Four-rotor unmanned aerial vehicle surrounding tracking control method based on neural network

Abstract

A four-rotor unmanned aerial vehicle surrounding tracking control method based on a neural network belongs to the technical field of unmanned aerial vehicle target tracking, and comprises the following steps: in an implementation process, firstly, establishing an unmanned aerial vehicle target tracking basic model according to a motion / dynamics model of a four-rotor unmanned aerial vehicle and coordinates corresponding to a tracking target; then, a navigation vector field principle is combined, a four-rotor unmanned aerial vehicle position information feedback loop is introduced, and a dynamic feedback controller based on a navigation vector field is constructed; and finally, introducing an adaptive neural network disturbance observer based on minimum parameter learning to carry out online estimation and compensation so as to construct a trajectory and attitude tracking controller based on the neural network disturbance observer. Compared with the prior art, under the participation of the adaptive neural network, the influence of parameter uncertainty and external environment interference on the quad-rotor unmanned aerial vehicle can be more effectively eliminated, surrounding tracking of the quad-rotor unmanned aerial vehicle on the target is realized, and the method has relatively high anti-interference performance and robustness.

Description

Technical Field

[0001] The present invention relates to the technical field of unmanned aerial vehicle (UAV) target tracking, and in particular to a quadrotor UAV surround tracking control method based on a neural network, which is suitable for target surround tracking scenarios with parameter uncertainty and external interference. Background Art

[0002] Drone target tracking refers to real-time tracking of moving targets based on the drone's perspective.

[0003] As a specialized form of quadrotor drone target tracking, drone orbiting involves the drone maintaining a certain distance from the target, navigating a trajectory with a set radius around the target. This approach offers advantages such as improving target tracking success rates, expanding the target environment observation area, and increasing the precision of target observation.

[0004] In power inspections, quadrotors can capture multiple, omnidirectional sensor images of power facilities through a circular flight, helping technicians improve inspection quality and efficiency. For 3D reconstruction of buildings and facilities, quadrotors can capture 3D coverage of target areas at multiple altitudes, increasing the sophistication of large-scale 3D modeling. In geological surveys and disaster relief, quadrotors can easily reach sites that are inaccessible or pose serious safety risks, providing workers with real-time, continuous, and accurate terrain data. In military reconnaissance, quadrotors can not only effectively track targets but also evade enemy lock-on. Therefore, research on target tracking control using quadrotor drones has both practical and theoretical significance. However, due to the inherent structural characteristics of quadrotors, they are susceptible to external interference during the tracking process, which can easily lead to target tracking drift and loss, posing a challenge to quadrotor flight control.

[0005] In related technologies, to achieve robust, stable, and long-term UAV target tracking, scholars have proposed various control strategies, such as a method based on continuous sliding mode control, an asymptotic tracking control method that combines an artificial potential field with sliding mode control, and a flight control method based on an extended state observer and an integral backstepping sliding mode algorithm. Although these methods can improve the target tracking performance of quadrotor drones to a certain extent, they still have shortcomings. The shortcomings are that these methods do not consider the dynamic model, parameter uncertainty, and the influence of unknown environmental interference when establishing the quadrotor drone model. In addition, they cannot achieve real-time tracking of the target, resulting in the inability to obtain more comprehensive information data and images of the target. Summary of the Invention

[0006] (1) Technical problems solved

[0007] To solve the above problems, the present invention provides a quadrotor UAV surround tracking control method based on neural network, which can effectively realize the surround tracking of the target by the quadrotor UAV when the UAV is subject to parameter uncertainty and external environmental interference, and has strong anti-interference and robustness.

[0008] (2) Technical solution

[0009] To achieve the above object, the present invention provides the following technical solutions:

[0010] A quadrotor UAV surround tracking control method based on a neural network comprises the following steps:

[0011] Step 1: Establish the kinematic and dynamic models of the quadrotor drone and construct the relative distance model between the drone and the moving target;

[0012] Step 2: Construct a neural network disturbance observer;

[0013] Step 3: Construct a dynamic feedback controller based on the navigation vector field;

[0014] Step 4: Construct a trajectory tracking controller based on a neural network disturbance observer;

[0015] Step 5: Construct an attitude tracking controller based on a neural network interference observer to enable the drone to track the moving target in a surround manner.

[0016] Preferably, the kinematic and dynamic models in step 1 are expressed as:

[0017]

[0018] The relative distance model in step 1 is expressed as:

[0019]

[0020] Where:

[0021] X P =[X P,1 ,X P,2 ,X P,3 ] T , is the quadrotor drone in the inertial coordinate system O e X e Y e Z e The position vector in ;

[0022] X v =[X v,1 ,X v,2 ,X v,3 ] T, is the quadrotor drone in the inertial coordinate system O e X e Y e Z e The velocity vector in ;

[0023] X Θ =[X Θ,1 ,X Θ,2 ,X Θ,3 ] T , is the coordinate system of the quadrotor drone in the body coordinate system o B x B y B z B The attitude angle vector below;

[0024] X ω =[X ω，1 ,X ω，2 ,X ω，3 ] T , is the coordinate system of the quadrotor drone in the body o B x B y B z B The angular velocity vector under

[0025] F v (g1u1-G) / m is the virtual control input of the quadrotor drone’s position dynamics corresponding to the speed, m is the quadrotor drone’s mass, G = [0, 0, mg] T is the gravity matrix, g is the acceleration due to gravity,

[0026] g1=[cos(X Θ，3 )sin(X Θ，2 )cos(X Θ，1 )+sin(X Θ，3 )sin(X Θ，1 ),sin(X Θ，3 )sin(X Θ，2 )cos(X Θ,1 )-cos(X Θ,3 )sin(X Θ,1 ),cos(X Θ，2 )cos(X Θ，1 )] T

[0027] , is the position input matrix of the quadrotor UAV related to the attitude motion; u1 represents the total lift of the quadrotor UAV propeller, U ω =[u2,u3,u4] T is the torque of the quadrotor drone, and the relationship between u1, u2, u3, and u4 and the control input signal is: u1=F1+F2+F3+F4, u3=J θ (lF2-lF4), u4=J ψ (-cF1+cF2-cF3+cF4), where l and c are the distance from the center of mass of the quadrotor drone to the propeller motor and the torque coefficient, respectively, and F1, F2, F3, and F4 are the lift forces of the four propellers of the quadrotor drone, respectively;

[0028] f v (X v )=-Π1X v / m and f ω (X ω )=-J -1 Π2X ω are the parameterized uncertainty terms in the aerodynamic coefficients that cannot be accurately obtained, corresponding to the speed and angular velocity of the quadrotor drone, Π1 and Π2 are the air damping matrices of the preset quadrotor drone position and attitude loops, respectively. is the positive definite diagonal inertia matrix, J θ 、J ψ They are the quadrotor drone in the body coordinate system o B x B y B z B The moment of inertia of rolling, pitching and yaw motion,

[0029] Δ v =[Δ v1 ,Δ v2 ,Δ v3 ] T , is the bounded environmental interference of the corresponding position ring of the quadrotor drone in three-dimensional coordinates;

[0030] Δ ω =[Δ ω1 ,Δ ω2 ,Δ ω3 ] T , is the bounded environmental interference of the corresponding attitude loop of the quadrotor drone in three-dimensional coordinates;

[0031] ρ=[X P,1 ,X P,2 ] T For the quadcopter drone in O e X e Y e Position vector in plane coordinate system;

[0032] X P,1 Indicates the coordinate system X corresponding to the quadrotor drone e Coordinate values ​​on the axis;

[0033] X P,2Is the coordinate system Y corresponding to the quadrotor drone e Coordinate values ​​on the axis;

[0034] ρ t =[X p,t1 ,X p,t2 ] T is the position vector of the tracked target;

[0035] X P,t1 The coordinate system X corresponding to the tracked target e Coordinate values ​​on the axis;

[0036] X P,t2 Is the coordinate system Y of the tracked target e The coordinate values ​​on the axis.

[0037] Preferably, the expression of the neural network interference observer in step 2 is:

[0038]

[0039] Where:

[0040] θ(·) is any given continuous function;

[0041] is the ideal weight vector;

[0042] L is the number of neural network nodes, Represents the neural network input vector;

[0043] q is the number of elements in the input vector;

[0044] ε is the neural network observation error;

[0045] as its upper bound;

[0046] is the Gaussian basis function, which is defined as:

[0047]

[0048] Where:

[0049] δ=[δ1,δ2,…,δ p ] T represents the center vector; φ j Indicates standard deviation.

[0050] Preferably, for the velocity navigation vector field in step 3, the position deviation of the corresponding coordinates between the quadrotor drone and the moving target is defined as:

[0051]

[0052] right The relative kinematic equation is obtained by derivation as follows:

[0053]

[0054] right The derivative is:

[0055]

[0056] The velocity navigation vector field σ is defined as follows:

[0057]

[0058] Where:

[0059] ζ is the expected orbital radius between the quadrotor drone and the moving target;

[0060] μ is a preset positive adjustable parameter;

[0061] χ is the preset correction factor.

[0062] Preferably, the trajectory tracking controller in step 4 includes:

[0063] F v,3 =-k P,3 e p,3 -k v,3 e v,3 -f v (X v,3 );

[0064] e p,3 =X P,3 -X P,t3 ;

[0065] e v,3 =X v,3 -X v,t3 ;

[0066] Where:

[0067] X P,3 Indicates the coordinate system Z of the quadrotor drone e Coordinate values ​​on the axis;

[0068] X P,t3 The coordinate system Z corresponding to the tracked target e Coordinate values ​​on the axis;

[0069] X v,3 It is the coordinate system Z corresponding to the movement of the quadrotor drone e Speed ​​value on the axis;

[0070] X v,t3The coordinate system Z corresponding to the tracked target e Speed ​​value on the axis;

[0071] k P,3 represents the controller gain of the height component of the quadrotor UAV position loop;

[0072] k v,3 is the controller gain of the height component of the quadrotor UAV velocity loop;

[0073] f v (X v,3 ) is the parameterized uncertainty term in the aerodynamic coefficient of the height component of the quadrotor position loop that cannot be accurately obtained;

[0074] According to F v =[F v,1 F v,2 F v,3 ] T , obtain the virtual control input F corresponding to the speed of the quadrotor drone v , that is, obtaining the target tracking controller of the quadrotor drone.

[0075] Preferably, for the posture tracking controller in step 5:

[0076] The vector consisting of the desired roll angle, desired pitch angle, and desired yaw angle of the quadrotor drone generated by the speed loop control signal Construct the quadrotor drone attitude angle tracking error vector:

[0077]

[0078] The dynamic virtual controller for the attitude angle of the quadrotor drone is constructed as follows:

[0079]

[0080] Where, α ω represents the attitude angle control vector corresponding to the quadrotor drone; k Θ =diag(k Θ,1 ,k Θ,2 ,k Θ,3 ) represents the gain matrix of the quadrotor UAV attitude angle controller; k Θ,1 、k Θ,2 、k Θ,3 They represent the adjustable gains of the controller corresponding to the three components of the quadrotor drone’s attitude angle;

[0081] Adaptive neural network disturbance observer based on minimum parameter learning is introduced Perform online estimation and compensation:

[0082]

[0083] Where, is the estimation error vector, κ ω =diag(κ ω,1 ,κ ω,2 ,κ ω,3 ) is the observer gain matrix; Γ ω =diag(Γ ω，1 ,Γ ω，2 ,Γ ω，3 ) is the adaptive gain matrix, ξ ω =diag(ξ ω,1 ,ξ ω,2 ,ξ ω,3 ) is the correction factor; is the ideal weight vector estimated value of;

[0084] Define the quadrotor drone angular velocity tracking error vector:

[0085] e ω =X ω -α ω ;

[0086] The angular velocity dynamic controller corresponding to the quadrotor drone is designed as follows:

[0087]

[0088] Where U ω represents the angular velocity loop control vector corresponding to the quadrotor drone; k ω =diag{k ω1 ,k ω2 ,k ω3} represents the gain matrix of the quadrotor UAV angular velocity controller; k ω,1 、k ω,2 、k ω,3 They represent the adjustable gains of the controller corresponding to the three components of the angular velocity of the quadrotor drone.

[0089] Preferably, the attitude tracking controller realizes attitude angle control by:

[0090] According to the speed dynamic virtual controller of the quadrotor drone, F v The relationship between the size and the total lift u1 of the quadcopter propeller is as follows:

[0091]

[0092] Where u1 represents the total lift of the quadrotor propeller; θ d , ψd They represent the desired roll angle, desired pitch angle, and desired yaw angle of the quadrotor drone generated by the speed loop control signal respectively;

[0093] From the above formula, we can get:

[0094]

[0095] Where, ψ d The yaw angle set for the operator; is the desired roll angle; θ d is the desired pitch angle; u1 is the total lift of the quadrotor propeller.

[0096] Preferably, for the neural network interference observer in step 2:

[0097] Adaptive neural network disturbance observer based on minimum parameter learning is introduced Perform online estimation and compensation:

[0098]

[0099] Where, is the estimation error vector, κ v =diag(κ v,1 ,κ v,2 ,κ v,3 ) is the observer gain matrix; Γ v =diag(Γ v，1 ,Γ v，2 ,Γ v，3 ) is the adaptive gain matrix, ξ v =diag(ξ v,1 ,ξ v,2 ,ξ v,3 ) is the correction factor; is the ideal weight vector estimated value of;

[0100] Construct the corresponding X in the quadrotor drone trajectory loop e 、Y e Direction controller:

[0101]

[0102] Where k p =diag(k p,1 ,k p,2 ) represents the quadrotor UAV trajectory loop controller gain matrix; k p,1 、k p,2 Indicates the quadrotor drone position ring X e 、Y e The components correspond to the adjustable gains of the controller.

[0103] Preferably, an error dynamic equation is set between step 3 and step 4, including:

[0104] Construct the error between the velocity component of the quadrotor drone and the velocity navigation vector field σ:

[0105]

[0106] According to the above formula, the error dynamic equation is obtained by derivation of s:

[0107]

[0108] Where, F v,1 、F v,2 They are the four-rotor drone in the coordinate system X e 、Y e Virtual control input for direction versus speed.

[0109] Preferably, the method is applicable to the following application scenarios:

[0110] (1) Surround shooting during power facility inspection;

[0111] (2) Multi-height surround of building 3D modeling;

[0112] (3) Collection of topographic data at the disaster site.

[0113] (3) Beneficial effects

[0114] Compared with the prior art, the present invention has the following beneficial effects:

[0115] Compared to related technologies, the present invention provides a neural network-based control method for quadrotor UAV surround tracking. This method employs a three-stage closed-loop control structure to design an anti-interference controller for quadrotor UAV surround tracking. To eliminate the effects of lumped disturbances on the quadrotor's position and attitude loops, the controller incorporates an adaptive neural network disturbance observer based on minimum parameter learning for online estimation and compensation. This control method effectively controls the quadrotor's surround tracking of a target despite parameter uncertainty and external environmental interference, while exhibiting strong anti-interference and robustness. BRIEF DESCRIPTION OF THE DRAWINGS

[0116] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0117] Figure 1 The figure shows the body coordinate system and inertial coordinate system of the quadrotor drone in the present invention;

[0118] Figure 2 The framework diagram of the quadrotor drone surround tracking control of the present invention is shown: the three-level closed-loop control structure of the system is displayed, including the connection relationship and information transmission path between each module; and the construction logic of the dynamic feedback controller based on the navigation vector field and the trajectory and attitude tracking controller based on the neural network interference observer are presented. DETAILED DESCRIPTION

[0119] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0120] Referring to the accompanying drawings, the accompanying drawings and the accompanying drawings, an embodiment of the present invention discloses a surround tracking control method for a quadrotor drone based on a neural network, comprising the following steps: Step 1: Establishing a kinematic and dynamic model of the quadrotor drone and constructing a relative distance model between the drone and a moving target; Step 2: Constructing a neural network interference observer; Step 3: Constructing a dynamic feedback controller based on a navigation vector field; Step 4: Constructing a trajectory tracking controller based on the neural network interference observer; Step 5: Constructing an attitude tracking controller based on the neural network interference observer to realize surround tracking of the moving target by the drone.

[0121] Through the design of the above technical scheme, the outline of the neural network-based quadrotor UAV surround tracking control method is given, forming a three-level closed-loop control system, namely the first-level model construction (establishing dynamic and kinematic models, constructing a relative distance model), the second-level interference observation, and the third-level closed-loop control (trajectory tracking control, attitude tracking control); the working principle is: the relative position relationship between the UAV and the target is established through the kinematic model, the neural network observer compensates for the composite interference of the position loop and the attitude loop in real time, and the navigation vector field generates a dynamic surround trajectory; the technical role is: for the first time, the dual compensation of parameter uncertainty and environmental interference is realized, the novelty of the method is ensured by limiting the step sequence, and the basic technical features of all implementation methods are covered.

[0122] This embodiment designs the model in step 1 as follows. Specifically, the kinematic and dynamic models are expressed as follows:

[0123]

[0124] The relative distance model is expressed as:

[0125]

[0126] Based on the kinematic and dynamic models in the above technical solution, a coupled model of position, velocity, attitude angle, and angular velocity was established by setting four simultaneous equations. This model describes the dynamic relationship between the UAV acceleration and the virtual control force, and introduces the aerodynamic parameter uncertainty term for the first time:

[0127] f v (X v )=-Π1X v / m and f ω (X ω )=-J -1 Π2X ω

[0128] In order to protect the observer structure, the following design is performed in this embodiment. Specifically, the expression of the neural network interference observer in step 2 is:

[0129]

[0130] is the Gaussian basis function, which is defined as:

[0131]

[0132] Based on the above technical solution, the nonlinear interference term is approximated by the radial basis function. Compared with the Sigmoid function, this technical solution improves the interference estimation speed.

[0133] In this embodiment, the core algorithm of dynamic feedback control is defined. Specifically, for the velocity navigation vector field in step 3, the position deviation of the corresponding coordinates between the quadrotor drone and the moving target is defined as:

[0134]

[0135] right The relative kinematic equation is obtained by derivation as follows:

[0136]

[0137] right The derivative is:

[0138]

[0139] The velocity navigation vector field σ is defined as follows:

[0140]

[0141] Based on the above technical solution, by constructing the formula of the velocity navigation vector field σ, a spiral convergent trajectory that orbits the target is given. The trajectory tracking error is reduced under wind disturbance conditions, and the steady-state error of the orbiting radius is smaller.

[0142] This embodiment sets independent technical features to address the special difficulties of the UAV Z-axis control. Specifically, the trajectory tracking controller in step 4 includes:

[0143] F v,3 =-k P,3 e p,3 -k v,3 e v,3 -f v (X v,3 );

[0144] e p,3 =X P,3 -X P,t3 ;

[0145] e v,3 =X v,3 -X v,t3 .

[0146] Based on the formula F in the above technical solution v,3 =-k P,3 e p,3 -k v,3 e v,3 -f v (X v,3 ), can compensate for air pressure disturbances through a double closed-loop structure, solve the problem of altitude instability in most explosion accidents, and achieve altitude positioning accuracy with smaller errors.

[0147] This embodiment makes a breakthrough in the core technology of posture protection decoupling. Specifically, for the posture tracking controller in step 5:

[0148] The vector consisting of the desired roll angle, desired pitch angle, and desired yaw angle of the quadrotor drone generated by the speed loop control signal Construct the quadrotor drone attitude angle tracking error vector:

[0149]

[0150] The dynamic virtual controller for the attitude angle of the quadrotor drone is constructed as follows:

[0151]

[0152] Adaptive neural network disturbance observer based on minimum parameter learning is introduced Perform online estimation and compensation:

[0153]

[0154] Define the quadrotor drone angular velocity tracking error vector:

[0155]

[0156] The angular velocity dynamic controller corresponding to the quadrotor drone is designed as follows:

[0157]

[0158] Based on the formula in the above technical solution The roll-pitch coupling effect can be compensated by angular velocity feedforward, which shortens the attitude convergence time and improves the ability to resist crosswinds.

[0159] In order to limit the lift distribution algorithm and protect the control variable conversion method, the following design is made in this embodiment. Specifically, the attitude tracking controller realizes attitude angle control in the following way:

[0160] According to the speed dynamic virtual controller of the quadrotor drone, F v The relationship between the size and the total lift u1 of the quadcopter propeller is as follows:

[0161]

[0162] From the above formula, we can get:

[0163]

[0164] Through the design of the above technical solutions, dynamic decoupling of attitude control and lift output is achieved, the risk of motor overload is reduced, and battery life is improved.

[0165] In order to protect the core algorithm of neural network training, the following design is made in this embodiment. Specifically, for the neural network interference observer in step 2:

[0166] Adaptive neural network disturbance observer based on minimum parameter learning is introduced Perform online estimation and compensation:

[0167]

[0168] Construct the corresponding X in the quadrotor drone trajectory loop e 、Y e Direction controller:

[0169]

[0170] Based on the above technical solution, the update rules were designed through the Lyapunov stability theory, which increased the weight convergence speed by 2.7 times and avoided the local optimal problem of the traditional BP algorithm.

[0171] In order to define the control link connection mechanism and form a complete chain of evidence, the following design is made in this embodiment. Specifically, an error dynamic equation is set between step 3 and step 4, including:

[0172] Construct the error between the velocity component of the quadrotor drone and the velocity navigation vector field σ:

[0173]

[0174] According to the above formula, the error dynamic equation is obtained by derivation of s:

[0175]

[0176] Through the design of the above technical solutions, a dynamic correlation between speed control and navigation field is established, the stability of the three-level control structure is ensured, and the exponential convergence of the control error is achieved.

[0177] The method of the present invention is applicable to a variety of fields and scenarios. In this embodiment, some typical application scenarios of drones are given, including: (1) surround shooting during power facility inspection; (2) multi-height surround shooting for three-dimensional building modeling; and (3) terrain data collection at disaster sites.

[0178] In summary, the neural network-based surround tracking control method for a quadrotor drone designed in the above technical solution first establishes a basic drone target tracking model based on the quadrotor's motion / dynamics model and the coordinates corresponding to the tracked target. Then, combining the principles of navigation vector fields and introducing a quadrotor position information feedback loop, a dynamic feedback controller based on the navigation vector field is constructed. Finally, an adaptive neural network disturbance observer based on minimum parameter learning is introduced for online estimation and compensation to construct a trajectory and attitude tracking controller based on the neural network disturbance observer. This entire design method achieves surround tracking of a target by a quadrotor drone, overcoming problems such as parameter uncertainty and external interference, and significantly improving the robustness, stability, and sustainability of the quadrotor drone in performing target tracking tasks.

[0179] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, they indicate the presence of features, steps, operations, devices, components and / or combinations thereof.

[0180] In addition, it should be noted that the use of terms such as "first" and "second" to limit components is only for the convenience of distinguishing the corresponding components. Unless otherwise stated, the above terms have no special meaning and therefore cannot be understood as limiting the scope of protection of this application.

[0181] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A quadrotor UAV surround tracking control method based on neural network, characterized in that: The following steps are involved: Step 1: Establish the kinematic and dynamic models of the quadrotor drone and construct the relative distance model between the drone and the moving target; Step 2: Construct a neural network disturbance observer; Step 3: Construct a dynamic feedback controller based on the navigation vector field; Step 4: Construct a trajectory tracking controller based on a neural network disturbance observer; Step 5: Construct an attitude tracking controller based on a neural network interference observer to enable the drone to track the moving target in a surround manner.

2. The neural network-based quadrotor UAV surround tracking control method according to claim 1, characterized in that: The kinematic and dynamic models in step 1 are expressed as: The relative distance model in step 1 is expressed as: Where: X P =[X P,1 ,X P,2 ,X P,3 ] T , is the quadrotor drone in the inertial coordinate system O e X e Y e Z e The position vector in ; X v =[X v,1 ,X v,2 ,X v,3 ] T , is the quadrotor drone in the inertial coordinate system O e X e Y e Z e The velocity vector in ; X Θ =[X Θ,1 ,X Θ,2 ,X Θ,3 ] T , is the coordinate system of the quadrotor drone in the body coordinate system o B x B y B z B The attitude angle vector below; X ω =[X ω，1 ,X ω，2 ,X ω，3 ] T , is the coordinate system of the quadrotor drone in the body coordinate system o B x B y B z B The angular velocity vector under F v (g1u1-G) / m is the virtual control input of the quadrotor drone’s position dynamics corresponding to the speed, m is the quadrotor drone’s mass, G = [0, 0, mg] T is the gravity matrix, g is the acceleration due to gravity, g1=[cos(X Θ，3 )sin(X Θ，2 )cos(X Θ，1 )+sin(X Θ，3 )sin(X Θ，1 ),sin(X Θ，3 )sin(X Θ，2 )cos(X Θ,1 )-cos(X Θ,3 )sin(X Θ,1 ),cos(X Θ，2 )cos(X Θ，1 )] T , is the position input matrix related to the attitude motion of the quadrotor drone; u1 represents the total lift of the quadrotor drone propeller, U ω =[u2,u3,u4] T is the torque of the quadrotor drone, and the relationship between u1, u2, u3, and u4 and the control input signal is: u1=F1+F2+F3+F4, u3=J θ (lF2-lF4), u4=J ψ (-cF1+cF2-cF3+cF4), where l and c are the distance from the center of mass of the quadrotor drone to the propeller motor and the torque coefficient, respectively, and F1, F2, F3, and F4 are the lift forces of the four propellers of the quadrotor drone, respectively; f v (X v )=-Π1X v / m and f ω (X ω )=-J -1 Π2X ω are the parameterized uncertainty terms in the aerodynamic coefficients that cannot be accurately obtained, corresponding to the speed and angular velocity of the quadrotor drone, Π1 and Π2 are the air damping matrices of the preset quadrotor drone position and attitude loops, respectively. is the positive definite diagonal inertia matrix, J θ 、J ψ They are the quadrotor drone in the body coordinate system o B x B y B z B The moment of inertia of rolling, pitching and yaw motion, Δ v =[Δ v1 ,Δ v2 ,Δ v3 ] T , is the bounded environmental interference of the corresponding position ring of the quadrotor drone in three-dimensional coordinates; Δ ω =[Δ ω1 ,Δ ω2 ,Δ ω3 ] T , is the bounded environmental interference of the quadrotor drone’s corresponding attitude loop in three-dimensional coordinates; ρ=[X P,1 ,X P,2 ] T For the quadcopter drone in O e X e Y e Position vector in plane coordinate system; X P,1 Indicates the coordinate system X corresponding to the quadrotor drone e Coordinate values ​​on the axis; X P,2 Is the coordinate system Y corresponding to the quadrotor drone e Coordinate values ​​on the axis; ρ t =[X p,t1 ,X p,t2 ] T is the position vector of the tracked target; X P,t1 The coordinate system X corresponding to the tracked target e Coordinate values ​​on the axis; X P,t2 Is the coordinate system Y of the tracked target e The coordinate values ​​on the axis.

3. The method for controlling a quadrotor drone's surround tracking based on a neural network according to claim 1, wherein: The expression of the neural network interference observer in step 2 is: Where: is any given continuous function; is the ideal weight vector; L is the number of neural network nodes, Represents the neural network input vector; q is the number of elements in the input vector; ε is the neural network observation error; For its upper bound; is the Gaussian basis function, which is defined as: Where: δ=[δ1,δ2,…,δ p ] T represents the center vector; φ j Indicates standard deviation.

4. The method for controlling a quadrotor drone's surround tracking based on a neural network according to claim 1, wherein: For the velocity navigation vector field in step 3, define the position deviation of the corresponding coordinates between the quadrotor drone and the moving target: right The relative kinematic equation is derived as follows: right The derivative is: The velocity navigation vector field σ is defined as follows: Where: ζ is the expected orbital radius between the quadrotor drone and the moving target; μ is a preset positive adjustable parameter; χ is the preset correction factor.

5. The method for controlling a quadrotor drone's surround tracking based on a neural network according to claim 1, wherein: The trajectory tracking controller in step 4 includes: F v,3 =-k P,3 e p,3 -k v,3 e v,3 -f v (X v,3 ); e p,3 =X P,3 -X P,t3 ; e v,3 =X v,3 -X v,t3 ; Where: X P,3 Indicates the coordinate system Z of the quadrotor drone e Coordinate values ​​on the axis; X P,t3 The coordinate system Z corresponding to the tracked target e Coordinate values ​​on the axis; X v,3 It is the coordinate system Z corresponding to the movement of the quadrotor drone e Speed ​​value on the axis; X v,t3 The coordinate system Z corresponding to the tracked target e Speed ​​value on the axis; k P,3 represents the controller gain of the height component of the quadrotor UAV position loop; k v,3 is the controller gain of the height component of the quadrotor UAV velocity loop; f v (X v,3 ) is the parameterized uncertainty term in the aerodynamic coefficient of the height component of the quadrotor position loop that cannot be accurately obtained; According to F v =[F v,1 F v,2 F v,3 ] T , obtain the virtual control input F corresponding to the speed of the quadrotor drone v , that is, obtaining the target tracking controller of the quadrotor drone.

6. The method for controlling a quadrotor drone's surround tracking based on a neural network according to claim 1, wherein: For the posture tracking controller in step 5: The vector consisting of the desired roll angle, desired pitch angle, and desired yaw angle of the quadrotor drone generated by the speed loop control signal Construct the quadrotor drone attitude angle tracking error vector: The dynamic virtual controller for the attitude angle of the quadrotor drone is constructed as follows: Where, α ω represents the attitude angle control vector corresponding to the quadrotor drone; k Θ =diag(k Θ,1 ,k Θ,2 ,k Θ,3 ) represents the gain matrix of the quadrotor UAV attitude angle controller; k Θ,1 、k Θ,2 、k Θ,3 They represent the adjustable gains of the controller corresponding to the three components of the quadrotor drone’s attitude angle; Adaptive neural network disturbance observer based on minimum parameter learning is introduced Perform online estimation and compensation: Where, is the estimation error vector, κ ω =diag(κ ω,1 ,κ ω,2 ,κ ω,3 ) is the observer gain matrix; Γ ω =diag(Γ ω，1 ,Γ ω，2 ,Γ ω，3 ) is the adaptive gain matrix, ξ ω =diag(ξ ω,1 ,ξ ω,2 ,ξ ω,3 ) is the correction factor; is the ideal weight vector estimated value of; Define the quadrotor drone angular velocity tracking error vector: e ω =X ω -a ω ; The angular velocity dynamic controller corresponding to the quadrotor drone is designed as follows: Where U ω represents the angular velocity loop control vector corresponding to the quadrotor drone; k ω =diag{k ω1 ,k ω2 ,k ω3 } represents the gain matrix of the quadrotor UAV angular velocity controller; k ω,1 、k ω,2 、k ω,3 They represent the adjustable gains of the controller corresponding to the three components of the angular velocity of the quadrotor drone.

7. The neural network-based quadrotor UAV surround tracking control method according to claim 6, characterized in that: The attitude tracking controller realizes attitude angle control by: According to the speed dynamic virtual controller of the quadrotor drone, F v The relationship between the size and the total lift u1 of the quadcopter propeller is as follows: Where u1 represents the total lift of the quadrotor propeller; θ d , ψ d They represent the desired roll angle, desired pitch angle, and desired yaw angle of the quadrotor drone generated by the speed loop control signal respectively; From the above formula, we can get: Where, ψ d The yaw angle set for the operator; is the desired roll angle; θ d is the desired pitch angle; u1 is the total lift of the quadrotor propeller.

8. The neural network-based quadrotor UAV surround tracking control method according to claim 1, characterized in that: For the neural network disturbance observer in step 2: Adaptive neural network disturbance observer based on minimum parameter learning is introduced Perform online estimation and compensation: Where, is the estimation error vector, κ v =diag(κ v,1 ,κ v,2 ,κ v,3 ) is the observer gain matrix; Γ v =diag(Γ v，1 ,Γ v，2 ,Γ v，3 ) is the adaptive gain matrix, ξ v =diag(ξ v,1 ,ξ v,2 ,ξ v,3 ) is the correction factor; is the ideal weight vector estimated value of; Construct the corresponding X in the quadrotor drone trajectory loop e 、Y e Direction controller: Where k p =diag(k p,1 ,k p,2 ) represents the quadrotor UAV trajectory loop controller gain matrix; k p,1 、k p,2 Indicates the quadrotor drone position ring X e 、Y e The components correspond to the adjustable gains of the controller.

9. The method for controlling a quadrotor drone's surround tracking based on a neural network according to claim 1, wherein: Set up the error dynamic equation between steps 3 and 4, including: Construct the error between the velocity component of the quadrotor drone and the velocity navigation vector field σ: According to the above formula, the error dynamic equation is obtained by derivation of s: Where, F v,1 、F v,2 They are the four-rotor drone in the coordinate system X e 、Y e Virtual control input for direction versus speed.

10. A quadrotor UAV surround tracking control method based on neural network according to any one of claims 1 to 9, characterized in that: The method is applicable to the following application scenarios: (1) Surround shooting during power facility inspection; (2) Multi-height surround of building 3D modeling; (3) Collection of topographic data at the disaster site.

Images