates that distributed observers can asymptotically estimate the consistency tracking error.

[0119] Step 1033: Using the reconstructed signal and the tracking error observation, determine the control input that matches the current agent as the control parameter.

[0120] As an optional example, step 1023 can be implemented by the following calculations (27) to (36):

[0121] (27)

[0122] (28)

[0123] (29)

[0124] (30)

[0125] (31)

[0126] (32)

[0127] (33)

[0128] (34)

[0129] (35)

[0130] (36)

[0131] in, Represents intelligent agents i Control input; express Adaptive gain in; express Coupling gain in; express The initial value; express The control gain matrix in; Let represent any positive number that satisfies (37); Represents a known constant (e.g., the output threshold of the agent's controller); Represents any positive number; Represents a nonlinear smooth function; A, B Represents intelligent agents i The system matrix.

[0132] In step 1033, the solution is obtained by solving the linear matrix inequality (32). Q Thus, the parameters in the above calculation formulas (28)~(31) and (33)~(36) are determined, and the control input of the intelligent agent represented by calculation formula (27) is determined. That is, an intelligent agent i Distributed control rate.

[0133] In step 104, the actuator utilizes this control input This enables the realization of intelligent agents. i The tracking movement relative to the leader in the cluster.

[0134] As mentioned above, in Figure 6 In this implementation, a distributed observer based on relative output is used to estimate the tracking error observation. The follower can obtain the difference between the actual system output and the actual output of the neighboring agent through the sensor, thus eliminating the need to establish a wireless communication link and avoiding the threat of network spoofing attacks.

[0135] To facilitate understanding of the above technical solutions of the present invention, the following detailed description of the above technical solutions of the present invention is provided through specific embodiments.

[0136] Example 1

[0137] To verify the effectiveness of the above-mentioned solution of the present invention, the designed algorithm is executed using a simulation platform (e.g., Matlab).

[0138] The simulation uses a short-period dynamic model of a general-purpose transport aircraft t(DC-8), where the system matrix is:

[0139] .

[0140] The aircraft system model is as follows:

[0141] ;

[0142] in, q i Indicates the pitch angle of the aircraft; These represent the attack angles of the aircraft; u i This represents the control input of the aircraft, which in Example 1 can specifically represent the aircraft's lift-descent angle.

[0143] The preset communication topology of the aircraft cluster is as follows: Figure 3 As shown, aircraft 5 is the leader, and aircraft 1-4 are followers. The aircraft used to verify the present invention are followers 1-4.

[0144] The leader's (i.e., aircraft 5) attack angle is as follows: Figure 7 As shown; the model of a fake data injection attack (i.e., a specific example of a network attack) is as follows:

[0145] .

[0146] During the simulation phase, the parameters selected based on the scheme of this invention are as follows:

[0147] , , , , , , , , .

[0148] Simulation calculations were performed using a simulation platform to obtain the tracking error between the followers (aircraft 1-4) and the leader (aircraft 5), as shown below. Figure 8 As shown, followers can catch up with the leader in a short time. The results also show that the tracking error converges to zero under the distributed control law designed based on the scheme of this invention. Figure 9 The effectiveness of the reconstructed signal in reconstructing the attack signal has been verified. Figure 10 For the control input of the followers (aircraft 1~4).

[0149] Therefore, the simulation results are consistent with the theoretical analysis, thus demonstrating the effectiveness of the present invention.

[0150] Based on the same inventive concept, and referring to Figure 11The present invention also provides a multi-agent distributed elastic cooperative control device based on attack reconstruction, comprising: a sliding mode observation module 1101, configured to: establish a system model of a sliding mode state observer using the system model of the current agent, the obtained actual system output value of the current agent, and preset sliding mode state observer rules; perform asymptotic estimation of the system state of the current agent to obtain system state observation values; wherein, the preset sliding mode state observer rules refer to the system matrix, observer gain, and sliding mode variables of the sliding mode state observer required by the established system model of the sliding mode state observer; wherein, the agent cluster includes a leader and multiple followers, the communication relationship between the leader and multiple followers satisfies a preset communication topology, and at least one follower in the communication topology is an out-neighbor of the leader; wherein, the out-neighbor of the leader is a follower in the preset communication topology that can directly obtain the actual system output value of the leader; the system model includes a system matrix representing the current agent and information on control inputs suspected of being tampered with by an attack signal; wherein, the system state observation values ​​include a first system state observation value and a second system state observation value. The attack signal reconstruction module 1102 is configured to: reconstruct the attack signal using the current agent's system state observation value and the actual system output value to obtain a reconstructed signal; the control parameter determination module 1103 is configured to: determine the control parameters of the current agent using the reconstructed signal and the distributed observation results of the current agent in response to the reconstructed signal being greater than a preset attack signal detection threshold; wherein the control parameters include compensation for the attack signal; the motion control execution module 1104 is configured to: control the current agent to perform tracking motion relative to the leader using the control parameters.

[0151] The above modules can be implemented by software, hardware, or a combination of both.

[0152] refer to Figure 12 The present invention also provides a multi-agent distributed resilient cooperative control device based on attack reconstruction, comprising: a memory and a processor, wherein the memory stores a program and the processor runs the program to execute the aforementioned control method.

[0153] The present invention also provides an intelligent agent, including the aforementioned multi-agent distributed elastic cooperative control device based on attack reconstruction.

[0154] The present invention also provides a program product that executes the aforementioned attack-reconstruction-based multi-agent distributed elastic cooperative control method when running on a processor.

[0155] The various embodiments in this invention are described in a progressive manner. For the same or similar parts between the various embodiments, please refer to each other. Each embodiment focuses on describing the differences from other embodiments.

[0156] The scope of protection of this invention is not limited to the embodiments described above. Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its scope and spirit. If these modifications and variations fall within the scope of the claims of this invention and their equivalents, then the intent of this invention also includes these modifications and variations.

Claims

1. A multi-agent distributed resilient cooperative control method based on attack reconstruction, characterized in that, include: Using the current agent's system model, the obtained actual system output value of the current agent, and the preset sliding mode state observer rules, a system model of the sliding mode state observer is established to perform asymptotic estimation of the system state of the current agent and obtain the system state observation value. Among them, the preset sliding mode state observer rules refer to the system matrix, observer gain, and sliding mode variables of the sliding mode state observer required by the system model of the established sliding mode state observer; The agent cluster comprises a leader and multiple followers. The communication relationship between the leader and the followers satisfies a preset communication topology, and at least one follower in the preset communication topology is an outgoing neighbor of the leader. The outgoing neighbor of the leader is a follower in the preset communication topology that can directly obtain the actual system output value of the leader. The system model includes a system matrix representing the current agent and information on control inputs that have been tampered with by suspected attack signals. The system state observations include a first system state observation and a second system state observation. The first system state observation is an estimate of the unmeasurable states within the system from the sliding mode state observer, and the second system state observation is an estimate of the directly measurable states within the system from the sliding mode state observer. The attack signal is reconstructed using the current system state observations and the actual system output values ​​of the agent to obtain the reconstructed signal. In response to the reconstructed signal being greater than a preset attack signal detection threshold, the control parameters of the current agent are determined using the reconstructed signal and the distributed observation results of the current agent; wherein the control parameters include compensation for the attack signal; Using the control parameters, the current agent is controlled to perform tracking motion relative to the leader.

2. The method according to claim 1, characterized in that, The process involves establishing a system model for the sliding mode state observer using the current agent's system model, the obtained actual system output value of the current agent, and preset sliding mode state observer rules. This model then performs asymptotic estimation of the current agent's system state to obtain system state observations, including: Based on the preset sliding mode state observer rules, the control input suspected of being tampered with by the attack signal, and the actual system output value of the current agent, a system model of the sliding mode state observer is established; wherein, the preset sliding mode state observer rules refer to the system matrix, observer gain, and sliding mode variables of the sliding mode state observer required by the established system model of the sliding mode state observer; Based on the system model of the sliding mode state observer, the system state observation value of the current agent is determined.

3. The method according to claim 2, characterized in that, The process of reconstructing the attack signal using the current system state observations and actual system outputs of the intelligent agent to obtain the reconstructed signal includes: Based on the observed values ​​of the second state of the system and the actual output values ​​of the current agent, the estimation error of the second state of the system is established. Using the system's second state estimation error as the objective function and the convergence of the objective function as the optimization objective, the sliding mode variables of the system model of the sliding mode state observer are adjusted until the objective function converges, and the sliding mode variables under the converged condition are used as the reconstruction signal.

4. The method according to claim 1, characterized in that, In response to the reconstructed signal exceeding a preset attack signal detection threshold, the control parameters of the current agent are determined using the reconstructed signal and the distributed observation results of the current agent, including: Obtain the actual system output difference between the current agent and its neighboring agents; wherein, the neighboring agent of the current agent is an agent in the preset communication topology whose actual system output value can be obtained by the current agent; Based on the actual output difference of the system and the system model of the distributed observer of the current agent, the tracking error observation value of the current agent is determined as the distributed observation result. Using the reconstructed signal and the tracking error observation, a control input matching the current agent is determined as the control parameter.

5. The method according to claim 4, characterized in that, The step of obtaining the actual system output difference between the current agent and its neighboring agents includes: Using the sensors of the current agent, the difference in the actual system output of the current agent relative to the neighboring agent is measured.

6. A multi-agent distributed elastic cooperative control device based on attack reconstructing, characterized in that, include: The sliding mode observation module is configured to use the system model of the current agent, the actual output value of the current agent's system obtained, and the preset sliding mode state observer rules to establish the system model of the sliding mode state observer, perform asymptotic estimation of the system state of the current agent, and obtain the system state observation value. Among them, the preset sliding mode state observer rules refer to the system matrix, observer gain, and sliding mode variables of the sliding mode state observer required by the system model of the established sliding mode state observer; The agent cluster comprises a leader and multiple followers. The communication relationship between the leader and the followers satisfies a preset communication topology, and at least one follower in the preset communication topology is an outgoing neighbor of the leader. The outgoing neighbor of the leader is a follower in the preset communication topology that can directly obtain the actual system output value of the leader. The system model includes a system matrix representing the current agent and information on control inputs that have been tampered with by suspected attack signals. The system state observations include a first system state observation and a second system state observation. The first system state observation is an estimate of the unmeasurable states within the system from the sliding mode state observer, and the second system state observation is an estimate of the directly measurable states within the system from the sliding mode state observer. The attack signal reconstruction module is configured to: reconstruct the attack signal using the current system state observation value of the agent and the actual output value of the system to obtain the reconstructed signal; The control parameter determination module is configured to: in response to the reconstructed signal being greater than a preset attack signal detection threshold, determine the control parameters of the current agent using the reconstructed signal and the distributed observation results of the current agent; wherein the control parameters include compensation for the attack signal; The motion control execution module is configured to: use the control parameters to control the current agent to perform tracking motion relative to the leader.

7. A multi-agent distributed elastic cooperative control device based on attack reconstructing, characterized in that, It includes a memory and a processor, the memory storing a program, and the processor running the program to perform the method of any one of claims 1 to 5.

8. An intelligent agent, characterized in that, Includes the multi-agent distributed resilient cooperative control device based on attack reconstruction as described in claim 6 or 7.

9. A program product, characterized in that, When the program product is run on the processor, it executes the multi-agent distributed elastic cooperative control method based on attack reconstruction as described in any one of claims 1 to 5.

Images