Multi-uav distributed adaptive formation and dynamic obstacle avoidance method against compound attack

Abstract

The application provides a multi-unmanned aerial vehicle (UAV) distributed adaptive formation and dynamic obstacle avoidance method against compound attacks, and relates to the technical field of multi-agent system cooperative control. The application constructs a communication topology and a UAV dynamics model considering a directed denial of service attack and a deception attack; an adaptive topology recovery mechanism is designed to smoothly recover the attacked link; a leader state is estimated based on neighbor node information, a distributed fuzzy state observer is designed to estimate unknown system dynamics, and an attack compensator is designed to offset the influence of the deception attack on the output signal; a distributed formation controller based on an event triggering mechanism is designed, and a dynamic obstacle avoidance mechanism is integrated. Through a unified nonlinear error function, dynamic obstacle speed estimation, attack compensation mechanism and fuzzy observer design, the application can realize safe, stable and cooperative operation of the multi-agent system in a complex environment with deception attacks and unknown dynamic obstacles, and realize high-precision and high-robustness formation control.

Description

Technical Field

[0001] This invention relates to the field of cooperative control technology for multi-agent systems, specifically to a method for distributed adaptive formation and dynamic obstacle avoidance of multi-UAVs to resist complex attacks. Background Technology

[0002] Multiple unmanned aerial vehicles (UAVs) can perform collaborative tasks such as reconnaissance, disaster response, environmental monitoring, and logistics, surpassing single-unit operation modes through full coverage, distributed sensing, redundancy backup, and complex task execution capabilities. Achieving these advantages relies on a robust communication network between UAVs to maintain state estimation, control coordination, and mission synchronization. However, in adversarial environments, among various cyber-physical threats, denial-of-service attacks and deception attacks are particularly serious.

[0003] Resource constraints faced by small unmanned aerial vehicles (UAVs) have spurred a trend towards event-triggered mechanisms replacing traditional time-triggered control protocols. In targeted denial-of-service attacks, attackers can selectively disrupt critical communication links between specific UAVs, causing a sharp decline in the performance of traditional fault-tolerant control methods based on global network information models. This has spurred research into resilient cooperative control. Deception attacks refer to a type of network attack where attackers forge or impersonate trusted information, identities, or sources to bypass system security mechanisms, gain unauthorized access, or induce targets to perform harmful operations. In actual mission execution facing these two types of attacks, dynamic obstacles often appear in formation missions, such as suddenly falling obstacles in rescue missions. Therefore, this study aims to investigate dynamic obstacle avoidance and communication methods based on adaptive fuzzy formation control, building upon research in resilient cooperative control.

[0004] When attacked, the system faces the following challenges: damage to specific links causes local connectivity failure, rendering the global spectrum analysis required by traditional methods ineffective and preventing smooth recovery of attacked links; in real-world environments, different formation modes need to be adjusted according to various complex terrains to achieve dynamic obstacle avoidance. Therefore, there is an urgent need for a dynamic obstacle avoidance and communication scheme that can intelligently adapt to topology changes, ensure stability and performance under strict communication constraints, effectively resist targeted denial-of-service attacks and spoofing attacks, and achieve adaptive fuzzy formation control. This scheme should simultaneously guarantee collision avoidance and obstacle avoidance, maintain communication distance, and eliminate the impact of network attacks. Summary of the Invention

[0005] To address the shortcomings of existing technologies, the present invention aims to propose a multi-UAV distributed adaptive formation and dynamic obstacle avoidance method resistant to complex attacks, including:

[0006] Step 1: Construct a drone formation, which includes a leader drone and follower drones, and the drones transmit position information to each other;

[0007] Step 2: Set residual link weights under denial-of-service attacks; monitor the communication status between drones in real time, and when the communication status is in recovery, introduce a smoothing factor to smoothly adjust the residual link weights until the communication status is fully recovered, then proceed to Step 3.

[0008] Step 3: Calculate the tracking error of the following drone. When the tracking error is greater than or equal to the preset error threshold, trigger communication and transmit the position information of the following drone to the drone with which it has a communication link. When the tracking error is less than the preset error threshold, no action is taken.

[0009] Step 4: When the distance between the follower drone and the obstacle is less than or equal to the preset distance, the same method as calculating the tracking error of the follower drone in Step 3 is used to calculate the tracking error of the follower drone at the current moment. If the tracking error is less than the preset error threshold, no action is taken; if the tracking error is greater than or equal to the preset error threshold, the current moment is taken as the trigger communication moment, the distributed control law of the follower drone at the trigger communication moment is calculated, the follower drone executes the distributed control law at the trigger communication moment, and the auxiliary signals of the follower drone are dynamically updated.

[0010] Optionally, the residual link weights in step 2 include the residual link weights following the drone. Residual link weight of the leader drone Residual link weight following the drone The link weight between follower drone i and follower drone j, and the residual link weight of the leader drone. The link weight between the following drone i and the leader drone.

[0011] Optionally, step 2 involves real-time detection of the communication status between drones, including:

[0012] The success rate of data packets transmitted between drones is statistically analyzed in real time using a sliding window. When the success rate is within a preset threshold range, the communication status is considered to be in recovery; when the success rate is the maximum value of the preset range, the communication status is considered to be fully recovered.

[0013] Optionally, step 2 involves smoothly adjusting the residual link weights using a smoothing factor, specifically achieved through the following formula:

[0014] ;

[0015] ;

[0016] in, This represents the adjusted link weight between follower drone i and follower drone j. The smoothing factor represents the smoothing factor for drone i and drone j following the drone; This indicates the adjusted link weight between the following drone i and the leader drone. This represents the confidence factor. This represents the smoothing factor for the following drone i and the leader drone.

[0017] Optionally, the calculation of the tracking error of the following drone in step 3 is achieved in the following way:

[0018] For each following drone i, a distributed fuzzy state observer is used to compute a local estimate of the leader drone's state. Specifically, this is achieved through the following formula:

[0019] ;

[0020] in, for The first derivative over time, where A is the parameter matrix. For the distributed fuzzy state observer gain, Here is the gain matrix. To follow the location information received by the drone i For the fuzzy parameter matrix, For fuzzy basis function vectors; When the communication link of the drone i is subjected to a spoofing attack, When the communication link following the drone i is not subjected to a spoofing attack, ; The state vector after compensation;

[0021] The fuzzy parameter matrix The adaptive law is updated online and takes the form of gradient descent.

[0022] ;

[0023] in The learning rate matrix is ​​positive definite. It is a positive definite matrix.

[0024] Optionally, the compensated state vector Obtain it through the following methods:

[0025] Design an attack compensator based on fuzzy logic rules to approximate an unknown nonlinear function introduced by a deception attack and compensate for the deviation of the output signal.

[0026] A composite attack model incorporating additive and multiplicative attacks is constructed. The measurable output affected by the deception attack is represented as follows:

[0027] ;

[0028] in, For unknown additive attack functions; For unknown multiplicative attack bonus; This indicates the location information of the drone i. This represents the state vector of the drone i following the drone;

[0029] Design a fuzzy attack compensator , is represented as:

[0030] ;

[0031] in For adaptive parameter matrix, For adaptive parameter vectors, For fuzzy basis function vectors;

[0032] Design an attack gain estimator , is represented as:

[0033] ;

[0034] in For projection operators, The value is a pre-defined positive number and is a constant. This is the compensated tracking error. Positive design parameters;

[0035] The compensated location information is as follows: ;

[0036] The compensated location information The compensated location information It consists of its derivatives.

[0037] Optionally, step 4, calculating the distributed control law of the following UAV at the trigger communication time, includes:

[0038] When the distance between the drone and the obstacle is less than or equal to a preset distance, the nonlinear obstacle avoidance error function is triggered. , is represented as:

[0039] ;

[0040] in, This represents the logarithmic term of the obstacle avoidance error. Indicates the distance between the drone and the obstacle. Indicates the real-time location of the obstacle; Indicates the minimum distance range for obstacle avoidance; Indicates the obstacle detection range;

[0041] Estimating obstacle velocity by designing an adaptive law. Specifically, this is achieved through the following formula:

[0042] ;

[0043] in, It is an estimate of the speed of the obstacle. express The first derivative in time, , For design parameters; This is the compensated tracking error. , For auxiliary compensation signal , for The first derivative, These are design parameters. It is an adaptive parameter vector. They are known basis function vectors; Used to limit the magnitude of the adaptive term to prevent excessive updates from causing system instability; It is a positive number and a constant, used to adjust the input scale of the function;

[0044] Calculate the leader's drone tracking item Agreement with neighbors Specifically, this is achieved through the following formula:

[0045] ;

[0046] ;

[0047] in, Indicates the time when communication is triggered. To estimate the tracking error of the drone i based on its position information, To estimate the tracking error based on the speed information of the drone i, It is a positive definite diagonal gain matrix. To follow the drones in formation, To estimate the tracking error based on the position information of the drone j, To estimate the tracking error based on the drone's speed information;

[0048] Design a distributed control law to follow the drone. , Represented as:

[0049] ;

[0050] in, This is a feedforward compensation term, including the tracking error at the moment of triggering communication. This represents the obstacle avoidance gain coefficient. Represents variables related to obstacle distance. This represents the unit direction vector that follows the drone i as it points toward the obstacle.

[0051] Optionally, the dynamic updating of the auxiliary signals for the following UAV in step 4 is achieved through the following formula:

[0052] ;

[0053] in, Indicates auxiliary signal, The derivative of the auxiliary signal. Indicates the attenuation coefficient. This represents the weighted sum coefficient of the formation.

[0054] The beneficial effects of adopting the above technical solution are as follows:

[0055] Through a three-state adaptive topology recovery model, the system can intelligently utilize the residual connectivity during attack intervals to achieve smooth degradation and recovery of communication and control, significantly outperforming the traditional binary switch model. The observer design relies solely on local neighbor information, completely eliminating dependence on global network spectrum information and enhancing the system's scalability, privacy, and survivability in adversarial environments. The trigger function-based observer significantly reduces the communication burden, achieving robustness under network attacks. The minimum trigger interval strictly eliminates the Zeno phenomenon, ensuring the physical realizability of control. Through a unified nonlinear error function, dynamic obstacle velocity estimation, attack compensation mechanism, and fuzzy observer design, the system can achieve safe, stable, and collaborative operation of multi-agent systems in complex environments with deception attacks and unknown dynamic obstacles. Attached Figure Description

[0056] Figure 1 This is one of the flowcharts illustrating the multi-UAV distributed adaptive formation and dynamic obstacle avoidance method for resisting composite attacks in this invention embodiment;

[0057] Figure 2 This is the second flowchart illustrating the multi-UAV distributed adaptive formation and dynamic obstacle avoidance method against composite attacks in this embodiment of the invention.

[0058] Figure 3 This is a time-varying directed topological graph in an embodiment of the present invention;

[0059] Figure 4 This is a dynamic obstacle avoidance diagram of a drone system simulating an actual mission in an embodiment of the present invention;

[0060] Figure 5 In this embodiment of the invention, the drone follows the other drone to complete a preset formation based on the neighbor node information in a simulated real-world task. Detailed Implementation

[0061] The specific embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples. The following examples are for illustrative purposes only and are not intended to limit the scope of the invention.

[0062] To address the problems existing in the prior art, this invention provides a multi-UAV distributed adaptive formation and dynamic obstacle avoidance method resistant to complex attacks, combined with... Figure 1 and Figure 2 This may include the following steps:

[0063] Step 1: Construct a drone formation, which includes a leader drone and follower drones, and the drones transmit position information to each other;

[0064] This invention also constructs a communication topology and UAV dynamics model that considers directed denial-of-service attacks and spoofing attacks. Specifically, it constructs a time-varying directed graph of UAV formations. ,like Figure 3 ,in It is a set of nodes, with each node corresponding to a drone. For the team leader's drone, To follow the drone's ensemble; arrows exist between nodes, with the head pointing in the direction of information transmission. For the edge set that includes all arrows, The adjacency matrix is ​​a weighted adjacency matrix, which includes the residual follower drone link weights under a denial-of-service attack. and residual leader drone link weight Among them, the residual follower drone link weight The link weight for information transmission between follower drones under a preset ideal state, and the residual leader drone link weight. The link weight for information transmission between the leader drone and the follower drones under a preset ideal state;

[0065] Constructing a dynamic model of a drone:

[0066]

[0067]

[0068] Where A, B, and C are parameter matrices. It is an unknown nonlinear dynamic function. It is necessary to approximate and estimate online using a fuzzy logic system (FLS) to represent model uncertainties, external disturbances, etc. It is the state vector following drone i. Indicates the location of the drone being followed. Indicates the speed at which the drone is following. This indicates the location information of the drone being followed, where t represents time. express The first derivative in time, It is a distributed control law. It is a vector signal that is calculated in real time by the controller based on the current state estimate, neighbor information, attack compensation, and obstacle avoidance requirements.

[0069] Constructing a dynamic model of the leader drone:

[0070]

[0071]

[0072] in, Represents the state vector of the leader drone ,in, This indicates the location information of the leader's drone. Indicates the speed of the leader's drone. express The first derivative in time, This indicates the location information output by the leader drone. The input is for an unknown but bounded leader drone, which may depend on its own state or be time-varying.

[0073] Define the system tracking error as (The control objective is for the error to approach zero; this is used for stability analysis, and robustness will be discussed later.) Let the state vector of the leader drone be... This is the time-varying formation offset vector.

[0074] To address denial-of-service attacks, this invention designs an adaptive topology recovery mechanism to smoothly restore attacked links, specifically including:

[0075] Maintain a three-state model for each directed communication link (j,i):

[0076] ;

[0077] Residual Follower Drone Link Weight The following formula determines the link weight after an attack:

[0078] ;

[0079] in, It is a recovery factor, mostly an exponential function with adjustable parameters. Includes the time when the link begins to recover. Recovery rate parameter Recovery time constant ;

[0080] For the leader drone - follower drone link, its weight Additional by Updated confidence factor Adjustments are needed:

[0081] ;

[0082] in, for; Includes smoothing parameters Update cycle , This represents the packet success rate calculated using a sliding window. It is calculated by statistically analyzing the success rate of receiving packets from neighboring or leader drones within the sliding window. It is a recovery factor.

[0083] Based on the above inventive concept, the present invention performs smooth recovery of attacked links, which is described in detail in step 2.

[0084] Step 2: Set residual link weights under denial-of-service attacks; monitor the communication status between drones in real time, and when the communication status is in recovery, introduce a smoothing factor to smoothly adjust the residual link weights until the communication status is fully recovered, then proceed to Step 3.

[0085] The residual link weights include the residual link weights of the drones. Residual link weight of the leader drone Residual link weight following the drone The link weight between follower drone i and follower drone j, and the residual link weight of the leader drone. The link weight between the following drone i and the leader drone;

[0086] This includes real-time monitoring of the communication status between drones, including:

[0087] The success rate of data packets transmitted between drones is statistically analyzed in real time using a sliding window. When the success rate is within a preset threshold range, the communication status is considered to be in recovery; when the success rate is the maximum value of the preset range, the communication status is considered to be fully recovered.

[0088] The residual link weights are smoothly adjusted using a smoothing factor, specifically through the following formula:

[0089] ;

[0090] ;

[0091] in, This represents the adjusted link weight between follower drone i and follower drone j. The smoothing factor represents the smoothing factor for drone i and drone j following the drone; This indicates the adjusted link weight between the following drone i and the leader drone. This represents the confidence factor. The smoothing factor represents the smoothing factor for the following drone i and the leader drone;

[0092] Step 3: Calculate the tracking error of the following drone. When the tracking error is greater than or equal to the preset error threshold, trigger communication and transmit the position information of the following drone to the drone with which it has a communication link. When the tracking error is less than the preset error threshold, no action is taken.

[0093] Specifically, the state of the leader drone is estimated based on neighbor node information, a distributed fuzzy state observer is designed to estimate the dynamics of the unknown system in response to deception attacks, and an attack compensator is designed to counteract the impact of deception attacks on the output signal.

[0094] For each following drone i, a distributed fuzzy state observer is used to compute a local estimate of the leader drone's state. Specifically, this is achieved through the following formula:

[0095] ;

[0096] in, for The first derivative over time, where A is the parameter matrix. For the distributed fuzzy state observer gain, Here is the gain matrix. To follow the location information received by the drone i For the fuzzy parameter matrix, It is a fuzzy basis function vector (which can be composed of preset membership functions); When the communication link of the drone i is subjected to a spoofing attack, When the communication link following the drone i is not subjected to a spoofing attack, ; The state vector after compensation;

[0097] The fuzzy parameter matrix The adaptive law is updated online and takes the form of gradient descent.

[0098] ;

[0099] in The learning rate matrix is ​​positive definite. It is a positive definite matrix;

[0100] Wherein, the compensated state vector Obtain it through the following methods:

[0101] Design an attack compensator based on fuzzy logic rules to approximate an unknown nonlinear function introduced by a deception attack and compensate for the deviation of the output signal.

[0102] A composite attack model incorporating additive and multiplicative attacks is constructed. The measurable output affected by the deception attack is represented as follows:

[0103] ;

[0104] in, For unknown additive attack functions; For unknown multiplicative attack bonus; This indicates the location information of the drone i. This represents the state vector of the drone i following the drone;

[0105] Design a fuzzy attack compensator : ,in For adaptive parameter matrix, For adaptive parameter vectors, For fuzzy basis function vectors;

[0106] Design an attack gain estimator: ,in This is a projection operator used to ensure that the estimated value is bounded (the bound is 0-). (Preset positive numbers) The value is a pre-defined positive number and is a constant. This is the compensated tracking error. Positive design parameters;

[0107] The compensated location information is as follows:

[0108] The compensated location information The compensated location information It consists of its derivatives.

[0109] Based on local estimation Calculate tracking error Specifically, this is achieved through the following formula:

[0110] ;

[0111] in, This represents the state vector of the drone i following the drone. This is the offset vector of the drone formation;

[0112] When the tracking error is greater than or equal to the preset error threshold, communication is triggered to transmit the position information of the following drone i to the drone that has a communication link with the following drone i.

[0113] This invention designs the event triggering condition for the observer to be the time after k trigger communications, and the time when the observer trigger function is greater than or equal to 0, expressed as:

[0114]

[0115] in, It is the observer trigger function (i.e., the tracking error);

[0116] The observer's trigger function is based on its own estimation error of the leader drone's state. When it detects that its estimation error of the leader drone is too large, the trigger function communicates with neighboring drones to ensure the accuracy of the state estimation.

[0117] Step 4: When the distance between the following drone and the obstacle is less than or equal to a preset distance, the tracking error of the following drone at the current moment is calculated using the same method as in Step 3. If the tracking error is less than a preset error threshold, no action is taken. If the tracking error is greater than or equal to the preset error threshold, the current moment is used as the trigger communication moment. The distributed control law of the following drone at the trigger communication moment is calculated, and the following drone executes the distributed control law at the trigger communication moment. The auxiliary signals of the following drone are dynamically updated. Under the premise of ensuring the reliability of obstacle avoidance sensor data, a distributed formation controller based on an event triggering mechanism is designed, integrating a dynamic obstacle avoidance mechanism.

[0118] An event-triggered communication mechanism is adopted. The controller's triggering condition is: the (c+1)th communication trigger occurs after the (c)th communication trigger, and the observer's trigger function is greater than or equal to 0. The controller's trigger function is activated when the system tracking error is found to be too large and may disrupt the formation. This balances control accuracy and communication burden.

[0119] ;

[0120] in, This indicates the time when the c-th communication is triggered. This represents an edge-based event triggering function that enforces a minimum triggering interval to avoid Zeno behavior.

[0121] When the tracking error of the drone is greater than or equal to the preset error threshold, the current moment is taken as the trigger communication moment, and the drone executes the distributed control law at the trigger communication moment.

[0122] The distributed control law for the following UAV, which is used to calculate the communication trigger moment, includes:

[0123] Step A1: When the distance between the following drone and the obstacle is less than or equal to a preset distance, trigger the nonlinear obstacle avoidance error function. , is represented as:

[0124] ;

[0125] in, This represents the logarithmic term of the obstacle avoidance error. Indicates the distance between the drone and the obstacle. Indicates the real-time location of the obstacle; Indicates the minimum distance range for obstacle avoidance; Indicates the obstacle detection range;

[0126] For example, if the detection distance is 100 meters and the minimum obstacle avoidance distance is 50 meters, and an obstacle appears at 100 meters, the function is triggered, and obstacle avoidance preparation begins. The obstacle avoidance pressure increases as the distance to the obstacle approaches the minimum distance. It is crucial to ensure that the distance to the obstacle never falls below the minimum obstacle avoidance distance. When this occurs, the nonlinear obstacle avoidance error function is activated, triggering obstacle avoidance behavior.

[0127] Step A2: Estimate obstacle velocity by designing an adaptive law. Specifically, this is achieved through the following formula:

[0128] ;

[0129] in, It is an estimate of the speed of the obstacle. express The first derivative in time, , For design parameters; This is the compensated tracking error. , For auxiliary compensation signal , for The first derivative, These are design parameters. It is an adaptive parameter vector. They are known basis function vectors; Used to limit the magnitude of the adaptive term to prevent excessive updates from causing system instability; It is a positive number and a constant, used to adjust the input scale of the function;

[0130] Step A3: Calculate the leader's drone tracking item Agreement with neighbors Specifically, this is achieved through the following formula:

[0131] ;

[0132] ;

[0133] in, Indicates the time when communication is triggered. To estimate the tracking error of the drone i based on its position information, To estimate the tracking error based on the speed information of the drone i, It is a positive definite diagonal gain matrix. To follow the drones in formation, To estimate the tracking error based on the position information of the drone j, To estimate the tracking error based on the drone's speed information;

[0134] Step A4: Design the distributed control law for the drone. , Represented as:

[0135] ;

[0136] in, This is a feedforward compensation term, including the tracking error at the moment of triggering communication. This represents the obstacle avoidance gain coefficient. Represents variables related to obstacle distance. This represents the unit direction vector of the drone i pointing towards the obstacle.

[0137] Includes the tracking error value from the communication at the most recent control trigger moment. Add obstacle avoidance items This allows the intelligent agent to automatically adjust its direction to avoid obstacles when it detects them;

[0138] The calculated distributed control law enables collaborative control with anti-attack capabilities, adaptive formation, and dynamic obstacle avoidance in various task scenarios.

[0139] The auxiliary signals for following the drone are dynamically updated, specifically through the following formula:

[0140] ;

[0141] in, Indicates auxiliary signal, The derivative of the auxiliary signal. Indicates the attenuation coefficient. This represents the weighted sum coefficient of the formation;

[0142] The system also includes obstacle avoidance information, further enhancing its response capability during obstacle avoidance, and is consistent with the first-level error surface in the backstepping control method. Relevant, control objectives :when Prove it by going back step by step.

[0143] like Figure 4 and Figure 5 All followers maintained tracking of the leader drone during obstacle avoidance and ensured formation stability through dynamic updates of auxiliary signals.

[0144] After the formation bypasses the obstacle, the obstacle avoidance function is deactivated, and the system automatically returns to normal formation control mode. All drones maintain low communication overhead while ensuring control accuracy without activating event-triggered mechanisms. The lead drone continues to guide the formation to the target area to complete the rescue mission.

[0145] The method of this invention can ensure the stable operation of a multi-UAV swarm system even under combined attacks. Its core lies in constructing the following three-layer protection system:

[0146] 1. Precise perception of state and dynamics based on fuzzy state observer: By designing a fuzzy state observer, the system can estimate its own state and the unknown nonlinear dynamics caused by model uncertainty and external disturbances in real time and accurately when it cannot be directly measured.

[0147] 2. Active Attack Signal Cancellation Based on Attack Compensator: An attack compensator is specifically designed to counter deception attacks that alter the output signal. It can estimate the attack gain online and correct the output signal affected by the deception attack in real time, thereby ensuring that control decisions are based on real data.

[0148] 3. Adaptive learning of unknown dynamics based on adaptive fuzzy logic: Utilizing the universal approximation property of adaptive fuzzy logic systems, online learning and compensation are performed on complex nonlinear components of the system that are difficult to model. This property makes the system robust.

[0149] The adaptive topology recovery mechanism is run to manage the communication link status with neighboring drones and the leader drone; the distributed fuzzy state observer and attack compensator are run to estimate the leader drone's status and resist attacks; the event-triggered distributed formation controller is run to calculate control commands to achieve formation tracking; and the dynamic obstacle avoidance mechanism is run to avoid obstacles in real time based on sensor information.

[0150] The multi-UAV distributed adaptive formation and dynamic obstacle avoidance method of this invention, which resists complex attacks, utilizes a three-state adaptive topology recovery model. This model enables the system to intelligently utilize residual connectivity during attack intervals, achieving smooth degradation and recovery of communication and control, significantly outperforming the traditional binary switch model. The observer design relies solely on local neighbor information, completely eliminating dependence on global network spectrum information and enhancing the system's scalability, privacy, and survivability in adversarial environments. The trigger function-based observer significantly reduces communication burden, ensuring robustness under attack. The minimum trigger interval strictly eliminates the Zeno phenomenon, ensuring the physical realizability of control. Through a unified nonlinear error function, dynamic obstacle velocity estimation, attack compensation mechanism, and fuzzy observer design, the method enables the safe, stable, and collaborative operation of multi-agent systems in complex environments with deceptive attacks and unknown dynamic obstacles. First, a communication topology and UAV dynamics model considering directed denial-of-service attacks and spoofing attacks are constructed. Next, an adaptive topology recovery mechanism is designed to smoothly restore attacked links. Then, the leader UAV's state is estimated based on neighbor node information, and a distributed fuzzy state observer is designed to estimate unknown system dynamics. An attack compensator is then designed to counteract the impact of spoofing attacks on the output signal. Finally, a distributed formation controller based on an event-triggered mechanism is designed, integrating a dynamic obstacle avoidance mechanism. Through a unified nonlinear error function, dynamic obstacle velocity estimation, attack compensation mechanism, and fuzzy observer design, the safe, stable, and collaborative operation of a multi-agent system can be achieved in complex environments with spoofing attacks and unknown dynamic obstacles.

[0151] The above description is merely a preferred embodiment of this disclosure and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in the embodiments of this disclosure is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described inventive concept. For example, technical solutions formed by substituting the above-described features with (but not limited to) technical features with similar functions disclosed in the embodiments of this disclosure.

Claims

1. A method for adaptive formation and dynamic obstacle avoidance of multi-UAVs resisting complex attacks, characterized in that, include: Step 1: Construct a drone formation, which includes a leader drone and follower drones, and the drones transmit position information to each other; Step 2: Set residual link weights under denial-of-service attacks; monitor the communication status between drones in real time, and when the communication status is in recovery, introduce a smoothing factor to smoothly adjust the residual link weights until the communication status is fully recovered, then proceed to Step 3. Step 3: Calculate the tracking error of the following drone. When the tracking error is greater than or equal to the preset error threshold, trigger communication and transmit the position information of the following drone to the drone with which it has a communication link. When the tracking error is less than the preset error threshold, no action is taken. Step 4: When the distance between the drone and the obstacle is less than or equal to the preset distance, the same method as in Step 3 is used to calculate the tracking error of the drone at the current moment. If the tracking error is less than the preset error threshold, no action is taken. When the tracking error is greater than or equal to the preset error threshold, the current time is taken as the trigger communication time, the distributed control law of the following UAV at the trigger communication time is calculated, the following UAV executes the distributed control law at the trigger communication time, and the auxiliary signals of the following UAV are dynamically updated.

2. The multi-UAV distributed adaptive formation and dynamic obstacle avoidance method for resisting complex attacks according to claim 1, characterized in that, The residual link weights mentioned in step 2 include the residual link weights following the drone. Residual link weight of the leader drone Residual link weight following the drone The link weight between follower drone i and follower drone j, and the residual link weight of the leader drone. The link weight between the following drone i and the leader drone.

3. The multi-UAV distributed adaptive formation and dynamic obstacle avoidance method for resisting complex attacks according to claim 1, characterized in that, Step 2 involves real-time monitoring of the communication status between drones, including: The success rate of data packets transmitted between drones is statistically analyzed in real time using a sliding window. When the success rate is within a preset threshold range, the communication status is considered to be in recovery; when the success rate is the maximum value of the preset range, the communication status is considered to be fully recovered.

4. The multi-UAV distributed adaptive formation and dynamic obstacle avoidance method for resisting complex attacks according to claim 1, characterized in that, Step 2, which involves smoothly adjusting the residual link weights using a smoothing factor, is specifically implemented using the following formula: ； ； in, This represents the adjusted link weight between follower drone i and follower drone j. The smoothing factor represents the smoothing factor for drone i and drone j following the drone; This indicates the adjusted link weight between the following drone i and the leader drone. This represents the confidence factor. This represents the smoothing factor for the following drone i and the leader drone.

5. The multi-UAV distributed adaptive formation and dynamic obstacle avoidance method for resisting complex attacks according to claim 1, characterized in that, Step 3, calculating the tracking error of the drone, is achieved in the following way: For each following drone i, a distributed fuzzy state observer is used to compute a local estimate of the leader drone's state. Specifically, this is achieved through the following formula: ； in, for The first derivative over time, where A is the parameter matrix. For the distributed fuzzy state observer gain, Here is the gain matrix. To follow the location information received by the drone i For the fuzzy parameter matrix, For fuzzy basis function vectors; When the communication link of the drone i is subjected to a spoofing attack, When the communication link following the drone i is not subjected to a spoofing attack, ; The state vector after compensation; The fuzzy parameter matrix The adaptive law is used for online updates, and the adaptive law takes the form of gradient descent: ； in The learning rate matrix is ​​positive definite. It is a positive definite matrix.

6. The multi-UAV distributed adaptive formation and dynamic obstacle avoidance method for resisting complex attacks according to claim 5, characterized in that, The compensated state vector Obtain it through the following methods: Design an attack compensator based on fuzzy logic rules to approximate an unknown nonlinear function introduced by a deception attack and compensate for the deviation of the output signal. A composite attack model incorporating additive and multiplicative attacks is constructed. The measurable output affected by the deception attack is represented as follows: ； in, For unknown additive attack functions; For unknown multiplicative attack bonus; This indicates the location information of the drone i. This represents the state vector of the drone i following the drone; Design a fuzzy attack compensator , is represented as: ； in For adaptive parameter matrix, For adaptive parameter vectors, For fuzzy basis function vectors; Design an attack gain estimator , is represented as: ； in For projection operators, The value is a pre-defined positive number and is a constant. This is the compensated tracking error. Positive design parameters; The compensated location information is as follows: ; The compensated location information The compensated location information It consists of its derivatives.

7. The multi-UAV distributed adaptive formation and dynamic obstacle avoidance method for resisting complex attacks according to claim 1, characterized in that, Step 4, which involves calculating the distributed control law of the following UAV at the trigger communication moment, includes: When the distance between the drone and the obstacle is less than or equal to a preset distance, the nonlinear obstacle avoidance error function is triggered. , is represented as: ； in, This represents the logarithmic term of the obstacle avoidance error. Indicates the distance between the drone and the obstacle. Indicates the real-time location of the obstacle; Indicates the minimum distance range for obstacle avoidance; Indicates the obstacle detection range; Estimating obstacle velocity by designing an adaptive law. Specifically, this is achieved through the following formula: ； in, It is an estimate of the speed of the obstacle. express The first derivative in time, , For design parameters; This is the compensated tracking error. , For auxiliary compensation signal , for The first derivative, These are design parameters. It is an adaptive parameter vector. They are known basis function vectors; Used to limit the magnitude of the adaptive term to prevent excessive updates from causing system instability; It is a positive number and a constant, used to adjust the input scale of the function; Calculate the leader's drone tracking item Agreement with neighbors Specifically, this is achieved through the following formula: ； ； in, Indicates the time when communication is triggered. To estimate the tracking error of the drone i based on its position information, To estimate the tracking error based on the speed information of the drone i, It is a positive definite diagonal gain matrix. To follow the drones in formation, To estimate the tracking error based on the position information of the drone j, To estimate the tracking error based on the drone's speed information; Design a distributed control law to follow the drone. , Represented as: ； in, This is a feedforward compensation term, including the tracking error at the moment of triggering communication. This represents the obstacle avoidance gain coefficient. Represents variables related to obstacle distance. This represents the unit direction vector that follows the drone i as it points toward the obstacle.

8. The multi-UAV distributed adaptive formation and dynamic obstacle avoidance method for resisting complex attacks according to claim 1, characterized in that, Step 4, which involves dynamically updating the auxiliary signals of the following drone, is specifically achieved through the following formula: ； in, Indicates auxiliary signal, The derivative of the auxiliary signal. Indicates the attenuation coefficient. This represents the weighted sum coefficient of the formation.

Images