Unmanned aerial vehicle control method, server and system in unmanned aerial vehicle cluster

Abstract

The invention relates to an unmanned aerial vehicle control method in an unmanned aerial vehicle cluster, a server and a system, and belongs to the field of unmanned aerial vehicle cluster path planning control. According to the technical scheme, the method comprises the steps that path ideal input parameters are designed for each unmanned aerial vehicle based on an artificial potential field method and an unmanned aerial vehicle kinetic model, each unmanned aerial vehicle optimizes nonlinear terms of a single unmanned aerial vehicle system through a neural network, and a first control law obtained after the nonlinear terms of the single unmanned aerial vehicle system are optimized is constructed; and optimizing the first control law through a gradient descent method to obtain a second control law, and controlling the unmanned aerial vehicle through the virtual control input equation by the second control law. By designing path control planning based on an artificial potential field method and flight sliding mode control planning based on a neural network, a multi-unmanned aerial vehicle cluster path planning effect can be realized, and the anti-interference capability and robustness of the system are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of path planning and control of unmanned aerial vehicle clusters, and in particular relates to a method, a server and a system for controlling unmanned aerial vehicles in an unmanned aerial vehicle cluster. Background Art

[0002] With the rapid development of drone technology, rotorcraft drones, which can carry many types of sensors and gimbals to perform flight missions, have played an indispensable role in various fields when faced with tasks that humans cannot accomplish.

[0003] The current method used for drones to identify obstacles and avoid them to continue flying mainly combines GPS and visual sensors to estimate the system status of the drone, infer whether there are obstacles and plan paths. However, this method is difficult to implement in urban environments where high-rise buildings exist, and it is easy to cause errors in system state estimation when encountering dynamic obstacles. For this reason, the prior art proposes an autonomous obstacle avoidance method for drones based on an artificial potential field method for single drones to overcome the problem of poor obstacle avoidance effect of drone status monitoring using GPS and visual sensors.

[0004] However, in actual drone applications, single drones have certain limitations. At this time, quadrotor drone clusters have attracted widespread attention. As a large and complex system, quadrotor drone clusters may encounter obstacles such as buildings and mountains during actual flight. The existence of these obstacles will threaten the flight safety of the quadrotor drone cluster. The quadrotor drone cluster needs to be able to make different decisions according to different environmental conditions to avoid threatening obstacles and other quadrotor drones at the same time. However, the stability and robustness of the obstacle avoidance method in the existing technology needs to be strengthened, and the analytical calculation is complex and requires global information. Summary of the invention

[0005] In view of the problems in the prior art, the present invention provides a method, server and system for controlling drones in a drone cluster, which solves the problem of insufficient stability and robustness when performing path planning and obstacle avoidance for a quad-rotor drone cluster in the prior art control method.

[0006] The technical solution adopted by the present invention is as follows: In a first aspect, the present application provides a method for controlling drones in a drone cluster, comprising the following steps: S1. Perform force analysis on the quadrotor drone and establish a virtual control input equation; S2, establish the dynamic model equation of the quadrotor drone in combination with the coordinate system rotation matrix; S3, design the ideal input parameters of the path based on the artificial potential field method and the UAV dynamics model; S4, optimizing the nonlinear terms of the single UAV system through neural network; S5, constructing the first control law after optimizing the nonlinear terms of the single UAV system; S6. Optimize the first control law by using the gradient descent method to obtain the second control law, and control the UAV through a virtual control input equation based on the second control law.

[0007] In an optional embodiment, the virtual control input equation is:

[0008] in, is the lift generated by each rotor of the quadcopter, i=1, 2, 3, 4, to is the rotation speed of the four rotors, is the lift generated by the rotor and, Represents the input value of the roll angle channel, Represents the input value of the pitch angle channel, represents the input value of the yaw angle channel, and b represents the anti-torque coefficient of the propeller motor; The dynamic model equation of the quadrotor drone is:

[0009] Among them, x, y, and z are the position variables in the directions of the three coordinate axes. , , is the linear velocity of the UAV on the three coordinate axes, , , is the linear acceleration of the UAV on the three coordinate axes, respectively The pitch, roll and yaw angles of the drone. are the pitch angular velocity, roll angular velocity and yaw angular velocity of the UAV, are the pitch acceleration, roll acceleration and yaw acceleration of the drone, are the linear velocity drag coefficients on the three coordinate axes, are the rotational resistance coefficients around the three coordinate axes, Indicates the farthest distance between the arm and the center of mass of the drone. is the moment of inertia of the rotor rotating around its axis, , g is the acceleration due to gravity, are the moments of inertia on the three coordinate axes respectively.

[0010] In an optional embodiment, step S3 includes the following steps: S3-1. Calculate the potential field function between drones and obstacle potential field function based on the artificial potential field method; S3-2. Determine the ideal input parameters based on the potential field function between machines and the potential field function of obstacles.

[0011] In an optional embodiment, the machine potential field function is:

[0012] in is the potential field between the i-th UAV and the j-th UAV, is the repulsive force between the i-th UAV and the j-th UAV, is the emergency obstacle avoidance repulsion between the i-th UAV and the j-th UAV, Expressed as and The distance between the following drones, is the influence range of the repulsive force between machines, The impact range of the emergency obstacle avoidance force; The obstacle potential field function is:

[0013] in is the obstacle potential field, Indicates The following drone's current position, Indicates the location of the obstacle. Expressed as The distance between the following drone and the obstacle, Indicates the influence range of the obstacle repulsion, Indicates the influence range of the emergency obstacle avoidance force, For the The obstacle repulsion between the drone and the obstacle, For the Emergency obstacle avoidance between a drone and obstacles.

[0014] In an optional implementation, the total potential field function of the UAV is obtained according to the potential field function between the UAVs and the potential field function of the obstacles:

[0015] The potential field force generated by the single UAV according to the artificial potential field and potential field force coefficient Multiply the displacement and deflection angle to get the initial expected position and attitude in the cluster control system Perform vector overlay to obtain ideal input parameters , the specific expression is as follows:

[0016]

[0017]

[0018] In an optional embodiment, the drone nonlinear system is:

[0019] Where A= , is a nonlinear function containing state quantities, k(A) is the control input coefficient, is the interference amount, , is the upper bound of the interference amount; The nonlinear interference terms are combined, and D(A)=f(A)+d(A) is approximated by RBF neural network:

[0020] Where A= represents the activation function of the hidden layer, is a center vector of nodes in the hidden layer with the same dimension as A, represents the width of the Gaussian basis function, which is a positive scalar. is the ideal network weight, is the network approximation error, The output of the neural network is the nonlinear interference term.

[0021] In an optional implementation, the first control law is:

[0022]

[0023] is the estimate of , s is the sliding surface, and c is a constant.

[0024] In an optional implementation, the second control law is:

[0025] Where i=1, 2, 3, 4, when i=1, T is a constant, A= , the second control law corresponding to the yaw angle; when i=2, T= ,A= The second control law corresponding to the roll angle in the y-axis direction when i=3, T= ,A= The second control law corresponding to the pitch angle in the x-axis direction when i=4, T= ,A= The second control law corresponding to the z-axis direction is used to adjust the position and attitude of the drone when avoiding obstacles.

[0026] In the second aspect, an embodiment of the present application also provides a drone cluster control server, which includes: a memory, a processor, and a drone cluster control program stored in the memory and executable on the processor. When the drone cluster control program is executed by the processor, the method proposed in the first aspect is implemented.

[0027] In a third aspect, an embodiment of the present application also provides a drone cluster control system, which includes several drones, a drone cluster control server, and a geodetic receiver. The geodetic receiver is used to receive satellite signals and generate differential positioning data, and send the differential positioning data to the drone cluster control server in real time. The drone cluster control server obtains a satellite positioning correction number based on the differential positioning data through the method proposed in the first aspect, and sends the satellite positioning correction number to the drone; the drone cluster control server receives waypoint tasks for controlling all drones separately, and sends the waypoint tasks to the corresponding drones.

[0028] It can be seen from the above technical solutions that the present invention has the following advantages: The application provides a method for planning the path of drones in a drone cluster through a neural network. By designing a path control plan based on an artificial potential field method and a flight sliding mode control plan based on a neural network, the path planning effect of a multi-drone cluster can be achieved, and the anti-interference ability and robustness of the system are improved. Through the designed inter-machine potential field function, when a single drone is avoiding obstacles, the remaining drones can monitor the movement trend of the obstacle-avoiding drone and avoid them synchronously. Through the set obstacle potential field function, the drone can be controlled to autonomously avoid obstacles, which can improve the stability of control. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] In order to more clearly illustrate the technical solution of the present invention, the accompanying drawings required for use in the description will be briefly introduced below. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0030] Figure 1 This is a schematic diagram of the forces acting on a quad-rotor drone in an embodiment of a specific implementation manner of the present invention.

[0031] Figure 2This is the x position tracking curve of the improved sliding mode controller in the specific implementation example of the present invention.

[0032] Figure 3 This is the y position tracking curve of the improved sliding mode controller in the specific implementation example of the present invention.

[0033] Figure 4 This is the z position tracking curve of the improved sliding mode controller in the specific implementation example of the present invention.

[0034] Figure 5 This is a yaw angle tracking curve of an improved sliding mode controller in a specific implementation example of the present invention.

[0035] Figure 6 This is a rolling angle tracking curve of an improved sliding mode controller in a specific implementation example of the present invention.

[0036] Figure 7 This is a pitch angle tracking curve of an improved sliding mode controller in a specific implementation example of the present invention.

[0037] Figure 8 These are time-varying signal response curves of various controllers when there is wind field interference in the specific implementation example of the present invention.

[0038] Fig. 9 It is the RBF neural network wind field interference estimation curve in the specific implementation example of the present invention.

[0039] Fig.10 This is the tracking performance when the actuator is damaged in the specific implementation example of the present invention (k=0%).

[0040] Fig.11 This is the tracking performance when the actuator is damaged in the specific implementation example of the present invention (k=30%).

[0041] Fig.12 This is a drone cluster obstacle avoidance path planning diagram in a specific implementation example of the present invention.

[0042] Fig.13 for Fig.12 A magnified image of the area near the obstacle.

[0043] Fig.14 for Fig.12 A magnified image of the area near obstacle B.

[0044] Fig.15 This is a control diagram for the path planning of a swarm of drones in a specific implementation example of the present invention. DETAILED DESCRIPTION

[0045] In the detailed description below, various embodiments of the present disclosure will be described more fully. The present disclosure may have various embodiments, and adjustments and changes may be made therein. However, it should be understood that there is no intention to limit the various embodiments of the present disclosure to the specific embodiments disclosed herein, but rather the present disclosure should be understood to cover all adjustments, equivalents and / or alternatives that fall within the spirit and scope of the various embodiments of the present disclosure.

[0046] In order to clearly describe the technical solutions of the embodiments of the present application, some terms and technologies involved in the embodiments of the present application are briefly introduced below: The artificial potential field method (APF) referred to in the embodiments of the present application is an algorithm for path planning, which was first proposed by Khatib in 1986. Its basic idea is to construct an obstacle repulsive potential field around the obstacle and a gravitational potential field around the target point. The controlled object (such as a robot, drone, etc.) is simultaneously subjected to repulsive and gravitational forces in the composite field composed of these two potential fields. The combined force direction of repulsive and gravitational forces (potential field gradient) guides the movement of the controlled object, thereby generating a collision-free path.

[0047] RBF neural network, or radial basis function neural network, is an efficient feedforward neural network. The following is a detailed explanation of RBF neural network: The basic principle of RBF neural network is to map the input vector to a high-dimensional space through a series of radial basis function transformations, and then map the result in the high-dimensional space to the output space through linear transformation. This transformation method enables RBF neural network to approximate any nonlinear function and has powerful nonlinear modeling capabilities.

[0048] For example, with the rapid development of drone technology, rotorcraft drones that can carry many types of sensors and gimbals to perform flight missions have played an indispensable role in various fields when faced with various tasks that humans cannot accomplish.

[0049] The current method used for drones to identify obstacles and avoid them to continue flying mainly combines GPS and visual sensors to estimate the system status of the drone, infer whether there are obstacles and plan paths. However, this method is difficult to implement in urban environments where high-rise buildings exist, and it is easy to cause errors in system state estimation when encountering dynamic obstacles. For this reason, the prior art proposes an autonomous obstacle avoidance method for a single drone based on an artificial potential field to overcome the problem of poor obstacle avoidance effect of drone status monitoring using GPS and visual sensors.

[0050] However, in actual drone applications, single drones have certain limitations. At this time, quadrotor drone clusters have attracted widespread attention. As a large and complex system, quadrotor drone clusters may encounter obstacles such as buildings and mountains during actual flight. The existence of these obstacles will threaten the flight safety of the quadrotor drone cluster. The quadrotor drone cluster needs to be able to make different decisions according to different environmental conditions to avoid threatening obstacles and other quadrotor drones at the same time. However, the stability and robustness of the obstacle avoidance method in the existing technology needs to be strengthened, and the analytical calculation is complex and requires global information.

[0051] In response to the above problems, this embodiment provides a method, server and system for controlling drones in a drone cluster. The method can further adjust the drone system control through an RBF neural network, solving the problem of insufficient stability and robustness in path planning and obstacle avoidance for a quad-rotor drone cluster in the prior art control method.

[0052] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. 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 creative work are within the scope of protection of the present invention.

[0053] Embodiment 1: A method for controlling drones in a drone cluster comprises the following steps: S1. Perform force analysis on the quadrotor drone and establish a virtual control input equation; S2, establish the dynamic model equation of the quadrotor drone in combination with the coordinate system rotation matrix; S3, design the ideal input parameters of the path based on the artificial potential field method and the UAV dynamics model; S4, optimizing the nonlinear terms of the single UAV system through neural network; S5, constructing the first control law after optimizing the nonlinear terms of the single UAV system; S6. Optimize the first control law by using the gradient descent method to obtain the second control law, and control the UAV through a virtual control input equation based on the second control law.

[0054] By designing path control planning based on the artificial potential field method and flight sliding mode control planning based on neural networks, the path planning effect of multiple UAV clusters can be achieved, improving the system's anti-interference ability and robustness.

[0055] The flight control of a quadrotor drone involves multiple variables, such as position, speed, attitude angle, etc. It is often difficult to accurately achieve the desired flight state by directly controlling physical quantities such as motor speed.

[0056] The role of virtual control input: As an intermediate link in the flight control system, the virtual control input equation can calculate the appropriate control instructions based on the error between the desired state and the actual state. After conversion, these instructions can more accurately control the flight state of the drone.

[0057] According to an embodiment of the present application, Figure 1 As shown, the virtual control input equation is:

[0058] in, is the lift generated by each rotor of the quadcopter, i=1, 2, 3, 4, to is the rotation speed of the four rotors, is the lift generated by the rotor and, Represents the input value of the roll angle channel, Represents the input value of the pitch angle channel, represents the input value of the yaw angle channel, and b represents the anti-torque coefficient of the propeller motor; It should be noted that the design and implementation of the neural network control algorithm needs to be based on the dynamic model. Through the dynamic model equation, the relationship between the control input and the motion state of the UAV can be derived, which provides a theoretical basis for the design of the neural network control algorithm, enabling the algorithm to calculate the appropriate control instructions based on the difference between the actual flight state and the target state.

[0059] According to an embodiment of the present application, the quad-rotor drone dynamics model equation is:

[0060] Among them, x, y, and z are the position variables in the directions of the three coordinate axes. , , is the linear velocity of the UAV on the three coordinate axes, , , is the linear acceleration of the UAV on the three coordinate axes, respectively The pitch, roll and yaw angles of the drone. are the pitch angular velocity, roll angular velocity and yaw angular velocity of the UAV, are the pitch acceleration, roll acceleration and yaw acceleration of the drone, are the linear velocity drag coefficients on the three coordinate axes, are the rotational resistance coefficients around the three coordinate axes, Indicates the farthest distance between the arm and the center of mass of the drone. is the moment of inertia of the rotor rotating around its axis, , g is the acceleration due to gravity, are the moments of inertia on the three coordinate axes respectively.

[0061] In one embodiment of the present invention, based on step S3, a possible embodiment is given below to illustrate its specific implementation scheme in a non-limiting manner, and step S3 includes the following steps: S3-1. Calculate the potential field function between drones and obstacle potential field function based on the artificial potential field method; S3-2. Determine the ideal input parameters based on the potential field function between machines and the potential field function of obstacles.

[0062] Design the artificial potential field repulsion function between UAVs, the expression is as follows:

[0063] in is the repulsive force scale factor, Indicates and The following drone's current position, Expressed as and The distance between the following drones, Indicates the influence range of the repulsive force between drones, generally refers to a circle with the following drone as the center and a radius of The circle of the machine can be differentiated by the above expression to obtain the expression of the repulsive force between the machines:

[0064] in .

[0065] From this, we can get the expression of the combined repulsive force between the machines:

[0066] In order to avoid unnecessary losses caused by collision, the emergency obstacle avoidance repulsion force is defined as follows:

[0067] in It can be seen that the emergency obstacle avoidance repulsion force is proportional to the magnitude of the repulsive force potential field between the drones. The shorter the distance between the drones, the greater the emergency obstacle avoidance repulsion force. The direction is in the opposite direction of the z-axis, which can achieve the effect of staggering up and down.

[0068] In this embodiment, the machine potential field function is:

[0069] in is the potential field between the i-th UAV and the j-th UAV, is the repulsive force between the i-th UAV and the j-th UAV, is the emergency obstacle avoidance repulsion between the i-th UAV and the j-th UAV, Expressed as and The distance between the following drones, is the influence range of the repulsive force between machines, is the influence range of the emergency obstacle avoidance force. Through the designed inter-drone potential field function, when a single drone is avoiding obstacles, the other drones can monitor the movement trend of the obstacle-avoiding drone and avoid it synchronously.

[0070] The expression of the repulsive potential field generated by obstacles in the flight area is:

[0071] in is the repulsive force scale factor, Indicates The following drone's current position, Indicates the location of the obstacle. Expressed as The distance between the following drone and the obstacle, It indicates the influence range of the obstacle repulsion, generally refers to the circle with the obstacle as the center and the radius as circle.

[0072] The repulsive force expression is obtained by taking the gradient of the repulsive potential field:

[0073] As the obstacle plays an absolutely dominant role when approaching the obstacle, the obstacle repulsion is directed along the drone to the rear side of the fuselage. The emergency obstacle avoidance force is perpendicular to the combined external repulsion. The specific direction needs to be determined according to the left and right direction of the current nose of the drone. Taking turning right as an example, the matrix of the repulsion force rotated 90° counterclockwise is defined as:

[0074] Then the relationship between the repulsive potential field and the emergency obstacle avoidance force is expressed as:

[0075] in is the speed of the obstacle relative to the UAV. Represents the influence range of the emergency obstacle avoidance force, and the obstacle artificial potential field function is obtained as follows:

[0076] in is the obstacle potential field, Indicates the location of the obstacle. Expressed as The distance between the following drone and the obstacle, Indicates the influence range of the emergency obstacle avoidance force, For the The obstacle repulsion between the drone and the obstacle, For the The emergency obstacle avoidance force between the drone and the obstacle. By setting the obstacle potential field function, the drone can be controlled to autonomously avoid obstacles.

[0077] Based on the above analysis, combined with the artificial potential field function between aircraft and the artificial potential field function of obstacles, the total potential field function of the UAV is obtained as follows:

[0078] The potential field force generated by the single UAV according to the artificial potential field and potential field force coefficient Multiply the displacement and deflection angle to get the initial expected position and attitude in the cluster control system Perform vector superposition to obtain ideal input parameters , the specific expression is as follows: ; ; .

[0079] In this embodiment, the sliding surface is designed: ; Design an adaptive reaching law: ; Where s is the sliding surface, is the adaptive reaching law, is a normal number, is the ideal input value, is the adaptive law adjustment parameter, is an adaptive gain parameter greater than zero.

[0080] The nonlinear system of the UAV is:

[0081] Where A= ， B= , is a nonlinear function containing state quantities, k(A) is the control input coefficient, is the interference amount, , is the upper bound of the interference amount; The nonlinear interference terms are combined, and D(A)=f(A)+d(A) is approximated by RBF neural network:

[0082] in represents the activation function of the hidden layer, is a center vector of nodes in the hidden layer with the same dimension as A, represents the width of the Gaussian basis function, which is a positive scalar. is the ideal network weight, is the network approximation error, The output of the neural network is the nonlinear interference term.

[0083] It should be noted that the learning speed of RBF neural network is relatively fast, which is mainly due to its fewer hidden layers and local approximation characteristics. During the training process, RBF neural network can find the optimal solution faster, thereby improving the real-time performance of the system. By approximating the disturbance term, RBF neural network can help quadcopters control their flight state more accurately. This helps to improve the flight accuracy and stability of the drone, enabling it to better cope with complex environments and tasks. When the drone is subject to unknown disturbances, the RBF neural network can predict and compensate based on its generalization ability. This helps to enhance the robustness of the system, enabling the drone to maintain a stable flight state in various environments.

[0084] It should be further explained that the adaptive reaching law shows that the general expression of the sliding mode control of the quadrotor UAV is:

[0085]

[0086] is the estimate of , s is the sliding surface, and c is a constant.

[0087] First, bring the above control law into the sliding surface derivative equation:

[0088] You can get:

[0089] in .

[0090] Select the Lyapunov function as:

[0091] Taking the derivative we get:

[0092] At this time, the adaptive law corresponding to the network weight is:

[0093] The derived formula is:

[0094] Since the network approximation error has an upper bound of: ; So we can take the coefficient , we can deduce:

[0095] And when hour, , according to the Russell invariance principle, the entire closed-loop system will tend to be stable, that is, hour The formula for the approximate nonlinear term of the RBF neural network is derived from the above, and the stability of the system is proved by combining the control law.

[0096] It should be further explained that the quadcopter is a typical nonlinear, highly coupled underactuated system with complex dynamic characteristics. The adaptive law can cope with the uncertainty of the system's internal parameters and the failure of the power mechanism by adjusting the controller parameters in real time, thereby improving the robustness of the entire system. When the system is disturbed or fails, the adaptive law can quickly adjust the control strategy to keep the drone in stable flight and avoid the occurrence of chattering.

[0097] It should be further explained that the variable parameter items of the control law based on the gradient descent optimization algorithm are designed as follows: First construct the performance error function:

[0098] Considering the gradient descent method, the adaptation equation is as follows:

[0099]

[0100] in, is the learning rate of the neural network convergence speed, and They are and The initial value of . According to the partial derivative chain rule, we can get:

[0101] The above two formulas have the following assumptions:

[0102] in is a Nabla operator, satisfying:

[0103] You can get and The expression is:

[0104]

[0105] Based on the above, the control law of the UAV yaw angle channel based on the gradient descent optimization algorithm is designed as follows. The other channels are similar. Approximation using RBF neural network:

[0106] Where i=1, 2, 3, 4, when i=1, T is a constant, A= , the second control law corresponding to the yaw angle; when i=2, T= ,A= The second control law corresponding to the roll angle in the y-axis direction when i=3, T= ,A= The second control law corresponding to the pitch angle in the x-axis direction when i=4, T= ,A= The second control law corresponding to the z-axis direction is used to adjust the position and attitude of the drone when avoiding obstacles.

[0107] It should be further explained that the gradient descent method continuously adjusts the parameters of the controller through iterative optimization to minimize the objective function (usually an error or loss function). In the control system of a quadcopter drone, this means that the ideal flight state can be more effectively approached, thereby improving the accuracy and stability of the control. Adaptive parameter adjustment can respond to various changes in the flight process in real time, ensuring that the drone can maintain a stable flight attitude in various environments. At the same time, the gradient descent method adaptive parameter adjustment has strong adaptive capabilities. It can detect changes in the flight process in real time and automatically adjust parameters according to these changes. This adaptive ability enables the drone to better adapt to various flight environments and mission requirements, improving the flexibility and practicality of the drone.

[0108] In order to illustrate the beneficial effects, the controller was built using the Matlab / Simulink simulation platform, and simulation and analysis were performed under the established motion model. Simulation experiments were conducted from two aspects, namely the tracking performance of a single UAV and the path planning effect of a cluster of UAVs.

[0109] Tracking performance of a single drone: A neural network-based controller is used to track the position and attitude of the following drone. Figure 2-7 is the corresponding tracking curve, where the tracking expectation signal during the simulation is set to:

[0110] It can be seen from the results that the designed controller can effectively track the target signal and has good tracking performance. In order to verify the anti-interference ability and robustness of the designed controller, the yaw angle channel of the UAV is taken as an example, and the wind field interference and actuator failure are present. Figure 8-11 is the corresponding tracking curve, the control experiment group is PID controller and backstepping controller, Figure 8 and Fig. 9 It can be seen that after adding wind field interference, other controllers have some degree of deviation and disturbance, while the controller based on the improved neural network has basically no deviation and can track the target signal well, which proves that the controller has a certain anti-interference ability. Fig.10 and Fig.11 It can be seen that when the actuator fails, both the PID and backstepping methods have a certain degree of overshoot, while the designed controller is still relatively stable and can track the signal relatively stably, proving that it has a certain degree of robustness.

[0111] Path planning effect of cluster drones: Fig.12For the simulation experiment of obstacle avoidance path planning of UAV, a cylindrical obstacle is set in the path planning of UAV. From the simulation results, it can be seen that the UAV can fly smoothly and orderly towards the target position in the cluster control system based on the leader-wingman method, and a certain safe distance can be maintained between UAVs. Fig.13 and Fig.14 This is a partial enlarged picture of the drone encountering an obstacle. It can be seen from the experimental picture that when the drone cluster encounters an obstacle, it can smoothly and smoothly bypass the obstacle and finally reach the destination. Fig.15 This is a simulation experiment for drone cluster path planning. During the flight, the overall multi-drone team realized cluster flight in a "heart shape" and a "pentagram shape" in turn, then made a 180-degree reverse turn, and finally achieved stable flight after the drone cluster climbed the slope. Finally, the cluster assembled in a "heart shape". It can be seen from the experimental results that the algorithm can achieve good cluster path planning control effects.

[0112] In some possible implementations, a drone cluster control server disclosed herein includes: a memory, a processor, and a drone cluster control program stored in the memory and executable on the processor. When the drone cluster control program is executed by the processor, the method proposed in the above embodiment is implemented.

[0113] In some possible implementations, the present invention discloses a drone cluster control system, which includes several drones, a drone cluster control server, and a geodetic receiver. The geodetic receiver is used to receive satellite signals and generate differential positioning data, and send the differential positioning data to the drone cluster control server in real time. The drone cluster control server obtains satellite positioning correction numbers according to the differential positioning data through the method proposed in the above embodiment, and sends the satellite positioning correction numbers to the drones; the drone cluster control server receives waypoint tasks for separately controlling all drones, and sends the waypoint tasks to the corresponding drones.

[0114] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for controlling drones in a drone cluster, characterized in that: The following steps are involved: S1. Perform force analysis on the quadrotor drone and establish a virtual control input equation; S2, establish the dynamic model equation of the quadrotor drone in combination with the coordinate system rotation matrix; S3, design the ideal input parameters of the path based on the artificial potential field method and the UAV dynamics model; S4, optimizing the nonlinear terms of the single UAV system through neural network; S5, constructing the first control law after optimizing the nonlinear terms of the single UAV system; S6. Optimize the first control law by using the gradient descent method to obtain the second control law, and control the UAV through a virtual control input equation based on the second control law.

2. The method for controlling drones in a drone cluster according to claim 1, characterized in that: The virtual control input equation is: in, is the lift generated by each rotor of the quadcopter, i=1, 2, 3, 4, to is the rotation speed of the four rotors, is the lift generated by the rotor and, Represents the input value of the roll angle channel, Represents the input value of the pitch angle channel, represents the input value of the yaw angle channel, and b represents the anti-torque coefficient of the propeller motor; The dynamic model equation of the quadrotor drone is: Among them, x, y, and z are the position variables in the directions of the three coordinate axes. , , is the linear velocity of the UAV on the three coordinate axes, , , is the linear acceleration of the UAV on the three coordinate axes, respectively The pitch, roll and yaw angles of the drone. are the pitch angular velocity, roll angular velocity and yaw angular velocity of the UAV, are the pitch acceleration, roll acceleration and yaw acceleration of the drone, are the linear velocity drag coefficients on the three coordinate axes, are the rotational resistance coefficients around the three coordinate axes, Indicates the farthest distance between the arm and the center of mass of the drone. is the moment of inertia of the rotor rotating around its axis, , g is the acceleration due to gravity, are the moments of inertia on the three coordinate axes respectively.

3. The method for controlling drones in a drone cluster according to claim 1, characterized in that: Step S3 includes the following steps: S3-1. Calculate the potential field function between drones and obstacle potential field function based on the artificial potential field method; S3-2. Determine the ideal input parameters based on the potential field function between machines and the potential field function of obstacles.

4. The method for controlling drones in a drone cluster according to claim 3, characterized in that: The potential field function between machines is: in is the potential field between the i-th UAV and the j-th UAV, is the repulsive force between the i-th UAV and the j-th UAV, is the emergency obstacle avoidance repulsion between the i-th UAV and the j-th UAV, Expressed as and The distance between the following drones, is the influence range of the repulsive force between machines, The impact range of the emergency obstacle avoidance force; The obstacle potential field function is: in is the obstacle potential field, Indicates The following drone's current position, Indicates the location of the obstacle. Expressed as The distance between the following drone and the obstacle, Indicates the influence range of the obstacle repulsion, Indicates the influence range of the emergency obstacle avoidance force, For the The obstacle repulsion between the drone and the obstacle, For the Emergency obstacle avoidance between a drone and obstacles.

5. The method for controlling drones in a drone cluster according to claim 4, characterized in that: The total potential field function of the UAV is obtained based on the potential field function between the aircraft and the obstacle potential field function: The potential field force generated by the single UAV according to the artificial potential field and potential field force coefficient Multiply the displacement and deflection angle to get the initial expected position and attitude in the cluster control system Perform vector overlay to obtain ideal input parameters , the specific expression is as follows: ； ； 。 6. The method for controlling drones in a drone cluster according to claim 5, characterized in that: The nonlinear system of the UAV is: Where A= ， B= , is a nonlinear function containing state quantities, k(A) is the control input coefficient, is the interference amount, , is the upper bound of the interference amount; Combine the nonlinear interference terms, let D(A)=f(A)+d(A), and approximate D(A) using RBF neural network: in Represents the activation function of the hidden layer is a center vector of nodes in the hidden layer with the same dimension as A, represents the width of the Gaussian basis function, which is a positive scalar. is the ideal network weight, is the network approximation error, The output of the neural network is the nonlinear interference term.

7. The method for controlling drones in a drone cluster according to claim 6, characterized in that: The first control law is: is the estimate of , s is the sliding surface, and c is a constant.

8. The method for controlling drones in a drone cluster according to claim 7, characterized in that: The second control law is: Where i=1, 2, 3, 4, when i=1, T is a constant, A= , the second control law corresponding to the yaw angle; when i=2, T= ,A= The second control law corresponding to the roll angle in the y-axis direction when i=3, T= ,A= The second control law corresponding to the pitch angle in the x-axis direction when i=4, T= ,A= The second control law corresponding to the z-axis direction is used to adjust the position and attitude of the drone when avoiding obstacles.

9. A drone cluster control server, characterized in that: The drone cluster control server includes: a memory, a processor, and a drone cluster control program stored in the memory and executable on the processor. When the drone cluster control program is executed by the processor, a method as claimed in any one of claims 1 to 8 is implemented.

10. A drone cluster control system, characterized in that: The drone cluster control system includes a plurality of drones, a drone cluster control server, and a geodetic receiver, wherein the geodetic receiver is used to receive satellite signals and generate differential positioning data, and send the differential positioning data to the drone cluster control server in real time, and the drone cluster control server obtains a satellite positioning correction number according to the differential positioning data by any one of the methods in claims 1 to 8, and sends the satellite positioning correction number to the drone; The drone cluster control server receives waypoint tasks for controlling all drones separately and sends the waypoint tasks to the corresponding drones.

Images