Non-cooperative target robust formation surrounding control method of multi-unmanned aerial vehicle system

Abstract

The invention discloses a non-cooperative target robust formation surrounding control method of a multi-unmanned aerial vehicle system. The method comprises the following steps: firstly, carrying out dynamic modeling on a multi-unmanned aerial vehicle system and a non-cooperative target, and constructing an unmanned aerial vehicle and target nonlinear dynamic model containing matching and non-matching uncertain items; modeling is carried out on a multi-unmanned aerial vehicle formation non-cooperative target hunting control error; further designing a matching / non-matching disturbance estimator based on a neural network, and dynamically estimating the composite uncertainty through the online updated adaptive weight; and then a multi-unmanned aerial vehicle system adaptive robust formation hunting controller is constructed, so that the unmanned aerial vehicle can hunting a target with a complex dynamic model into a convex hull of the unmanned aerial vehicle. According to the method, neural network fitting and sign-like function items are combined, the convergence performance of non-cooperative target robust formation hunting control of the multi-unmanned aerial vehicle system is improved, and it is ensured that hunting actions are still effectively and reasonably regulated and controlled under the non-cooperative multi-target condition.

Description

Technical Field

[0001] The present invention belongs to the technical field of cooperative control of multiple unmanned aerial vehicle systems, and in particular relates to a non-cooperative target robust formation capture control method for multiple unmanned aerial vehicle systems. Background Art

[0002] With the rapid development of intelligent agent technology, multi-agent systems have gradually become an important tool for solving complex problems. As an important research direction in this field, multi-UAV collaborative control aims to study how to enable UAV groups to achieve specific common goals through coordination and cooperation. The collaborative cooperation between multiple UAVs can significantly improve the efficiency and reliability of the system. The collaborative control of multiple UAVs mainly includes consistency control, formation control, encirclement control, etc. The present invention studies the capture control of a multi-UAV system of non-cooperative targets. The UAVs track multiple targets of interest and capture them into the convex hull formed by the UAVs. The capture control can attack the target or help the target explore the surrounding environment information, which has important application value in military and civilian fields. With the further development of technology, although scholars have designed a series of capture control methods for multi-UAV systems, the existing control strategies still have the following problems:

[0003] First, the existing UAVs considered in the capture control of multiple UAV systems are mostly ideal models. Although methods such as sliding mode control, adaptive control, extended least squares estimation, neural network fitting, and disturbance observer are used to solve the nonlinear terms in the model in the collaborative control research related to capture control, most of them do not consider the composite uncertainty composed of external disturbances and model uncertainty at the same time.

[0004] Second, the problem of non-cooperative multi-target in the existing UAV system capture control is still an open topic. When the targets are uncertain and high-order information is unknown, they may adopt various strategies to avoid the capture of UAVs. Therefore, it is necessary to further design a robust formation capture control method based on neural networks to improve the capture effect. Summary of the invention

[0005] The purpose of the present invention is to overcome the shortcomings of the prior art and provide a non-cooperative target robust formation capture control method for a multi-UAV system based on adaptive neural network estimation, which enables the intelligent agent to capture multiple non-autonomous non-cooperative targets in a certain formation configuration when the target high-order information is unknown. At the same time, the stability of the entire closed-loop system is also proved. It is worth mentioning that this scheme can not only capture multiple targets, but also take into account the model uncertainty and external disturbances of the target, thereby being able to solve more complex real-world problems.

[0006] The objective of the present invention is achieved through the following technical solution: A non-cooperative target robust formation capture control method for a multi-UAV system, comprising the following steps:

[0007] S1. Dynamics modeling of multi-UAV system and non-cooperative targets; details are as follows:

[0008] S11. Construct the undirected communication topology between each drone in the multi-drone system and the directed sensing topology between the drone and the target; assume that there are N+M drones in total, of which M are targets and N are ordinary drones (unless otherwise specified, drones in the following text refer to ordinary drones); F = {1, 2, ..., N} is the index set of ordinary drones, and E = {N+1, N+2, ..., N+M} is the index set of targets; It means that there is an undirected communication topology between drones and a directed sensing topology between drones and targets; is a set of points, is an edge set; is a weighted adjacency matrix, if there is an edge from node j to i in the set ε, then element a ij =1, otherwise 0;

[0009] The edge in is represented by v ij =(v i ,v j ) indicates that i≠j; represents the neighbor set of node vi; is the in-degree matrix of the system, expressed as i=1,2,...,N+M, where d i Expressed as is the Laplace matrix of the system, Further expressed as:

[0010]

[0011] in It is the Laplace matrix unique to UAVs, reflecting the communication between them; is the Laplace matrix between the UAV and the target, reflecting the acquisition of target information by the UAV; The elements in l ij It is expressed as:

[0012]

[0013] S12. Use a neural network architecture to represent any continuous function f(x):

[0014] f(x)=K Tθ(x)+σ(x), (3)

[0015] in is the neural network weight, is the radial basis function, represents a set of real numbers, h is the number of neurons; the i-th element θ in θ(x) i (x) = exp[-(xd i ) T (xd i ) / f i 2 ], i = 1, 2, ..., h, where d i =[d i1 ,d i2 ,...,d ih ] T is the center of the receptive field, f i is the width of the Gaussian function, is the neural network fitting error;

[0016] The optimal neural network weights to minimize the fitting error;

[0017] S13, the drone measures the target's status information x through the onboard sensor k and v k , can track the trajectory of the target in a certain formation configuration according to the information, and capture the target; adopt a distributed cooperative control scheme based on adaptive neural network estimation, so that the UAV can achieve the capture task of non-cooperative multiple targets when the target control input is unknown;

[0018] Assume that at least one of the drones can obtain information about all targets through onboard sensors, and that the drone has a path to other drones in the communication topology; the state variables x of the drone and the target are i , state change rate v i and its derivatives They are all bounded;

[0019] If the multi-UAV system meets the following conditions, it means that the multi-UAV has completed the roundup control:

[0020]

[0021] in The status information of the drone; is the target status information; h i (t) = [h ix (t),h iv (t)] T is a predefined formation, h ix(t), h iv (t) are the position deviation and speed deviation that satisfy the formation, ε0 is the tracking error, η0 is the tracking error, Time to achieve roundup; γ k To satisfy The coefficient of

[0022] S14. Consider the following second-order nonlinear multi-UAV system dynamics model:

[0023]

[0024] where i∈F; u i (t)∈D u are the state and control input vector of UAV i respectively; m=1; item and is an unknown smooth bounded vector-valued function; suppose d(m i ) / dt is bounded; τ 1i (t) and τ 2i (t) is the disturbance, satisfying ‖τ 1i (t)‖≤d1 and‖τ 2i (t)‖≤d2, d1 and d2 are the upper bounds of the perturbation;

[0025] The dynamic model of the target is:

[0026]

[0027] in u k (t)∈D u are the state and control input of target k, k∈E; is an unknown smooth bounded vector-valued function; suppose d(p k ) / dt is bounded; τ 3k (t) and τ 4k (t) is the disturbance, satisfying

[0028] ‖τ 3k (t)‖≤d3 and‖τ 4k (t)‖≤d4, d3 and d4 are the upper bounds of the perturbation;

[0029] S2. Modeling of non-cooperative target capture errors of multi-UAV formations: Introducing auxiliary variables ζ x (t), ζ v (t),ψ x (t), where ζ x (t), ζ v(t) represent the actual state capture error and control input capture error, respectively, x (t) represents the tracking error of the target when the UAVs are in a certain formation;

[0030]

[0031] in u(t) and u E (t) are the control input signals of the UAV and the target respectively; col is defined as:

[0032]

[0033] k = N+1, N+2, ... N+M;

[0034] S3. Design of matching / non-matching disturbance estimator based on neural network estimation: Use neural network to fit nonlinear terms in dynamic model; there are:

[0035]

[0036] To minimize the fitting error, the optimal neural network weight matrix is ​​i=1,2,3;

[0037] is the radial basis function, For K * The estimate, The estimation error is expressed as diag is defined as: is the neural network fitting error, which is a bounded value;

[0038] S4. Design of adaptive robust formation capture controller for multi-UAV system; the details are as follows:

[0039] S41. Design a distributed formation capture controller based on neural network. Before designing the controller, define several important variables: for drone i (i∈F), we have:

[0040] Ω 1i (t) = ζ xi (t), (12)

[0041]

[0042] In order to deal with the integral disturbance, a robust virtual controller is introduced as follows:

[0043]

[0044] J 1ii are the elements of the diagonal matrix J1, the J1 matrix satisfies J1-(∈1 / 2+∈0+∈2+∈3)I≥0, and 3 are positive parameters designed below, I represents the identity matrix; Expressed as in Is The diagonal elements of a positive definite matrix are is a diagonal matrix Elements of and They are used to handle the position state disturbances of multiple UAVs and targets respectively. θ 1i (t) is θ 1i (ζ xi (t)) abbreviation;

[0045] S42. To avoid the derivation of the above virtual controller, the following low-pass filter is introduced for the i-th UAV:

[0046]

[0047] in is the virtual control signal and filtered signal of the i-th UAV; g ii is the element of the diagonal matrix G, the matrix G satisfies G -1 -(1 / (2∈1)-∈4 / 2)I>0;∈4 is a positive parameter of the design;

[0048] The filtering error is expressed as:

[0049]

[0050] for The matrix formed is the virtual control signal of all drones;

[0051] Then get

[0052]

[0053] S43. Define the following parameters:

[0054]

[0055] The adaptive update rate of the neural network is:

[0056]

[0057] is a diagonal matrix The diagonal elements of For i=1,2,3 diagonal matrix Ψ i Satisfy λ min (Ψ i )>0; is a positive definite diagonal matrix;

[0058] S45. Based on the inspiration from (11)-(18), the control rate of UAV i is set to:

[0059]

[0060] ρ0,ρ1 are positive design parameters; J 2ii > 0, is the element of the diagonal matrix J2; is the adaptive parameter of the symbolic function,

[0061] Respectively for estimates;

[0062] In the topology diagram Under this condition, by adopting a low-pass filter (15) and a distributed capture controller based on a neural network, the multi-UAV system (6) can achieve capture control of multiple targets (7), that is, satisfying conditions (4) and (5).

[0063] The beneficial effects of the present invention are: providing a non-cooperative target robust formation capture control method for a multi-UAV system, combining neural network fitting and quasi-symbolic function terms, improving the convergence performance of the capture control of a multi-UAV system with complex uncertainties, and ensuring that the capture action can still be effectively and reasonably regulated in the case of non-cooperative multi-targets. Compared with the prior art, the innovation of the present invention is reflected in the following two aspects:

[0064] (1) A capture scheme for non-cooperative targets is designed based on neural network fitting. The adaptability, real-time performance and flexibility of the radial basis function neural network are used to fit the model uncertainty of the agent and the target. The composite uncertainty is dynamically estimated through online updated adaptive weights to ensure stable capture performance in complex environments. A distributed capture control protocol with arbitrarily small errors is proposed.

[0065] (2) To achieve the capture of multiple non-autonomous and non-cooperative targets, the designed control rate can solve the problem of unknown non-zero control input of the target, ensuring that the UAV system can still effectively capture the target under the condition of uncertainty. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] Figure 1 A flowchart of a non-cooperative target robust formation capture control method for a multi-UAV system of the present invention;

[0067] Figure 2 The communication topology diagram for 8 UAVs and 3 targets;

[0068] Figure 3 The state diagram of different time targets and multiple UAVs, T k represents the target k, A i represents drone i; DETAILED DESCRIPTION

[0069] In order to better describe the content of the invention, some symbols in the following text are explained here: Let R represent a real number set. Let 1 N is a column vector of size N, whose elements are all 1. N represents the identity matrix of size N. Let 0 M×N is a matrix of appropriate size with all elements set to zero. The matrix inequality A>B (A≥B) implies that AB is positive (semipositive) definite. Let λ max (M) and λ min (M) are the maximum eigenvalue and minimum eigenvalue of the real symmetric matrix M. The vector The Euclidean norm ||α|| of is defined as vector The norm of ||α||1 is defined as vector The norm of ||α|| ∞ Defined as ||α|| ∞ =max i (|α i |). Define a column vector. Define a diagonal matrix. Let Indicates the Kronecker product. The "." above the letter, e.g. etc., all represent derivatives.

[0070] The technical solution of the present invention is further described below in conjunction with the accompanying drawings.

[0071] like Figure 1 As shown, a non-cooperative target robust formation capture control method for a multi-UAV system of the present invention comprises the following steps:

[0072] S1. Dynamics modeling of multi-UAV system and non-cooperative targets; details are as follows:

[0073] S11. Construct the undirected communication topology between each drone in the multi-drone system and the directed sensing topology between the drone and the target; assume that there are N+M drones in total, of which M are targets and N are ordinary drones; F = {1, 2, ..., N} is the index set of ordinary drones, and E = {N+1, N+2, ..., N+M} is the index set of targets; It means that there is an undirected communication topology between drones and a directed sensing topology between drones and targets; is a set of points, is an edge set; is a weighted adjacency matrix, if there is an edge from node j to i in the set ε, then element a ij is positive, otherwise it is 0;

[0074] The edge in is represented by v ij =(v i ,v j ) indicates that i≠j; represents the neighbor set of node vi; is the in-degree matrix of the system, expressed as where d i Expressed as is the Laplace matrix of the system, Further expressed as:

[0075]

[0076] in It is the Laplace matrix unique to UAVs, reflecting the communication between them; is the Laplace matrix between the UAV and the target, reflecting the acquisition of target information by the UAV; The elements in l ij It is expressed as:

[0077]

[0078] S12. Use a neural network architecture to represent any continuous function f(x):

[0079] f(x)=K T θ(x)+σ(x), (3)

[0080] in is the neural network weight, is the radial basis function, represents a set of real numbers, h is the number of neurons; the i-th element θ in θ(x) i (x) = exp[-(xd i ) T (xd i ) / f i 2 ], i = 1, 2, ..., h, where d i =[d i1 ,d i2 ,...,d ih ]T is the center of the receptive field, f i is the width of the Gaussian function, is the neural network fitting error;

[0081] The optimal neural network weights to minimize the fitting error;

[0082] S13, the drone measures the target's status information x through the onboard sensor k and v k , can track the trajectory of the target in a certain formation configuration according to the information, and capture the target; adopt a distributed cooperative control scheme based on adaptive neural network estimation, so that the UAV can achieve the capture task of non-cooperative multiple targets when the target control input is unknown;

[0083] Assume that at least one of the drones can obtain information about all targets through onboard sensors, and that the drone has a path to other drones in the communication topology; the state variables x of the drone and the target are i , state change rate v i and its derivatives They are all bounded;

[0084] If the multi-UAV system meets the following conditions, it means that the multi-UAV has completed the roundup control:

[0085]

[0086] in The status information of the drone; is the target status information; h i (t) = [h ix (t),h iv (t)] T is a predefined formation, h ix (t), h iv (t) are the position deviation and speed deviation that satisfy the formation, η0 is the tracking error, Time to achieve roundup; γ k To satisfy The coefficient of

[0087] S14. Consider the following second-order nonlinear multi-UAV system dynamics model:

[0088]

[0089] where i∈F; u i (t)∈D uare the state and control input vector of UAV i respectively; m=1; item and is an unknown smooth bounded vector-valued function; suppose d(m i ) / dt is bounded; τ 1i (t) and τ 2i (t) is the disturbance, satisfying ‖τ 1i (t)‖≤d1 and‖τ 2i (t)‖≤d2, d1 and d2 are the upper bounds of the perturbation;

[0090] The dynamic model of the target is:

[0091]

[0092] in u k (t)∈D u are the state and control input of target k, k∈E; is an unknown smooth bounded vector-valued function; suppose d(p k ) / dt is bounded; τ 3k (t) and τ 4k (t) is the disturbance, satisfying

[0093] ‖τ 3k (t)‖≤d3 and‖τ 4k (t)‖≤d4, d3 and d4 are the upper bounds of the perturbation;

[0094] S2. Error modeling of non-cooperative target capture by multi-UAV formation: Introducing auxiliary variable ζ x (t), ζ v (t),ψ x (t), where ζ x (t), ζ v (t) represent the actual state capture error and control input capture error, respectively, x (t) represents the tracking error of the target when the UAVs are in a certain formation;

[0095]

[0096] in u(t) and u E (t) are the control input signals of the UAV and the target respectively; col is defined as:

[0097]

[0098] k = N+1, N+2, ... N+M;

[0099] S3. Design of matching / non-matching disturbance estimator based on neural network estimation: Use neural network to fit nonlinear terms in dynamic model; there are:

[0100] To minimize the fitting error, the optimal neural network weight matrix is ​​i=1,2,3; is the radial basis function, For K * The estimate, The estimation error is expressed as diag is defined as: is the neural network fitting error, which is a bounded value;

[0101] S4. Design of adaptive robust formation capture controller for multi-UAV system; the details are as follows:

[0102] S41. Design a distributed formation capture controller based on neural network. Before designing the controller, define several important variables: for drone i (i∈N), we have:

[0103] Ω 1i (t) = ζ xi (t), (12)

[0104]

[0105] In order to deal with the integral disturbance, a robust virtual controller is introduced as follows:

[0106]

[0107] J 1ii are the elements of the diagonal matrix J1, the J1 matrix satisfies J1-(∈1 / 2+∈0+∈2+ε3)I≥0, and ∈3 are positive parameters, I represents the identity matrix; Expressed as in Is The diagonal elements of a positive definite matrix are is a diagonal matrix Elements of and They are used to handle the position state disturbances of multiple UAVs and targets respectively. θ 1i (t) is θ 1i (ζ xi (t)) abbreviation;

[0108] S42. To avoid the derivation of the above virtual controller, the following low-pass filter is introduced for the i-th UAV:

[0109]

[0110] where χ 1i (t) is the virtual control signal and filtered signal of the i-th UAV; g ii is the element of the diagonal matrix G, the matrix G satisfies G -1 -(1 / (2∈1)-∈4 / 2)I>0;∈4 is a positive parameter of the design;

[0111] The filtering error is expressed as:

[0112]

[0113] Then get

[0114]

[0115] S43. Define the following parameters:

[0116]

[0117] The adaptive update rate of the neural network is:

[0118]

[0119] is a diagonal matrix The diagonal elements of For i=1,2,3 diagonal matrix Ψ i Satisfy λ min (Ψ i )>0; is a positive definite diagonal matrix;

[0120] S45. Based on the inspiration from (11)-(18), the control rate of UAV i (i∈N) is set to:

[0121]

[0122] ρ0,ρ1 are positive design parameters; J 2ii > 0, is the element of the diagonal matrix J2; is the adaptive parameter of the symbolic function,

[0123] Respectively for estimates;

[0124] In the topology diagram Under this condition, by adopting a low-pass filter (15) and a distributed capture controller based on a neural network, the multi-UAV system (6) can achieve capture control of multiple targets (7), that is, satisfying conditions (4) and (5).

[0125] S5. Construction and stability analysis of the closed-loop error system for formation capture; details are as follows:

[0126] S51. Taking the derivative of Ω1(t), we can obtain from (9):

[0127]

[0128] Ω1(t) is Ω 1i (t) The formed matrix represents the matrix formed by the variables of all drones;

[0129] Assume that the following uncertainty or integral perturbation is bounded, which means that there is a positive upper limit satisfying the following condition

[0130]

[0131] 1 N-1 represents the identity matrix of size N-1;

[0132] Based on (14)-(16) and (18), system (21) becomes:

[0133]

[0134] Consider the first candidate Lyapunov function The derivative of V1(t) is:

[0135]

[0136] According to the inequality And combined with (18)(19), we have:

[0137]

[0138] Where ε1,ε3 are arbitrarily small positive numbers; according to the inequality ±2x T y≤x T Zx+y T Z -1 y, ±x T y≤‖x‖1‖y‖ ∞ ,get:

[0139]

[0140] and

[0141]

[0142] definition where σ 1,N is the upper bound of ||σ1||, σ 3,N is an upper bound of ||σ3||, ∈0, ∈1, ∈2, and ∈3 are positive scalars, and we obtain:

[0143]

[0144] I2 represents the identity matrix of size 2;

[0145] S52. According to (8), (11), and (18), we obtain:

[0146]

[0147] Substituting (20) into (33), we obtain:

[0148]

[0149] Consider the second candidate Lyapunov function

[0150] The derivative of V2(t) is:

[0151]

[0152] for The formed matrix is ​​the low-pass filter for all drones;

[0153] Assume τ2(t),σ2, and

[0154] is bounded; we get:

[0155]

[0156] Notice Through (35) to (37), we get:

[0157]

[0158] Consider the following candidate Lyapunov function V(t) for the entire system (6)(7) with protocol (20):

[0159]

[0160] By taking the derivative of V(t) through (32) and (38), we obtain:

[0161]

[0162] From (17) and the inequality ±2x T y≤x T Zx+y T Z -1 y, ±x T y≤‖x‖1‖y‖ ∞ ., we get:

[0163]

[0164] Where ∈4 is a positive scalar; according to the properties of the matrix trace, we have

[0165] Substituting (41) into (40), we obtain:

[0166]

[0167] Substituting (19) into (42), we obtain:

[0168]

[0169] Using the theorem conclusion of matrix trace, we get:

[0170]

[0171] Where 0<μ i <2λ min (Ψ i ) / λ max (Ψ i ),i=1,2,3;λ max (Ψ) and λ min (Ψ) are the maximum and minimum eigenvalues ​​of Ψ respectively; let l i =λ min (Ψ i )-μ i λ max (Ψ i ) / 2; then:

[0172]

[0173] in, c0=min{λ1,λ2,λ3,λ4,λ5,λ6},λ1=-λ min (J1-(1 / 2+∈0+∈2+∈3)I),λ2=-λ min (J2),λ3=-λ min (G -1 -(1 / 2+∈4 / 2)I), λ4=l1, λ5=l2, λ6=l3;

[0174] S53, Order So Inequality (45) can be rewritten as:

[0175]

[0176] Integrating (46), we get:

[0177]

[0178] According to the conditions have And the following two conditions hold:

[0179] (1) z(t) is uniformly bounded, which means there exists a positive constant z0 such that z(t)≤z0;

[0180] (2) For a positive constant η>0, there exists a fixed time T η >0, so that 0<z(t)<η / 4, Through (1) and (2), we get:

[0181]

[0182] In addition, there are

[0183]

[0184] and

[0185]

[0186] Select T D >T η , so that get as well as Therefore, there is Right now where lim t→∞ η=0; therefore, we get

[0187]

[0188] S6. Realize the robust formation capture control of non-cooperative targets; the details are as follows: From (46) to (50), we get: If i∈F holds, then V(t) is uniformly bounded; more precisely, According to the definition of V(t), the state is uniformly bounded; because is a constant and bounded optimal matrix, so we get: is also bounded; therefore, the state of the entire system is bounded; in addition, according to ζ x(t), ζ v The definitions of (t) and χ1(t) are:

[0189] as well as μ1, μ2, μ3 are positive constants; in addition, we also get ζ x (t) and ζ v (t) represents the actual tracking error,

[0190] Therefore, we get From the equation And express η0=μ1+μ3 to get This means that the target capture control of multiple UAV systems (6)(7) under protocol (20) has been achieved.

[0191] In this example, consider a Figure 2 A topology with eight drones and three targets is shown.

[0192] Assume that the nonlinear terms and disturbances in the UAV dynamics model are:

[0193] m i (x i (t),t)=(3cos(0.5t)+2sin(0.1t)+5sin(0.05t))*0.1,

[0194] n i (x i (t),v i (t),t)=(sin(0.5t)+cos(0.05t)+sin(0.05t))*0.1

[0195] τ 1i (t)=(3sin(0.2t)+0.2icos(20t))*10 -3 ,τ 2i (t) = cos(0.2t)*10 -3 ,(i∈N)

[0196] Assume that the nonlinear terms and disturbances in the target dynamics model are:

[0197] p k (x k (t),t)=0.5sin(t)

[0198] τ 3k (t)=(3sin(0.2t)+0.2icos(20t))*10 -2 ,τ 4k(t)=(7sin(0.2t)+0.3icos(20t))*10 -2 ,h i (t) is set to h i (t) = [h ix (t),h iv (t)] T =[1.5cos(1.5t+2π(i-1) / 8),-2.25sin(1.5t+2π(i-1) / 8)] T

[0199] The parameters in the algorithm are ∈0=0.5,∈1=1.5,∈2=0.2,∈3=0.3,∈4=3,J 1ii =20, J 2ii =50, g ii =0.05, ρ0=ρ1=1. The parameters in the neural network are h=11, f i =5. Figure 3 It shows the drone's capture status of the target at t=0s, t=1s, t=20s, and t=60s.

[0200] Those skilled in the art will appreciate that the embodiments described herein are intended to help readers understand the principles of the present invention, and should be understood that the protection scope of the present invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific variations and combinations that do not deviate from the essence of the present invention based on the technical revelations disclosed by the present invention, and these variations and combinations are still within the protection scope of the present invention.

Claims

1. A non-cooperative target robust formation capture control method for a multi-UAV system, characterized in that: The following steps are involved: S1. Dynamics modeling of multi-UAV system and non-cooperative targets; details are as follows: S11. Construct the undirected communication topology between each drone in the multi-drone system and the directed sensing topology between the drone and the target; assume that there are N+M drones in total, of which M are targets and N are ordinary drones; F = {1, 2, ..., N} is the index set of ordinary drones, and E = {N+1, N+2, ..., N+M} is the index set of targets; It means that there is an undirected communication topology between drones and a directed sensing topology between drones and targets; is a set of points, is an edge set; is a weighted adjacency matrix, if there is an edge from node j to i in the set ε, then element a ij =1, otherwise 0; The edge in is represented by v ij =(v i ,v j ) indicates that i≠j; Represents node v i The neighbor set of is the in-degree matrix of the system, expressed as where d i Expressed as is the Laplace matrix of the system, Further expressed as: in It is the Laplace matrix unique to UAVs, reflecting the communication between them; is the Laplace matrix between the UAV and the target, reflecting the acquisition of target information by the UAV; The elements in l ij It is expressed as: S12. Use a neural network architecture to represent any continuous function f(x): f(x)=K T θ(x)+σ(x)(3) in is the neural network weight, is the radial basis function, represents a set of real numbers, h is the number of neurons; the i-th element θ in θ(x) i (x) = exp[-(xd i ) T (xd i ) / f i 2 ], i = 1, 2, ..., h, where d i =[d i1 ,d i2 ,...,d ih ] T is the center of the receptive field, f i is the width of the Gaussian function, is the neural network fitting error; The optimal neural network weights to minimize the fitting error; S13, the drone measures the target's status information x through the onboard sensor k and v k , can track the trajectory of the target in a certain formation configuration according to the information, and capture the target; adopt a distributed cooperative control scheme based on adaptive neural network estimation, so that the UAV can achieve the capture task of non-cooperative multiple targets when the target control input is unknown; Assume that at least one of the drones can obtain information about all targets through onboard sensors, and in the communication topology, the drone has a path to other drones; The state variables x of the drone and the target i , state change rate v i and its derivatives They are all bounded; If the multi-UAV system meets the following conditions, it means that the multi-UAV has completed the roundup control: in The status information of the drone; is the target status information; h i (t) = [h ix (t),h iv (t)] T is a predefined formation, h ix (t), h iv (t) are the position deviation and speed deviation that satisfy the formation, η0 is the tracking error, Time to achieve roundup; γ k To satisfy The coefficient of S14. Consider the following second-order nonlinear multi-UAV system dynamics model: where i∈F; u i (t)∈D u are the state and control input vector of UAV i respectively; m=1; item and is an unknown smooth bounded vector-valued function; suppose d(m i ) / dt is bounded; τ 1i (t) and τ 2i (t) is the disturbance, satisfying ‖τ 1i (t)‖≤d1 and‖τ 2i (t)‖≤d2, d1 and d2 are the upper bounds of the perturbation; The dynamic model of the target is: in are the state and control input of target k, k∈E; is an unknown smooth bounded vector-valued function; suppose d(p k ) / dt is bounded; τ 3k (t) and τ 4k (t) is the disturbance, satisfying ‖τ 3k (t)‖≤d3 and‖τ 4k (t)‖≤d4, d3 and d4 are the upper bounds of the perturbation; S2. Modeling of non-cooperative target capture errors of multi-UAV formations: Introducing auxiliary variables ζ x (t), ζ v (t),ψ x (t), where ζ x (t), ζ v (t) represent the actual state capture error and control input capture error, respectively, x (t) represents the tracking error of the target when the UAVs are in a certain formation; in u(t) and u E (t) are the control input signals of the UAV and the target respectively; col is defined as: k = N+1, N+2, ... N+M; S3. Design of matching / non-matching disturbance estimator based on neural network estimation: Use neural network to fit nonlinear terms in dynamic model; have: To minimize the fitting error, the optimal neural network weight matrix is ​​i=1,2,3; is the radial basis function, For K * The estimate, The estimation error is expressed as diag is defined as: is the neural network fitting error, which is a bounded value; S4. Design of adaptive robust formation capture controller for multi-UAV system; the details are as follows: S41. Design a distributed formation capture controller based on neural network. Before designing the controller, define several important variables: for UAV i, i∈N, we have: Oh 1i (t)=ζ xi (t), (12) In order to deal with the integral disturbance, a robust virtual controller is introduced as follows: J 1ii are the elements of the diagonal matrix J1, the J1 matrix satisfies J1-(∈1 / 2+∈0+∈2+ε3)I≥0, and ∈3 are positive parameters, I represents the identity matrix; Expressed as is a diagonal matrix Elements of and They are used to handle the position state disturbances of multiple UAVs and targets respectively. θ 1i (t) is θ 1i (ζ xi (t)) abbreviation; S42. To avoid the derivation of the above virtual controller, the following low-pass filter is introduced for the i-th UAV: where χ 1i (t) is the virtual control signal and filtered signal of the i-th UAV; g ii is the element of the diagonal matrix G, the matrix G satisfies G -1 -(1 / (2∈1)-∈4 / 2)I>0;∈4 is a positive parameter of the design; The filtering error is expressed as: Then get S43. Define the following parameters: The adaptive update rate of the neural network is: is a diagonal matrix The diagonal elements of For i=1,2,3, the diagonal matrix Ψ i Satisfy λ min (Ψ i )>0; is a positive definite diagonal matrix; S45. Based on the inspiration from (11)-(18), the control rate of UAV i (i∈N) is set to: ρ0,ρ1 are positive design parameters; J 2ii > 0, is the element of the diagonal matrix J2; is the adaptive parameter of the symbolic function, Respectively for estimates; In the topology diagram Under this condition, by adopting a low-pass filter (15) and a distributed capture controller based on a neural network, the multi-UAV system (6) can achieve capture control of multiple targets (7), that is, satisfying conditions (4) and (5).

Images