Distributed prediction control method for unmanned aerial vehicle cluster system under hybrid network attack

Abstract

The invention provides a distributed predictive control method for an unmanned aerial vehicle cluster system under hybrid network attack. The method comprises the following steps: firstly, decoupling a global cost function, and approximately converting a distributed prediction control problem of an unmanned aerial vehicle cluster into a centralized prediction control architecture of independent operation of each unmanned aerial vehicle; on the basis, a state observer is designed in an off-line mode to estimate the system state, a multi-step control invariant set is introduced based on the estimated state, and a distributed predictive control strategy is constructed. Furthermore, a health judgment mechanism is designed, the neighbor relation between the unmanned aerial vehicles is dynamically updated, and nodes suffering from network attacks are isolated from the communication topology in a fault isolation mode. Meanwhile, through introduction of compatibility constraints, asymptotic stability of a global system is ensured. And finally, the feasibility and superiority of the proposed method in a hybrid network attack environment are verified in a mode of combining simulation with a physical experiment.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) swarm system control, and mainly to a predictive control method for UAV swarm systems under hybrid network attacks. Background Technology

[0002] Unmanned aerial vehicle (UAV) swarm systems are intelligent mission systems composed of multiple UAVs that autonomously network, coordinate decision-making, and implement distributed control. Their core lies in achieving global mission coordination through local information exchange, offering significant advantages such as strong environmental adaptability, high system robustness, and excellent mission scalability. As a strategic emerging field integrating next-generation artificial intelligence and aerospace technology, UAV swarms not only provide crucial support for national defense and security but also serve as a new driving force for economic and social development, demonstrating significant application potential in various scenarios such as intelligent logistics, precision agriculture, and smart cities.

[0003] Currently, research on UAV swarm systems is still in its developmental stage, and security and reliability remain the main bottlenecks restricting the large-scale application of this technology. In particular, various network attacks and constraints existing in communication networks pose a severe challenge to the stable operation of UAV swarm systems. Existing research has not fully covered the distributed predictive control problem of UAV swarms under hybrid network attacks, and related systematic methods are still lacking. Furthermore, the multi-step control invariant set method has not been effectively applied in UAV swarm control. This method can improve system control performance through multi-step feedback gain, providing a new solution to such problems. To address these research gaps, this study proposes a distributed predictive control method for UAV swarms resistant to hybrid network attacks, aiming to enhance the stability and reliability of the system under complex interference.

[0004] A search revealed application CN118192661A, which discloses a distributed cooperative predictive control method for UAVs in communication-impaired scenarios. This method establishes a multi-UAV system model for cooperative flight missions, configuring a network model and corresponding transmission channel buffer models for each UAV in a communication-impaired scenario. An improved robust multivariate observer is used to estimate the UAV's state after disturbance and the interference information in the output channel. The estimated state is used as the input to the distributed cooperative predictive controller to compensate for the corresponding interference, thus obtaining a control input generation and compensation strategy for communication-impaired scenarios, achieving distributed cooperative predictive control for UAVs in such scenarios. This invention effectively filters out measurement disturbances and FDI attack effects in the output channel and obtains estimated values ​​of the actual state, effectively addressing communication impairment caused by random packet loss in the transmission channel, reducing the computational burden in cooperative flight missions, and exhibiting high reliability in unknown interference and complex network environments.

[0005] CN118192661A employs a coupled distributed architecture with an undecoupled global cost function. Nodes must jointly optimize and solve this problem, leading to high computational complexity and a tendency for iterative non-convergence when communication is compromised, making it unsuitable for the real-time requirements of large-scale clusters. This invention decouples the global cost function, transforming the cluster control problem into a centralized predictive control where each node operates independently. This solution relies solely on local information, significantly reducing computational and communication burdens, ensuring real-time control performance even in communication-impaired scenarios, and improving applicability for large-scale deployments.

[0006] Regarding network attack response, CN118192661A only passively filters out interference and FDI attack effects through observers, lacking an attack node identification and isolation mechanism, which can easily lead to the spread of erroneous information and global instability. This invention designs a health assessment mechanism that accurately identifies attack nodes by monitoring indicators such as state estimation errors, dynamically updates neighbor relationships, and isolates attack nodes, blocking the spread of errors at the source. This achieves a shift from "passive anti-interference" to "active attack prevention," improving the cluster's robustness against attacks.

[0007] Regarding stability assurance, CN118192661A relies solely on disturbance compensation and a buffer model, without introducing control invariant set constraints, making it prone to state drift and unable to guarantee global asymptotic stability. This invention designs an observer offline and introduces a multi-step control invariant set, combined with compatibility constraints, to theoretically prove global asymptotic stability, thus addressing its weak stability support. Summary of the Invention

[0008] This invention aims to solve the problems of the prior art. It proposes a distributed predictive control method for unmanned aerial vehicle (UAV) swarm systems under hybrid network attacks. The technical solution of this invention is as follows:

[0009] A distributed predictive control method for a drone swarm system under hybrid network attacks includes the following steps:

[0010] Step a: Design the state observer offline and estimate the system state;

[0011] Step b: Introduce a multi-step control invariant set to design a distributed control strategy;

[0012] Step c establishes a compatible constraint condition that makes the global system asymptotically stable.

[0013] Step d involves updating the neighbor relationships between drones through a health assessment mechanism and using "fault isolation" to isolate nodes that have been attacked from the formation communication.

[0014] Furthermore, in step a, the mathematical model of the UAV swarm system is established as follows:

[0015] Consider a swarm of unmanned aerial vehicles (UAVs) that can be decoupled into M subsystems, where the state-space equation of UAV i is:

[0016]

[0017]

[0018] These represent the system's state and output, as well as the control input; It is a known constant matrix.

[0019] Hybrid attacks occurring in the communication network between controllers and actuators are described as follows:

[0020]

[0021] Indicates control input before being attacked. , Indicates the probability of an attack occurring. Indicates an intermediate variable.

[0022] The specific expressions for the observer and controller are as follows:

[0023]

[0024]

[0025] in, , These are the observed values ​​of the system state and the output of the observer, respectively. and These represent the observer gain and the feedback gain, respectively. Let x be the value of x predicted at time k+t, as given at time k. The expression represents the set {0, 1, ..., N-1}, where N is a specified scalar.

[0026] Referring to the design of centralized predictive control objectives, Divided into ,in For UAV The control objective is:

[0027] .

[0028] Represents the set {0,1, ...,M}. Representing the state matrices of adjacent drones and the current drone, respectively. Represents the weight matrix, It is an intermediate variable.

[0029] Furthermore, in obtaining the observer gain Based on this, we study a distributed predictive controller based on a multi-step control invariant set.

[0030] The design based on multi-step control invariant sets can be divided into the following steps:

[0031] i. Incorporate the system state into the ellipsoidally invariant set:

[0032]

[0033]

[0034] For a set of ellipsoidal invariant sets, , , These are intermediate matrix variables.

[0035] ii. Define a time-varying Lyapunov function and force the Lyapunov function to decrease while satisfying the following conditions:

[0036]

[0037] Represents the Lyapunov function. , This represents the weight matrix.

[0038] Define a time-varying Lyapunov function:

[0039]

[0040] in, , , , , , For intermediate matrix variables; , It is a specified scalar. This is the observation error. Therefore:

[0041]

[0042] therefore, Finite-time cost function can be used To approximate;

[0043] iii. Through multi-step control sets Introduce system state into the control invariant set The conditions are:

[0044]

[0045]

[0046] These are intermediate matrix variables.

[0047] iv. Define an upper bound on the maximum value of the objective function, as follows:

[0048]

[0049] in

[0050]

[0051]

[0052]

[0053] The cost function for the ultimate control objective For intermediate matrix variables, This is the weight matrix.

[0054] Furthermore, step c considers sufficient conditions for decoupling the asymptotic stability of the global system to design compatibility constraints:

[0055] Whenever the neighbor's UAV In time When solving optimization problems, let

[0056]

[0057] when You can get

[0058] .

[0059] In a feasible state For the optimal state of the system, For the specified scalar, This is the optimal system control input.

[0060] Furthermore, step d employs a health assessment mechanism to update the communication relationships within the drone cluster, specifically including:

[0061] First, the drone swarm formation is abstracted into several nodes. Then, a health assessment mechanism is used to reset the weight of the edge connected to the node with communication failure to zero.

[0062] First, the communication relationships between drones are updated through a health assessment mechanism. Then, "fault isolation" is used to isolate nodes subjected to network attacks from the formation communication. This is described as follows:

[0063]

[0064] in, , Describe the communications prior to the attack. Describe the communication after an attack. This indicates the communication relationship prior to the attack. The value is determined by the health assessment mechanism;

[0065] The health assessment mechanism can be described as follows:

[0066] Step 1: In At any moment, drones Check its own health status. If the device is in perfect health, it can be set... ,otherwise .

[0067] Step 2: When drones transmit information (such as their position and speed) through communication, health information can be... It is incorporated into the information being transmitted.

[0068] Step 3: Drone Receive drone Information, if , place ;like That is, including or (If no message is received), then set .

[0069] Step 4: Output The value of . Let Return to Step 1.

[0070] Finally, based on communication failure handling strategies and control schemes, and using distributed predictive control theory, a cluster control algorithm under network attacks is designed.

[0071] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, implements a distributed predictive control method for a drone swarm system under hybrid network attacks as described in any one of the claims.

[0072] A non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements a distributed predictive control method for a drone swarm system under hybrid network attacks as described in any one of the claims.

[0073] The advantages and beneficial effects of this invention are as follows:

[0074] This invention relates to a distributed predictive control method for a drone swarm system under hybrid network attacks. First, the global cost function is decoupled, approximating the distributed predictive control problem of the drone swarm as a centralized predictive control architecture where each drone operates independently. Based on this, an offline state observer is designed to estimate the system state, and a multi-step control invariant set is introduced based on the estimated state to construct a distributed predictive control strategy. Furthermore, a health assessment mechanism is designed to dynamically update the neighbor relationships between drones, isolating nodes subjected to network attacks from the communication topology through "fault isolation." Simultaneously, the asymptotic stability of the global system is ensured by introducing compatibility constraints. Finally, the feasibility and superiority of the proposed method in a hybrid network attack environment are verified through a combination of simulation and physical experiments.

[0075] Innovation Point 1: Offline Design of State Observer + Construction of Distributed Control Strategy with Multi-Step Control Invariant Sets – Corresponding to steps a and b of claims. This invention innovatively employs an "offline design observer" to pre-calibrate core parameters, significantly reducing online computational pressure. Simultaneously, it introduces "multi-step control invariant sets" to ensure the predicted system state always remains within the safe and feasible region, theoretically mitigating state drift risks. This combination of "offline pre-design + rigid constraints" breaks away from the conventional "online adaptive + flexible compensation" approach, representing an adaptive design that combines invariant set theory from control theory with a distributed architecture.

[0076] Innovation Point 2: Achieving Global Asymptotic Stability through Compatible Constraint Construction – Corresponding to step c of claim 2. This invention does not rely on the traditional "information interaction enhancement" approach, but instead innovatively designs "compatibility constraints." By constraining the compatibility of local control strategies at each node, global asymptotic stability can be achieved without frequent information interaction. This "constraint-based interaction" design approach precisely solves the stability problem of distributed architectures in complex network environments. It requires a deep understanding of the coupling relationship between global and local optimization, representing a cross-dimensional technological innovation.

[0077] Innovation Point 3: Proactive Anti-Attack Design of Health Judgment Mechanism + Fault Isolation – Corresponding to step d of claim 2. This invention proposes a proactive defense strategy of "health judgment mechanism + fault isolation": identifying attacking nodes by monitoring multi-dimensional indicators such as state estimation error and communication data consistency, and then isolating attacking nodes by dynamically updating neighbor relationships. This proactive defense logic of "identification-isolation" requires the integration of cross-domain technologies such as dynamic adjustment of communication topology and quantifiable evaluation of node health, breaking through the conventional thinking of "only filtering out interference without dealing with the attack source," and is a unique innovative design specifically for hybrid network attack scenarios. Attached Figure Description

[0078] Figure 1 A diagram illustrating the subsystem framework under a network attack;

[0079] Figure 2 For unmanned aerial vehicle (UAV) experimental control platform;

[0080] Figure 3 To simulate the movement trajectory of a drone swarm formation;

[0081] Figure 4 The vertical speeds of the four drones;

[0082] Figure 5 Results of drone motion trajectory frame extraction;

[0083] Figure 6 For displaying the trajectory on the graphic station; Detailed Implementation

[0084] The technical solutions of the embodiments of the present invention will be clearly and thoroughly described below with reference to the accompanying drawings. The described embodiments are merely some embodiments of the present invention.

[0085] The technical solution of the present invention to solve the above-mentioned technical problems is:

[0086] first, Figure 1 This is a diagram of the subsystem framework under a network attack. Figure 2 It serves as an experimental control platform for unmanned aerial vehicles (UAVs). Figure 3 To simulate the movement trajectory of a drone swarm formation, Figure 4 The vertical velocities of the four drones represent the state response of the drone swarm system. It is evident that all drones have reached their designated positions, with their velocities in both the x and y directions converging to zero. Figure 5 The results of the drone motion trajectory frame extraction show that the drones can effectively complete formation tasks and maintain an equal safe distance between adjacent drones. Figure 6The trajectory display on the graphics station shows the movement trajectory of each drone. Due to signal interference, there are some vibrations and deviations compared to the reference trajectory, but these are still within a controllable range.

[0087] The specific implementation includes the following steps:

[0088] Step a: Design the state observer offline and estimate the system state;

[0089] Step b: Introduce a multi-step control invariant set to design a distributed control strategy;

[0090] Step c establishes a compatible constraint condition that makes the global system asymptotically stable.

[0091] Step d involves updating the neighbor relationships between drones through a health assessment mechanism and using "fault isolation" to isolate nodes that have been attacked from the formation communication.

[0092] Furthermore, in step a, the mathematical model of the UAV swarm system is established as follows:

[0093] Consider a drone swarm system that can be decoupled into M subsystems, where the state-space equation of drone i is:

[0094]

[0095]

[0096] These represent the system's state and output, as well as the control input; It is a known constant matrix.

[0097] Hybrid attacks occurring in the communication network between controllers and actuators are described as follows:

[0098]

[0099] , Indicates the probability of an attack occurring. Indicates an intermediate variable.

[0100] The specific expressions for the observer and controller are as follows:

[0101]

[0102]

[0103] in, , These are the observed values ​​of the system state and the output of the observer, respectively. and These represent the observer gain and the feedback gain, respectively. Let x be the value of x predicted at time k+t, as given at time k. The expression represents the set {0, 1, ..., N-1}, where N is a specified scalar.

[0104] Referring to the design of centralized predictive control objectives, Divided into ,in For UAV The control objective is:

[0105]

[0106] Represents the set {0,1, ...,M}. Representing the state matrices of adjacent drones and the current drone, respectively. Represents the weight matrix, It is an intermediate variable.

[0107] Furthermore, in obtaining the observer gain Based on this, we study a distributed model predictive controller based on a multi-step control invariant set.

[0108] The design based on multi-step control invariant sets can be divided into the following steps:

[0109] i. Incorporate the system state into the ellipsoidally invariant set:

[0110]

[0111]

[0112] For a set of ellipsoidal invariant sets, , , These are intermediate matrix variables.

[0113] ii. Define a time-varying Lyapunov function and force the Lyapunov function to decrease while satisfying the following conditions:

[0114]

[0115] Represents the Lyapunov function. , This represents the weight matrix.

[0116] Define a time-varying Lyapunov function:

[0117]

[0118] in, , , , , , For intermediate matrix variables; , It is a specified scalar. This is the observation error. Therefore:

[0119]

[0120] therefore, Finite-time cost function can be used To approximate.

[0121] iii. Through multi-step control sets Introduce system state into the control invariant set The conditions are:

[0122]

[0123]

[0124] These are intermediate matrix variables.

[0125] iv. Define an upper bound on the maximum value of the objective function, as follows:

[0126]

[0127] in

[0128]

[0129]

[0130]

[0131] The cost function for the ultimate control objective For intermediate matrix variables, This is the weight matrix.

[0132] Furthermore, we consider designing compatibility constraints based on sufficient conditions for the asymptotic stability of the decoupled global system.

[0133] Whenever the neighbor's UAV In time When solving optimization problems, let

[0134]

[0135] when You can get

[0136]

[0137] In a feasible state For the optimal state of the system, For the specified scalar, This is the optimal system control input.

[0138] Furthermore, a health assessment mechanism is used to update neighbor relationships within the drone swarm:

[0139] First, the drone swarm formation is abstracted into several nodes. Then, a health assessment mechanism is used to reset the weight of the edge connected to the node with communication failure to zero.

[0140] First, a health assessment mechanism is used to update the neighbor relationships between drones. Then, "fault isolation" is employed to isolate nodes subjected to network attacks from the formation communication. The communication differs before and after the attack, and can be described as follows:

[0141]

[0142] in, , Describe the communications prior to the attack. Describe the communication after an attack. This indicates the communication relationship prior to the attack. The value is determined by the health assessment mechanism.

[0143] The health assessment mechanism can be described as follows:

[0144] Step 1: In At any moment, drones Check its own health status. If the device is in perfect health, it can be set... ,otherwise .

[0145] Step 2: When drones transmit information (such as their position and speed) through communication, health information can be... It is incorporated into the information being transmitted.

[0146] Step 3: Drone Receive drone Information, if , place ;like That is, including or (If no message is received), then set .

[0147] Step 4: Output The value of . Let Return to Step 1.

[0148] Finally, based on communication failure handling strategies and control schemes, and using distributed predictive control theory, a cluster control algorithm under network attacks is designed.

[0149] The systems, devices, modules, or units described in the above embodiments can be implemented by computer chips or entities, or by products with certain functions.

[0150] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0151] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0152] The above embodiments should be understood as illustrative only and not as limiting the scope of protection of the present invention. After reading the description of the present invention, those skilled in the art can make various alterations or modifications to the present invention, and these equivalent changes and modifications also fall within the scope defined by the claims of the present invention.

Claims

1. A distributed predictive control method for a drone swarm system under hybrid network attacks, characterized in that, Includes the following steps: Step a: Design the state observer offline and estimate the system state; Step b: Introduce a multi-step control invariant set to design a distributed control strategy; Step c: Construct a compatible constraint condition to make the global system asymptotically stable; Step d involves updating the neighbor relationships between drones through a health assessment mechanism and using "fault isolation" to isolate nodes that have been attacked from the formation communication.

2. The distributed predictive control method for a drone swarm system under hybrid network attacks according to claim 1, characterized in that, In step a, the mathematical model of the UAV swarm system is established as follows: Consider a drone swarm system that can be decoupled into M subsystems, where the state-space equation of drone i is: These represent the system's state and output, as well as the control input; A known constant matrix; Hybrid attacks occurring in the communication network between controllers and actuators are described as follows: Indicates control input before being attacked. , Indicates the probability of an attack occurring. Indicates intermediate variables; The specific expressions for the observer and controller are as follows: in, , These are the observed values ​​of the system state and the output of the observer, respectively. and These represent the observer gain and the feedback gain, respectively. Let x be the value of x predicted at time k+t, as given at time k. This represents the set {0, 1, ..., N-1}, where N is a specified scalar; Referring to the design of centralized predictive control objectives, Divided into ,in For UAV The control objective is: ； Represents the set {0,1, ...,M}. Representing the state matrices of adjacent drones and the current drone, respectively. Represents the weight matrix, It is an intermediate variable.

3. The distributed predictive control method for a drone swarm system under hybrid network attacks according to claim 2, characterized in that: In obtaining observer gain Based on this, a distributed model predictive controller based on multi-step control invariant sets is designed; The design based on multi-step control invariant sets consists of the following steps: i. Incorporate the system state into the ellipsoidally invariant set: For a set of ellipsoidal invariant sets, , , For intermediate matrix variables; ii. Define a time-varying Lyapunov function and force the Lyapunov function to decrease while satisfying the following conditions: Represents the Lyapunov function. , Represents the weight matrix; Define a time-varying Lyapunov function: in, , , , , , For intermediate matrix variables; , It is a specified scalar. It is an observation error; therefore, we have: therefore, Finite-time cost function can be used To approximate; iii. Through multi-step control sets Introduce system state into the control invariant set The conditions are: For intermediate matrix variables; iv. Define an upper bound on the maximum value of the objective function, as follows: in The cost function for the ultimate control objective For intermediate matrix variables, This is the weight matrix.

4. The distributed predictive control method for a drone swarm system under hybrid network attacks according to claim 3, characterized in that: In order to design compatibility constraints, sufficient conditions to ensure the asymptotic stability of the global system were considered. Whenever the neighbor's UAV In time When solving optimization problems, let when You can get . In a feasible state For the optimal state of the system, For the specified scalar, This is the optimal system control input.

5. The distributed predictive control method for a drone swarm system under hybrid network attacks according to claim 4, characterized in that, Step d employs a health assessment mechanism to update the communication relationships within the drone cluster, specifically including: First, the drone swarm formation is abstracted into several nodes. Then, a health assessment mechanism is used to reset the weight of the edge connected to the node with communication failure to zero. First, the communication relationships between drones are updated through a health assessment mechanism. Then, "fault isolation" is used to isolate nodes subjected to network attacks from the formation communication. This is described as follows: in, , Describe the communications prior to the attack. Describe the communication after an attack. This indicates the communication relationship prior to the attack. The value is determined by the health assessment mechanism; The health assessment mechanism can be described as follows: Step 1: In At any moment, drones Check its own health status. If the device is in perfect health, it can be set... ,otherwise ; Step 2: When transmitting information between drones, including their position and speed, health information can be... Incorporate it into the transmitted information; Step 3: Drone Receive drone Information, if , place ;like That is, including or If no message is received, then set ; Step 4: Output The value; let Return to Step 1; Finally, based on communication failure handling strategies and control schemes, and using distributed predictive control theory, a cluster control algorithm under network attacks is designed.

6. An electronic device, characterized in that, It includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the distributed predictive control method for a drone swarm system under hybrid network attacks as described in any one of claims 1 to 5.

7. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the distributed predictive control method for a drone swarm system under hybrid network attacks as described in any one of claims 1 to 5.

Images