Plant protection unmanned aerial vehicle cooperative operation spraying area control method

Abstract

The application relates to the field of intelligent operation unmanned planes, and discloses a plant protection unmanned plane cooperative operation spraying area control method, which constructs a multi-unmanned plane cooperative hanging type spraying system; a Lagrange method is used to establish an underactuated dynamics model of a single plane hanging load system, and mutual constraints among the multi-unmanned planes are considered under the multi-unmanned plane framework; a barrier Lyapunov function (BLF) is designed, and full-range constraints are applied to single plane load displacement and relative distances among the multi-unmanned planes; RBF neural network is used to compensate system nonlinearity and external disturbance on line; an adaptive thrust control law of multi-plane coordination is designed, high-precision trajectory tracking of each unmanned plane, strict constraints of load displacement, and collision-free guarantee among the medicine boxes are realized; the application can effectively eliminate the collision risks among the unmanned planes and among the medicine boxes by pre-planning reasonable trajectories, and can avoid medicine waste caused by repeated spraying.

Description

Technical Field

[0001] This invention relates to the field of intelligent agricultural drones, specifically a method for controlling the spraying area of ​​agricultural drones in collaborative operations. Background Technology

[0002] The application of drone technology is a new and important engine for promoting the growth of the low-altitude economy. Drone flight control technology is one of the key and core technologies for the future development of the low-altitude economy. Drone technology has become a new field and track for enhancing innovation capabilities, which will have important application value for improving the efficiency ratio and industrial benefits of industrial production.

[0003] Sling-load drone systems, with their unique characteristics, are suitable for intelligent operation scenarios such as logistics delivery, agricultural and forestry plant protection, and emergency rescue. However, sling-load drone systems have high requirements for safety and stability during flight load operations, especially in multi-drone collaborative operation scenarios, where the challenges are even more severe: when multiple drones are spraying pesticides simultaneously, the pesticide tanks carried by each drone are prone to collisions, causing mission failure or even safety accidents; at the same time, the spraying range of each drone needs to be strictly constrained to avoid pesticide waste and environmental pollution caused by repeated spraying; in addition, the sling-load system of a single drone is itself an underactuated system, and its number of control inputs is usually less than the number of degrees of freedom of the system. Coupled with the strong coupling characteristics caused by load swing and the uncertain aerodynamic environment, precise control is extremely difficult. Traditional single-drone control methods mainly focus on suppressing swing angle or end-effector swing, but it is difficult to achieve hard constraints on the full displacement of the load and ensure collision avoidance between multiple drones. Against this backdrop, this invention proposes an adaptive constraint control method for a crop protection drone collaborative operation and hanging spraying system. By using a barrier Lyapunov function (BLF) to simultaneously constrain the load displacement of a single drone and the relative distance between multiple drones, combined with adaptive compensation from an RBF neural network, the method theoretically ensures safety and, in engineering practice, enables collaborative operation of multiple drones. Summary of the Invention

[0004] The purpose of this invention is to provide a method for controlling the spraying area of ​​agricultural drones in collaborative operation, so as to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for controlling the spraying area of ​​agricultural drones in collaborative operation, comprising the following steps:

[0006] Step S1: Construct a collaborative suspended pesticide spraying system of N quadcopter drones with N≥2. Each drone is connected to a pesticide tank load via a cable. The system includes a sensor system on each drone, an embedded real-time computing platform, and a wireless communication module for inter-drone information exchange.

[0007] Step S2: Establish underactuated dynamic models of the load-bearing system for each UAV using the Lagrange method, and establish relative positional constraints between UAVs within the multi-UAV framework to ensure that the distance between any two medicine boxes is not less than the preset minimum safety interval. ;

[0008] Step S3: Design a multi-layer constraint barrier Lyapunov function (BLF) and apply full-process hard constraints on the horizontal displacement of the single-machine load and the relative distance between multiple machines.

[0009] Step S4: Use RBF neural network as an adaptive parameter estimator to estimate and compensate for unknown nonlinear terms, multi-machine coupling effects and external disturbances in each UAV system online;

[0010] Step S5: Based on Lyapunov stability theory, design an adaptive update law for the weights and biases of the neural network that considers inter-machine coupling.

[0011] Step S6: Design a nonlinear adaptive thrust control law that integrates single-machine constraint terms, multi-machine collision avoidance constraints, and neural network compensation terms to achieve high-precision trajectory tracking of each UAV, strict load displacement constraints, and collision-free guarantee between UAVs.

[0012] Preferably, the hybrid drag model of the i-th UAV in step S2 is expressed as: (1)

[0013] in, The coefficient is a linear drag coefficient. This is the load resistance coefficient. The length of the cable. These are the projection coefficients of the load oscillation around the x-axis and y-axis, respectively. For the saturated bounded term coefficients, the relative position constraints between multiple machines are:

[0014] d ij (t) = q(x i − x j ) 2 + (y i -y j ) 2 + (z i - z j ) 2 ≥ d min (2)

[0015] Where, d ij Let d be the distance between the medicine boxes of the i-th and j-th drones. min This is the preset minimum safety interval.

[0016] Preferably, the barrier Lyapunov function in step S3 is:

[0017] (3)

[0018] in, To constrain the stringency parameter, To pre-set constraint boundaries, This represents the displacement error of the load in the x, y, and z directions.

[0019] Preferably, the output of the RBF neural network in step S4 for the i-th UAV is:

[0020] (4)

[0021] in, It is an input vector containing the local state error and the relative distance to neighboring units. The Gaussian radial basis functions are updated using the following adaptive law:

[0022] (5)

[0023] (6)

[0024] in, This represents the multi-machine coupling gain coefficient. It is a positive definite gain matrix.

[0025] Preferably, the cooperative thrust control law for the i-th UAV in step S6 is:

[0026] (7)

[0027] in, Let i be the single-unit load constraint term for the i-th UAV. Collision avoidance and repulsion terms generated for multi-machine relative distance constraints.

[0028] According to the above-mentioned agricultural drone collaborative spraying area control method, the sensor system in step S1 includes: an inertial measurement unit (IMU) for acquiring the drone's attitude and angular velocity; a GNSS / RTK positioning module for acquiring the drone's position information; a laser TOF sensor for real-time measurement of the distance between the load and the surrounding environment; a millimeter-wave radar for detecting adjacent drones and obstacles in front; an ultrasonic anemometer for measuring real-time wind speed; and a wireless communication module for enabling status information exchange between drones.

[0029] Preferably, the system comprises N quadcopter drones, where N≥2, each drone is equipped with an embedded real-time computing platform OrangePi and a flight control drive module, and the drones are connected to each other via wireless communication links to execute the multi-drone cooperative constraint control method described in claims 1-6.

[0030] Preferably, the embedded real-time computing platform on each UAV is a single-board computer running the ROS robot operating system, configured to perform distributed or centralized multi-UAV cooperative adaptive control laws, neural network parameter estimation, and interaction with state information from other UAVs.

[0031] A non-transitory computer-readable storage medium storing a computer program, which, when executed by a processor, implements the aforementioned multi-UAV cooperative constraint control method.

[0032] Compared with the prior art, the beneficial effects of this invention are as follows:

[0033] This invention achieves synchronous, full-process hard constraints on the load displacement of a single drone and the relative distance between drones by constructing a barrier Lyapunov function within a multi-drone collaborative framework. This overcomes the limitations of single-drone methods and traditional multi-drone control, ensuring that the displacement of each spray tank is always confined within a preset safety boundary throughout the entire collaborative operation, while the distance between any two spray tanks is never less than the safety interval. This fundamentally eliminates the collision risk and repeated spraying problems in multi-drone collaborative spraying. Through the adaptive mechanism of the RBF neural network, the system can compensate in real time for the dynamic coupling, aerodynamic parameter uncertainties, and wind field disturbances between drones, demonstrating strong robustness and scalability at both theoretical and engineering levels, providing a new solution for large-scale agricultural drone swarm operations. Attached Figure Description

[0034] Figure 1 This is a flowchart of the method of the present invention;

[0035] Figure 2 This is a schematic diagram of the three-dimensional model structure of the suspension system of the present invention;

[0036] Figure 3 This is a simplified XOZ plane model diagram of the present invention;

[0037] Figure 4 This is a physical image of the experimental platform of this invention;

[0038] Figure 5 This is the overall control block diagram of the transportation system based on unmanned aerial vehicles (UAVs) of the present invention;

[0039] Figure 6 This is a schematic diagram of a multi-drone coordinated spraying operation according to the present invention. Detailed Implementation

[0040] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0041] Example

[0042] Please see Figures 1-5 The diagram illustrates a method for controlling the spraying area using agricultural drones in a collaborative operation. Taking N=2 (two drones) as an example, the method is similar when extended to N≥3, and includes the following steps:

[0043] Step S1: Construct a collaborative suspended pesticide spraying system for two quadcopter drones. Each drone carries a pesticide tank load via a cable and is equipped with a sensor system, an embedded real-time computing platform, and an inter-drone information exchange module via a wireless communication link. The two drones are labeled Agent 1 and Agent 2, respectively.

[0044] Step S2: Establish underactuated dynamic models of the load-bearing systems of the two UAVs using the Lagrange method, including mixed drag terms, and establish relative position constraints between the UAVs: ,in =1.2m (considering the size of the medicine box and safety margin);

[0045] The aforementioned multi-UAV system uses N (N≥2) quadcopter UAVs as its core. Each UAV is connected to a medicine box payload via a fixed-length cable and communicates with the payload through a wireless communication link. It integrates multiple sensors and a real-time computing platform to achieve stable and safe multi-UAV collaborative payload operations, while ensuring accurate tracking of preset trajectories by each UAV and collision avoidance constraints between multiple medicine boxes. Key components of the system include:

[0046] Single-unit layer: Each UAV is equipped with an inertial measurement unit (IMU) for attitude feedback, a high-precision GNSS / RTK positioning module for providing trajectory error feedback, a laser TOF sensor for real-time measurement of the distance between the payload and the environment, a millimeter-wave radar for detecting surrounding obstacles and other aircraft, an ultrasonic anemometer for measuring real-time wind speed, and an embedded real-time computing platform (OrangePi) for running adaptive control laws.

[0047] Collaboration Layer: The system enables the exchange of status information between drones through wireless communication modules (such as WiFi or data transmission links). Each drone periodically exchanges key status quantities such as relative position, speed, and load displacement, which are used for distance constraints and collaborative control inputs between computers.

[0048] Specifically, in terms of system modeling, for the i-th UAV (i=1,2,...,N), the underactuated dynamic equations of its suspended load system are established using the Lagrange method:

[0049] Wherein, the state vector The position and load swing angle of the i-th UAV; This represents the mutual constraints between multiple machines (generated by BLF constraint terms); the meanings of other parameters are the same as before.

[0050] Furthermore, within the framework of multiple UAVs, relative position constraints between the UAVs are established. Let the positions of the medicine boxes corresponding to the i-th and j-th UAVs (i≠j) be respectively... Then the distance constraint between machines is:

[0051]

[0052] in, The constraint, which sets a minimum safe interval (including the size of the medicine box and the safety margin), ensures that no two medicine boxes will collide with each other, thereby avoiding the risk of collisions during collaborative operations.

[0053] To address the distortion problem of traditional linear drag models at low airspeeds, a hybrid drag model containing linear terms and saturated bounded terms is proposed (for the i-th UAV):

[0054]

[0055] The linear drag term, coupled drag term, and saturated bounded term have the same meanings as in the single-aircraft case. This hybrid model can cover the aerodynamic characteristics of the entire speed range, especially the low-speed region, providing a basis for accurate modeling of multiple aircraft in formation flight or intensive operations.

[0056] In terms of control design, the sway suppression problem is transformed into a boundary constraint problem of load displacement by reconstructing the load displacement constraints. The known horizontal displacement error of the load is:

[0057]

[0058] Traditional control objective: Suppressing swing angle The goal after displacement constraint reconstruction is to reconstruct the state vector. Decoupling is for horizontal displacement ( ) and vertical height ( Two independent subsystems, horizontal displacement constraint (X / Y): constraining the horizontal displacement of the load. Vertical height constraint (Z): Constrains the drone's altitude tracking error. Spatial displacement error of constraint load Strictly less than the safety margin, i.e. .

[0059]

[0060] Where x and y are the coordinates of the UAV in the inertial frame, and l is the length of the rope. Is the load along The swing angle of the axis (state vector) (part of) The desired location of the load; For safety margin.

[0061] To effectively suppress the load swing angle, an adaptive tracking method needs to be designed for the UAV translation control subsystem. Its Lyapunov function will be constructed based on the system state error, which is defined as follows:

[0062]

[0063] Where, x d y d z d γ xd γ yd Let x, y, z, γ represent the components of the desired state. x ,γ y For the actual state of the drone, specifically, x d y d z d It originates from a preset trajectory, while γ xd γ yd The expected value is zero.

[0064] Furthermore, within the framework of multiple UAVs, a barrier Lyapunov function (BLF) with integrated constraints is designed. This function comprises two constraints: (1) single-UAV load displacement constraints, ensuring that the displacement of the medicine boxes carried by each UAV is limited within its respective safe range; and (2) inter-UAV relative distance constraints, ensuring that the distance between any two medicine boxes is always not less than the preset minimum safe interval, thereby preventing collisions during collaborative operations. The expression for the multi-UAV constraint BLF is:

[0065]

[0066] The first term applies constraints to the load displacement of all N UAVs. Let be the displacement error of the i-th UAV's load in the k-th direction; the second term applies to all Constraints are imposed on the relative distance between drones, when the distance d between any two medicine boxes... ij →d min When the corresponding term approaches infinity, a strong repulsive force is generated to prevent collisions. This multi-level constraint design ensures the stability of a single machine and the safety of multiple machines.

[0067] when When the BLF term and its derivative are in the same state, they will generate an exponential repulsive force, forcing the corresponding error away from the boundary, thus achieving a hard constraint throughout the entire process.

[0068] In terms of control structure, to overcome the dynamic coupling, drag parameter uncertainty, and external disturbances in multi-UAV systems, an RBF neural network is used as a distributed or centralized adaptive parameter estimator. For the i-th UAV, the input of the RBF network is the state information of the UAV and its neighboring UAVs. (Including the local state error and relative position information with other bodies), the hidden layer performs nonlinear mapping through Gaussian radial basis functions, outputting estimates of the uncertainties of the local system and the influence of neighboring bodies:

[0069]

[0070] To ensure the approximation accuracy of the RBF network under multi-machine collaboration, an adaptive update law considering inter-machine coupling is designed:

[0071]

[0072]

[0073] Where, α ij This is the coupling gain coefficient, used to balance local control and multi-machine coordination.

[0074] Furthermore, the output of the RBF neural network (estimated values ​​of unknown nonlinear terms and multi-machine coupling terms) Embedding the thrust control law of the i-th UAV, design the following multi-UAV cooperative adaptive thrust control law:

[0075] Among them, (1) nominal feedback through the proportion-differential term (2) Ensure single-machine tracking error convergence; (3) Dynamic feedforward is compensated by gravity. With model acceleration feedforward Improve trajectory tracking performance; (3) Neural network compensation term Online estimation of nonlinearity and multi-machine coupling effects of the local system; (4) Single-machine constraint terms Based on the repulsive force generated by BLF, the displacement of the medicine box of the i-th UAV is constrained within a safe range; (5) Multi-aircraft collision avoidance terms The corresponding constraint force occurs when the distance between the two medicine boxes is approximately d. min (6) Coupling resistance compensation term generates repulsive effect; and It is used to compensate for the aerodynamic coupling effect caused by load swing and relative motion between multiple machines.

[0076] Through this control law that integrates multi-layer constraints and RBF adaptive compensation, the i-th UAV can not only complete its own trajectory tracking and load constraints, but also coordinate with other UAVs to maintain a safe distance, thereby achieving the goal of group coordinated spraying.

[0077] Regarding the stability proof of the system, for the nonlinear adaptive robust tracking control law of multi-UAV cooperation, the signal boundedness and asymptotic convergence of the closed-loop system are rigorously proved using Lyapunov stability theory and Barbalat's theorem. The specific conclusions are as follows: Under the multi-UAV cooperative framework, a global Lyapunov function is constructed:

[0078]

[0079] in, For the single-unit load constraint BLF term of the i-th UAV, it is used to constrain its load displacement within a safe range; The BLF term represents the relative distance constraint between the i-th and j-th UAVs, used to ensure that the distance between the UAVs is not less than the minimum safe interval.

[0080] The global Lyapunov function forms a unified constraint framework by simultaneously considering the individual load constraints and the relative distance constraints between all UAVs. This framework is applied when the load displacement of any UAV approaches its constraint boundary or the relative distance between any two UAVs approaches d. min When this occurs, the corresponding BLF term and its derivative will generate a strong repulsive force, ensuring that the constraint is rigidly satisfied.

[0081] By designing a multi-machine cooperative thrust control law, the time derivative of the global Lyapunov function is made to satisfy:

[0082]

[0083] Where, k i and E multi Both are greater than 0. According to Barbalat's lemma, we can obtain... (t→∞, i=1,2,...,N), thus the state error of all UAVs asymptotically converges to zero.

[0084] Furthermore, by utilizing the constraint properties of BLF, it can be rigorously proven that:

[0085] Single-machine load displacement constraints: For all t≥0 and i=1,2,...,N, ensure that the displacement of each medicine box is always within a safe range;

[0086] Multi-machine relative distance constraint: d ij (t)≥d min For all t≥0 and i≠j, ensure that no two medicine boxes collide.

[0087] asymptotic convergence: This means that all drones can track their own trajectories with high precision and load sway is completely suppressed.

[0088] Furthermore, an analysis of the boundary layer and initial conditions for the aforementioned constraints reveals that, although BLF can asymptotically guarantee the constraints... The constraint always holds true, but strict constraint satisfaction also requires the initial state to meet specific conditions. Specifically, to ensure that the constraint is valid throughout the entire process from t=0, the initial value should satisfy:

[0089]

[0090] Where i≠j, The boundary layer thickness parameter is used to avoid approaching the constraint boundary at the initial moment. In engineering practice, trajectory planning ensures that the UAV starts from an initial state that is sufficiently far away from the constraint boundary, which can satisfy the above conditions.

[0091] When δ k →0 + and δ d →0 + When the initial conditions tend to and At this point, the constraints have reached a critical state at the initial moment. To ensure the practical feasibility of the system, it is recommended to choose δ in the engineering design. k and δ d It should be no less than 3-5 times the accuracy of the sensor.

[0092] The selection criteria for the coupling gain coefficient, in multi-machine cooperative control, the coupling gain coefficient α ij The interaction strength between machines and the stability margin of the system are directly affected. Reasonable selection criteria are as follows:

[0093] Given the coupling gain matrix Its spectral radius ρ(A) should satisfy:

[0094]

[0095] Where λ min (Kp) is the scaling gain matrix. The smallest eigenvalue, L ϕLet be the Lipschitz constant of the Gaussian radial basis functions. This is the bound of the gradient constrained by the relative distance.

[0096] In practice, for a uniform formation of N drones, the following adaptive coupling gain is used:

[0097]

[0098] in Based on the coupling gain, d max =2d min The key characteristic of this adaptive design, which defines the effective influence range of adjacent units, is: when the distance d between units... ij Approaching the minimum safety interval d min When d approaches its maximum value α0, the coupling gain tends to maximize the impact of the collision avoidance constraint; conversely, when d approaches its maximum value α0, the coupling gain tends to maximize the impact of the collision avoidance constraint. ij Exceeding the effective range of influence d max When the coupling gain is completely attenuated to zero, the virtual coupling over long distances is completely eliminated, thereby reducing the communication and computing burden.

[0099] Quantifying the finite-time convergence speed of RBF neural networks: To ensure that the parameter learning speed of the neural network is sufficient to support the real-time satisfaction of constraints, a quantitative analysis of the convergence of RBF is required. Under the Lyapunov stability framework, the learning dynamics of the network weights are as follows:

[0100]

[0101] in, For the weight estimation error, when the learning gain When I is the identity matrix, the exponential convergence rate of the weight error is:

[0102]

[0103] in Let g be the Gram matrix of the Gaussian radial basis functions. To ensure that the constraints are effectively initiated in a finite time, the convergence time constant should satisfy:

[0104]

[0105] Where t is the expected time for constraint activation, which is determined by the formation speed, initial formation distance, and maximum interval range. In the design, the convergence time constant should not exceed 1 / 5 of the constraint activation time to ensure that the parameter learning of the neural network does not lag behind the triggering of the constraint.

[0106] Specific learning gain γ θThe number of radial basis function centers should be selected according to the following principles: the gain setting needs to balance stability and convergence speed. Too large a gain will lead to learning instability and parameter oscillation, while too small a gain will lead to slow convergence and difficulty in timely support of constraint implementation. The number of radial basis function centers also needs to balance accuracy and computational complexity. Too few centers will not be able to fully cover the state space and the approximation accuracy will be insufficient, while too many centers will result in large computational load and insufficient real-time performance. Therefore, the optimal choice of gain and number of centers should be determined by offline simulation testing on the embedded platform used to simultaneously meet the requirements of approximation accuracy and real-time performance.

[0107] An analysis of the impact of communication latency on distributed architecture reveals that, in a distributed collaborative framework, each device exchanges relative position information via wireless links, inevitably resulting in end-to-end latency τ. d This delay will directly affect the calculation accuracy of the relative distance constraint.

[0108] Suppose that when the i-th UAV obtains the position information of the j-th UAV at time t, this information actually comes from time t. The actual distance is Under the assumption of constant relative velocity, the estimation error bound is:

[0109]

[0110] To ensure a safety margin for the constraints, the estimation error caused by the delay should be accounted for. Design incorporating actual safety intervals:

[0111]

[0112] Where k is the safety factor and k is greater than or equal to 2, usually taken as 2 or 3 to cope with uncertainty, and the communication delay τ d The specific value depends on the wireless communication technology used (WiFi, data link, 5G, etc.) and the specific environment in which the system is deployed, and should be determined through actual measurement rather than estimation.

[0113] To improve latency robustness, a prediction-correction mechanism can be adopted: linear or higher-order predictions are made based on past velocity history to compensate for communication latency. The effectiveness of prediction compensation depends on the predictability of the target motion and the accuracy of the prediction model. It is recommended to verify its actual improvement effect through simulation and experiments.

[0114] The above evidence shows that the multi-machine cooperative control law of the present invention achieves the goals of group trajectory tracking and cooperative operation while ensuring safety constraints, forming a complete multi-machine stability framework of "safety constraints-signal boundedness-asymptotic convergence".

[0115]

[0116] Where, x d (t), y d (t), z d (t) represents the desired trajectory of the UAV, 0 indicates that the sway angle is completely suppressed; control objective: , The UAV's center of mass C tracks the desired trajectory: ξ d =[x d (t),y d (t),z d (t)] T .

[0117] By combining the barrier Lyapunov function (BLF) and neural network adaptive compensation, the above method achieves high-precision tracking and disturbance suppression while ensuring state constraints, forming a complete stability framework of "safety constraints → signal boundedness → asymptotic convergence", which provides a strict theoretical guarantee for the safety control of UAV sling systems.

[0118] The specific implementation of this invention is analyzed as follows, taking N=2 (two drones) as an example for discussion. The same applies when it is extended to the case of N≥3.

[0119] In terms of system architecture, this invention supports two collaborative control architectures:

[0120] Distributed architecture: Each UAV independently calculates its adaptive thrust control law based on its own sensor feedback and relative position information from interactions with neighboring UAVs. This architecture has the advantages of distributed computation and strong robustness to communication latency, and is suitable for large-scale multi-UAV formations with N≥3.

[0121] Centralized architecture: The global state information of multiple UAVs is gathered in the central control unit, which uniformly calculates the multi-layer constraint BLF function, RBF neural network compensation and thrust control law of each UAV, and then sends the control command to each UAV for execution. The advantage of this architecture is global optimality, but the computation and communication burden is heavy, and it is suitable for medium-sized formations with N≤5.

[0122] In terms of sensing and communication infrastructure, the system integrates the following main modules:

[0123] State awareness layer: Each UAV is equipped with an inertial measurement unit (IMU) for attitude feedback, a high-precision positioning module (GNSS / RTK) for position estimation, a distance measurement sensor for load and environment distance detection, multi-aircraft situational awareness equipment (such as millimeter-wave radar) for adjacent aircraft position detection, and a meteorological sensor for wind field measurement.

[0124] Collaborative Communication Layer: Real-time status information exchange between machines is achieved through wireless communication links. In a distributed architecture, each machine periodically broadcasts status variables such as position, speed, and load displacement estimates, and adjacent machines calculate relative position constraints accordingly. In a centralized architecture, global status information is aggregated to the central processing unit, which then calculates the constraint forces and issues control commands.

[0125] At the level of control algorithm implementation, the core operations of the system include:

[0126] First, multi-machine state estimation and information fusion are performed; based on the outputs of each sensor, the position p of the i-th UAV is estimated. (i) ,speed Load displacement error In critical states, through information exchange, the relative distance d to other organisms is calculated. ij and its time derivative .

[0127] Next, the Lyapunov function of the multi-layer constraint barrier is calculated, and the single-machine layer constraint term is:

[0128]

[0129] Its gradient gives the repulsive force that constrains the load displacement of the machine.

[0130] The multi-layer constraint terms are:

[0131]

[0132] The repulsive force of the relative distance constraint between its gradient generators.

[0133] The RBF neural network adaptive compensation calculation is performed again, with the network input including the local state error and the relative position information of neighboring organisms: Through nonlinear mapping of Gaussian radial basis functions, the nonlinearity and multi-machine coupling effect of the online learning system are addressed, and the weights and biases are updated according to the following adaptive law:

[0134]

[0135]

[0136] in, This represents the multi-machine coupling gain coefficient.

[0137] Finally, the thrust control law is synthesized and executed to drive the quadcopter motors:

[0138] This control law integrates standard feedback control, dynamic compensation, constraint force, and neural network compensation to ensure that the state error of the i-th UAV converges, the load displacement constraint is satisfied, and the relative distance constraint with other aircraft is satisfied. This is achieved by adjusting the coupling gain coefficient α online. ij The system can adaptively adjust according to real-time communication status and environmental changes, improving its robustness to communication delays and parameter uncertainties.

[0139] Through the coordinated operation of the above multi-level and multi-module systems, N UAVs achieve high-precision tracking of their pre-planned trajectories in a group-based collaborative spraying operation, while strictly ensuring load displacement constraints and multi-UAV collision avoidance constraints. Compared with single-UAV control methods, this invention breaks through the limitations of traditional passive suppression of swing angles and actively constrains the load displacement throughout the entire process. Compared with traditional multi-UAV collaborative methods, this invention innovatively combines multi-layer BLF and RBF adaptive compensation with a distributed / centralized hybrid architecture, forming a more complete and robust theoretical and engineering framework. The variables in the above formulas are explained as follows: q represents the system state vector. x, y, z represent the position coordinates of the UAV in the inertial frame, in meters; γ x γ y These represent the oscillation angles of the load around the X and Y axes, respectively, in rad; l represents the cable length, in m; d x d y d z These represent the linear drag coefficients, in units of N·s / m; d p The load resistance coefficient is expressed in N·s / m; k0 represents the saturation bounded term coefficient; C x S x They represent cosγ respectively x， sinγ x C y S y They represent cosγ respectively y， sinγ y ;d ij d represents the distance between the i-th and j-th UAVs, in meters; min This indicates the preset minimum safety interval, in meters (m). b represents the displacement of the load of the i-th UAV in the k-th direction, in meters; k This represents the preset constraint boundary, in meters (m). The parameter representing the constraint strictness of the i-th drone; This represents the input vector of the RBF neural network for the i-th drone; Represents the Gaussian radial basis functions; Represents the variance of the radial basis functions; Indicates the multi-machine coupling gain coefficient; Represent the proportional-differential gain matrix of the i-th UAV; M represents the thrust output of the i-th UAV, in N; q The value indicates the mass of the drone, in kg; m indicates the payload mass, in kg.

[0140] It should be noted that, in this invention, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus.

[0141] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for controlling a spraying area in a cooperative operation of an agricultural unmanned aerial vehicle, characterized in that, Includes the following steps: Step S1: Construct a collaborative suspended pesticide spraying system of N quadcopter drones with N≥2. Each drone is connected to a pesticide tank load via a cable. The system includes a sensor system on each drone, an embedded real-time computing platform, and a wireless communication module for inter-drone information exchange. Step S2, respectively, using the Lagrange method to establish each unmanned aircraft hanging load system under-actuated dynamic model, and in the multi-unmanned aircraft frame to establish the relative position constraints between the machines, to ensure that the distance between any two drug boxes is not less than the minimum safety interval ; Step S3: Design a multi-layer constraint barrier Lyapunov function (BLF) and apply full-process hard constraints on the horizontal displacement of the single-machine load and the relative distance between multiple machines. Step S4: Use RBF neural network as an adaptive parameter estimator to estimate and compensate for unknown nonlinear terms, multi-machine coupling effects and external disturbances in each UAV system online; Step S5: Based on Lyapunov stability theory, design an adaptive update law for the weights and biases of the neural network that considers inter-machine coupling. Step S6: Design a nonlinear adaptive thrust control law that integrates single-machine constraint terms, multi-machine collision avoidance constraints, and neural network compensation terms to achieve high-precision trajectory tracking of each UAV, strict load displacement constraints, and collision-free guarantee between UAVs.

2. The method for controlling the spraying area of ​​a plant protection drone in collaborative operation according to claim 1, characterized in that: The hybrid drag model for the i-th UAV in step S2 is expressed as follows: （1） in, The linear drag coefficient is... This is the load resistance coefficient. The length of the cable. These are the projection coefficients of the load oscillation around the x-axis and y-axis, respectively. For the saturated bounded term coefficients, the relative position constraints between multiple machines are: d ij (t) = q(x i − x j ) 2 + (y i − y j ) 2 + (z i − z j ) 2 ≥ d min (2) wherein d ij is the inter-drone distance between the i-th and j-th drones, d min is the preset minimum safety separation.

3. The method for controlling the spraying area of ​​a plant protection drone in collaborative operation according to claim 2, characterized in that: The barrier Lyapunov function in step S3 is: (3) in, To constrain the stringency parameter, To pre-set constraint boundaries, This represents the displacement error of the load in the x, y, and z directions.

4. The method for controlling the spraying area of ​​a crop protection drone in collaborative operation according to claim 3, characterized in that: The output of the RBF neural network for the i-th UAV in step S4 is: （4） in, It is an input vector containing the local state error and the relative distance to neighboring units. The Gaussian radial basis functions are updated using the following adaptive law: （5） （6） in, This represents the multi-machine coupling gain coefficient. It is a positive definite gain matrix.

5. The method for controlling the spraying area of ​​a plant protection drone in collaborative operation according to claim 4, characterized in that: The cooperative thrust control law for the i-th UAV in step S6 is as follows: （7） in, Let i be the single-unit load constraint term for the i-th UAV. Collision avoidance and repulsion terms generated for multi-machine relative distance constraints.

6. A method for controlling the spraying area of ​​a crop protection drone in collaborative operation according to any one of claims 1-5, characterized in that: The sensor system in step S1 includes: an inertial measurement unit (IMU) for acquiring the attitude and angular velocity of the UAV; a GNSS / RTK positioning module for acquiring the UAV's position information; a laser TOF sensor for measuring the distance between the payload and the surrounding environment in real time; a millimeter-wave radar for detecting adjacent UAVs and obstacles in front; an ultrasonic anemometer for measuring real-time wind speed; and a wireless communication module for enabling status information exchange between the UAVs.

7. A collaborative control system for agricultural drones, characterized in that: It includes N quadcopter drones, where N≥2, each drone is equipped with an embedded real-time computing platform OrangePi and a flight control drive module, and the drones are connected to each other through wireless communication links, for executing the multi-drone cooperative constraint control method described in claims 1-6.

8. The control system according to claim 7, characterized in that: The embedded real-time computing platform on each drone is a single-board computer running the ROS robot operating system, configured to execute distributed or centralized multi-drone cooperative adaptive control laws, neural network parameter estimation, and interact with the state information of other drones.

9. A non-transitory computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by the processor, it implements the multi-UAV cooperative constraint control method according to any one of claims 1-6.

Images