Multi-unmanned aerial vehicle control method and terminal based on hybrid control strategy

Abstract

The application discloses a kind of multi-unmanned aerial vehicle control method and terminal based on hybrid control strategy, and it is related to unmanned aerial vehicle intelligent control technical field.The method comprises: the dynamic model of unmanned aerial vehicle is established in advance, and task parameters are set according to task demand;Through model predictive control method, all unmanned aerial vehicles are cooperatively controlled based on task parameters, and the predicted control input of each unmanned aerial vehicle is obtained;Through iterative learning control method, all unmanned aerial vehicles are cooperatively controlled based on task parameters, and the instant control input of each unmanned aerial vehicle is obtained;According to the predicted control input and instant control input, the target control input of each unmanned aerial vehicle is synthesized.The application fully combines the real-time optimization capability of model predictive control method and the experience learning advantage of iterative learning control method.Can give consideration to high-precision trajectory tracking of single task and adaptive error correction of multiple tasks, solve the problem that single control method is limited in performance under complex environment.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of unmanned aerial vehicle intelligent control, and particularly relates to a multi-unmanned aerial vehicle control method and terminal based on a hybrid control strategy. BACKGROUND

[0002] With the development of unmanned aerial vehicle technology, multi-unmanned aerial vehicle systems have become the core technology support in many fields due to their efficient task execution capability and flexible deployment characteristics. In practical applications, multi-unmanned aerial vehicle cooperative control mainly includes the following methods:

[0003] Traditional control methods are realized through a proportion integration differentiation (PID) controller. The structure is simple, but it is difficult to handle multi-variable coupling and complex constraints, and the error is large in high-precision trajectory tracking scenarios.

[0004] Model predictive control (MPC) can effectively handle constraint conditions through rolling optimization, has strong real-time response capability, and can realize high-precision trajectory tracking for a single task. However, it strongly depends on the accuracy of the system dynamics model, and the control performance significantly decreases when there is an error in the model or external disturbance.

[0005] Iterative learning control (ILC) can realize adaptive error correction for multiple tasks by using historical error iteration, and can improve the tracking accuracy of repeated tasks. However, it has weak real-time response capability and is difficult to cope with sudden disturbances and non-periodic tasks.

[0006] It can be seen that a single multi-unmanned aerial vehicle cooperative control method has limited performance in a complex environment, and it is difficult to balance high-precision trajectory tracking for a single task and adaptive error correction for multiple tasks.

[0007] Therefore, the prior art still needs to be improved and developed. SUMMARY

[0008] The technical problem to be solved by the present application is to provide a multi-unmanned aerial vehicle control method and terminal based on a hybrid control strategy to solve the problem of a single multi-unmanned aerial vehicle cooperative control method in the prior art having limited performance in a complex environment and being difficult to balance high-precision trajectory tracking for a single task and adaptive error correction for multiple tasks.

[0009] The technical solution adopted by the present application to solve the problem is as follows:

[0010] In a first aspect, the present application provides a multi-unmanned aerial vehicle control method based on a hybrid control strategy, which comprises:

[0011] A dynamic model of the UAVs is pre-established, and a task parameter is set according to a task requirement; the task parameter includes an expected trajectory;

[0012] The model predictive control method is used to cooperatively control all the UAVs based on the task parameter, to obtain a predicted control input of each UAV; the model predictive control method includes: predicting a future state based on a current observation state of a system and the dynamic model, and obtaining an optimal control sequence in the future by solving an optimization problem, and taking a first control input of the optimal control sequence as the predicted control input; wherein the optimization problem is constructed based on the task parameter; the current observation state of the system is used to reflect an observation result of a system state;

[0013] The iterative learning control method is used to cooperatively control all the UAVs based on the task parameter, to obtain an instant control input of each UAV; the iterative learning control method includes: correcting an expected trajectory in a next round based on a full-time-domain historical error, and generating the instant control input according to an error between a current state of the system and the expected trajectory; wherein the current state of the system is used to reflect a set of internal variables of the system at a current time;

[0014] The predicted control input and the instant control input are combined to synthesize a target control input of each UAV.

[0015] In an embodiment, the type of the dynamic model includes: a linear state space model and a full dynamic model.

[0016] In an embodiment, predicting a future state based on a current observation state of a system and the dynamic model, and obtaining an optimal control sequence in the future by solving an optimization problem, includes:

[0017] In each control period, a state sequence in a future time step is recursively predicted based on the dynamic model, starting from the current observation state of the system;

[0018] In each optimization period, a target function of a weighted polynomial is constructed according to an expected trajectory and a control target, and an optimization problem is established according to the target function;

[0019] The state sequence and the optimization problem are solved to obtain the optimal control sequence.

[0020] In an embodiment, the constraints corresponding to the optimization problem include: a dynamic constraint, an input physical boundary, an expandable state constraint, and an output constraint.

[0021] In an embodiment, in the model predictive control method, each unmanned aerial vehicle adopts objectification and arrayed batch management, and each unmanned aerial vehicle independently runs the model predictive control method.

[0022] In an embodiment, the expected trajectory of the next round is corrected based on the full-time domain historical error, and the instant control input is generated according to the error between the current state of the system and the expected trajectory, including:

[0023] By the trajectory correction learning unit of the upper layer, after the end of each round of task execution, the expected trajectory of the next round is corrected based on the full-time domain historical error, and the corrected trajectory is taken as the expected trajectory of the next round; the correction range includes each trajectory point and / or each spatial dimension.

[0024] By the real-time tracking controller of the lower layer, at each time, the instant control input is generated according to the error between the current state of the system and the expected trajectory.

[0025] In an embodiment, the target control input of each unmanned aerial vehicle is synthesized according to the prediction control input and the instant control input, including:

[0026] The prediction control input and the instant control input are weighted and fused to synthesize the target control input of each unmanned aerial vehicle.

[0027] In an embodiment, the corresponding weight distribution mode of the weighted fusion includes a static weight distribution mode and / or a dynamic weight adaptive distribution mode.

[0028] In an embodiment, the dynamic weight adaptive distribution mode includes:

[0029] In the initial stage, the weight of the prediction control input is greater than the weight of the instant control input.

[0030] In the error convergence stage, the weight of the instant control input is gradually increased.

[0031] In the disturbance or mutation stage, the weight of the prediction control input is increased.

[0032] In a second aspect, an embodiment of the present application further provides a terminal, including a memory and one or more processors; the memory stores one or more programs; the program contains instructions for executing the multi-unmanned aerial vehicle control method based on the hybrid control strategy as described in any of the above; the processor is used to execute the program.

[0033] The beneficial effects of the present application: the embodiment of the present application establishes the dynamic model of the unmanned aerial vehicle in advance, and sets the task parameters according to the task requirements; through the model predictive control method, all unmanned aerial vehicles are cooperatively controlled based on the task parameters, and the predicted control input of each unmanned aerial vehicle is obtained; through the iterative learning control method, all unmanned aerial vehicles are cooperatively controlled based on the task parameters, and the instant control input of each unmanned aerial vehicle is obtained; the target control input of each unmanned aerial vehicle is synthesized according to the predicted control input and the instant control input. The present application fully combines the real-time optimization capability of the model predictive control method and the experience learning advantage of the iterative learning control method. The high-precision trajectory tracking of single task and the adaptive error correction of multiple tasks can be considered, and the performance limitation problem of a single control method in a complex environment is solved. BRIEF DESCRIPTION OF DRAWINGS

[0034] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or prior art description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments described in the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0035] Figure 1 is the basic flow diagram of the multi-unmanned aerial vehicle control method based on the hybrid control strategy provided by the embodiment of the present application.

[0036] Figure 2 is the rolling optimization and time domain sliding diagram of the model predictive control method provided by the embodiment of the present application.

[0037] Figure 3 is the double-layer structure diagram of the controller in the iterative learning control method provided by the embodiment of the present application.

[0038] Figure 4 is the complete flow diagram of the multi-unmanned aerial vehicle control method based on the hybrid control strategy provided by the embodiment of the present application.

[0039] Figure 5 is the average tracking error performance comparison diagram of different control methods provided by the embodiment of the present application.

[0040] Figure 6 is the principle block diagram of the terminal provided by the embodiment of the present application. DETAILED DESCRIPTION

[0041] The application discloses a multi-unmanned aerial vehicle control method and a terminal based on a hybrid control strategy, and the application is further described in detail below with reference to the drawings and examples. It should be understood that the specific examples described herein are only used to explain the application and do not limit the application.

[0042] Those skilled in the art can understand that, unless specifically stated otherwise, the singular forms "a", "an", and "the" as used herein can also include plural forms. It should be further understood that the phrase "comprising" used in the specification of the application means that the features, integers, steps, operations, elements, and / or components exist, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof. It should be understood that when we say that an element is "connected" or "coupled" to another element, it can be directly connected or coupled to the other element, or there can be intermediate elements. In addition, "connected" or "coupled" as used herein can include wireless connection or wireless coupling. The phrase "and / or" as used herein includes all or any single unit and all combinations of the associated listed items.

[0043] Those skilled in the art can understand that, unless otherwise defined, all terms (including technical terms and scientific terms) used herein have the same meaning as that generally understood by those skilled in the art to which the application belongs. It should also be understood that terms such as those defined in a general dictionary should be understood as having meanings consistent with those in the context of the prior art, and should not be interpreted in an idealized or overly formal sense unless specifically defined as such.

[0044] In view of the problem that a single multi-unmanned aerial vehicle cooperative control method has limited performance in a complex environment and is difficult to balance high-precision trajectory tracking in a single task and adaptive error correction in multiple tasks, the application fuses the real-time optimization capability of a model predictive control method (MPC) and the experience learning advantage of an iterative learning control method (ILC) to propose a hybrid control strategy. The hybrid control strategy can realize high-precision and strong-robust cooperative trajectory planning of multi-unmanned aerial vehicles in a complex environment, and makes up for the limitations of a single control method.

[0045] As shown in Figure 1 The multi-unmanned aerial vehicle control method based on the hybrid control strategy provided by the application specifically includes the following steps:

[0046] In step S100, a dynamic model of an unmanned aerial vehicle is established in advance, and task parameters are set according to task requirements; the task parameters include an expected trajectory.

[0047] Specifically, taking a multi-UAV formation cooperative trajectory tracking task as an example, the task requirement (also known as training content) is for multiple UAVs to cooperate in a line formation to track a preset three-dimensional desired trajectory while maintaining a fixed distance and formation structure. First, a dynamic model of the UAVs is established, where the UAVs can be quadcopters. Then, based on the task requirements, task parameters and states are set, i.e., the task parameters and states are initialized, including but not limited to: the desired trajectory (also known as the reference trajectory), formation parameters, control inputs, and parameters.

[0048] Furthermore, the types of dynamic models include: linear state-space models and full dynamic models. The two dynamic models can be flexibly switched according to a preset model switching mechanism. For example, the model switching mechanism can be based on scene type; a linear state-space model is used for low-speed, small-scale scenes, while a full dynamic model is used for high-speed, large-scale scenes.

[0049] For linear state-space models, it is assumed that the UAV moves within a small range with minimal changes in attitude angles, and higher-order nonlinear factors such as air resistance and external disturbances are ignored. Under this assumption, taking a quadcopter UAV as an example, the UAV's three-dimensional spatial motion can be approximated as the motion of a point mass in inertial space. Therefore, the system state mainly includes position and velocity, and the control input mainly includes three-axis acceleration.

[0050] Assuming the ground coordinate system is the global reference system, the system state and control inputs are defined as follows:

[0051] System status : ;

[0052] in, This refers to the position of the UAV's center (also known as the UAV's centroid) in the ground coordinate system. This refers to the velocity components of the UAV in each axis direction (also known as the linear velocity of the UAV in each axis direction in the ground coordinate system).

[0053] Control input : ;

[0054] in,( , , ) represents the acceleration components of the UAV in each axis direction.

[0055] For a simplified linear model of a quadcopter UAV, the system dynamics can be described as follows:

[0056] ;

[0057] in, These are the velocity components of the UAV in each axis direction. Let represent the acceleration components of the UAV in each axis direction. This is the acceleration due to gravity.

[0058] Discrete-time modeling is used, with a sampling period of . Then the state update equation is:

[0059] ;

[0060] in, for The system state at any given moment; For the system in Time-based control input.

[0061] Furthermore, the system matrix With control matrix for:

[0062] ;

[0063] ;

[0064] Among them, the system matrix This reflects the autoregressive relationship between states. The first three rows indicate that the current position is recursively derived from the position and velocity of the previous step, while the last three rows represent the inertia of velocity retention. (Control matrix) This reflects the effect of the control input; controlling acceleration not only has an immediate impact on velocity, but also... The item has a cumulative effect on the position.

[0065] Furthermore, the output matrix Commonly used to extract position output:

[0066] ;

[0067] Final output ,in, This refers to the position of the drone in three-dimensional space.

[0068] For the full dynamics model, the system state and control input are defined as follows:

[0069] System status :

[0070] ;

[0071] in, Euler angles for the drone's attitude. For roll angle (Roll), Pitch, Yaw. is the angular velocity of the UAV in its own coordinate system, is the angular velocity of rotation around axis, is the angular velocity of rotation around axis, is the angular velocity of rotation around axis; is the transpose symbol, which converts a row vector into a column vector.

[0072] Control input :

[0073] ;

[0074] where, is the total thrust (the resultant force points to the body axis positive direction); is the moment control input around axis, is the moment control input around axis, is the moment control input around axis; is the transpose symbol, which converts a row vector into a column vector.

[0075] As a rigid body system, the dynamics behavior of quadrotor UAV can be described by Newton-Euler equations, including the equations of center of mass motion and attitude motion. In the ground inertial coordinate system , the center of mass motion of the UAV is affected by gravity and the thrust generated by the rotors, and the equation of center of mass motion is as follows:

[0076] ;

[0077] where, is the mass of the UAV; is the second derivative; is the position of the center of mass of the UAV in the ground coordinate system; is the acceleration of gravity; is the total thrust; is the rotation matrix from the body coordinate system to the ground coordinate system, and the rotation matrix is defined by Euler angles as follows:

[0078] ;

[0079] where, , , , , , .

[0080] The attitude motion equations reflect the relationship between the change of Euler angles and the angular velocity of the aircraft. The attitude motion equations are as follows:

[0081] ;

[0082] in, Let Euler angular velocity be the attitude of the drone. For roll angular velocity, For pitch angular velocity, Yaw angular velocity; To bypass Angular velocity of the axis of rotation, To bypass Angular velocity of the axis of rotation, To bypass Angular velocity of the axis of rotation. The transformation matrix from Euler angles to angular velocity is specifically represented as follows:

[0083] .

[0084] Furthermore, the Euler equations are used to describe the rotation of the UAV about its own axes, and the moment of inertia under each coordinate axis is defined. for:

[0085] ;

[0086] in, This refers to the angular acceleration of the UAV in its own coordinate system. To bypass Angular acceleration of the axis of rotation, To bypass Angular acceleration of the axis of rotation, To bypass Angular acceleration of the axis of rotation.

[0087] In practical engineering, to simplify analysis, the cross-inertia term is often ignored or approximated by the damping coefficient. The rotational speed of each rotor in a quadcopter determines the magnitude of the total thrust and torque. The following relationship is typically observed:

[0088] Total thrust : ;

[0089] in, This is the thrust coefficient; For the first quadcopter drone The rotational speed of each rotor.

[0090] Torque: ;

[0091] where, is the rotor to center of mass distance, is the counter torque coefficient.

[0092] To facilitate the implementation of flight simulation on digital computers, the full dynamics model needs to be discretized. Taking the sampling period as an example, the Euler method is used for approximation:

[0093] ;

[0094] ;

[0095] ;

[0096] ;

[0097] where, is the system state at time ; is the velocity component of the UAV in the axis direction at time ; is the acceleration component of the UAV in the axis direction at time ; is the roll angle of the UAV attitude at time ; is the roll angular velocity of the UAV attitude at time ; is the angular velocity of the UAV rotating around the axis in its own coordinate system at time ; is the angular acceleration of the UAV rotating around the axis in its own coordinate system at time .

[0098] Further, in actual simulation, damping factors and noise terms can also be introduced to better simulate real factors such as air resistance, environmental disturbance and sensor noise, so that the simulation result is closer to the flight characteristics of the real UAV.

[0099] Step S200, by the model predictive control method, the task parameters are used to cooperatively control all UAVs, and the predicted control input of each UAV is obtained; the model predictive control method includes: predicting the future state based on the current observed state of the system and the dynamics model, and obtaining the optimal control sequence in the future by solving an optimization problem, and taking the first control input of the optimal control sequence as the predicted control input; wherein the optimization problem is constructed based on the task parameters; the current observed state of the system is used to reflect the observation result of the system state.

[0100] This embodiment employs a hybrid control strategy for multi-UAV cooperative control, including a model predictive control (MMC) method. MMC is a feedback control method that utilizes a mathematical model of the controlled object to perform rolling optimization over a finite future time domain. Specifically, this embodiment uses the desired trajectory from the task parameters to perform cooperative control of all UAVs, and defines the control input obtained based on the MMC method as the predictive control input. The process of using the MMC method is as follows: at each sampling time, the MMC method predicts the system behavior over a future period based on the current observed state of the system and the known dynamic model. By solving the optimization problem, a series of optimal control sequences for the future are obtained. In actual execution, only the first optimized control input is applied to the system, and then the process repeats as it rolls into the next sampling period.

[0101] In one implementation, the future state is predicted based on the current observed state of the system and the dynamic model, and the optimal control sequence for the future is obtained by solving an optimization problem, including:

[0102] In each control cycle, starting from the current observed state of the system, the state sequence within a few future time steps is recursively predicted based on the dynamic model.

[0103] In each optimization cycle, a weighted polynomial objective function is constructed based on the desired trajectory and control objective, and an optimization problem is established based on the objective function.

[0104] The optimal control sequence is obtained by solving the state sequence and the optimization problem.

[0105] Specifically, the model predictive control method mainly includes the following three execution processes:

[0106] The process of predicting future states: In each control cycle, starting from the current observed state of the system, the future is recursively predicted based on the dynamic model of the UAV. The state sequence within a step, for example, setting =10, sampling period =0.1s. The prediction process strictly follows the laws of discrete dynamics.

[0107] The optimization decision-making process of the objective function: In each optimization cycle, the system constructs the following general prototype of a weighted, polynomial objective function around the desired trajectory and control objective:

[0108] ;

[0109] in, The objective function is... is a step count; is a three-dimensional spatial position at a time instant, is a reference three-dimensional spatial position at a time instant; is a three-dimensional velocity component at a time instant, is a reference three-dimensional velocity component at a time instant; is a control input at a time instant; is a weight coefficient for position, is a weight coefficient for position and velocity error; is a weight coefficient for control input energy consumption, is a weight coefficient for input variation. In the embodiment, all weight parameters are diagonal matrices.

[0110] Rolling horizon optimization and closed-loop feedback process: only the first control input of the control input sequence obtained by optimization is applied to the system in each cycle, and the system state is updated, entering the next cycle, repeating the prediction, optimization and application process, forming rolling horizon optimization and dynamic closed-loop feedback.

[0111] For example, the embodiment adopts a system composed of five four-rotor unmanned aerial vehicles, and simulation verification is respectively performed based on a linear state space model and a full dynamics model. The rolling optimization and time domain sliding principle of the model predictive control method is as shown in Figure 2 .

[0112] In an implementation manner, the constraint corresponding to the optimization problem includes: a dynamics constraint, an input physical boundary, an expandable state constraint and an output constraint.

[0113] Specifically, the embodiment can also realize integrated processing of the constraints, that is, strictly integrating the dynamics constraint, the input physical boundary, the expandable state constraint and the output constraint in the optimization problem.

[0114] For example, for the linear state space model, the unmanned aerial vehicle acceleration constraint is . is a minimum allowed value of acceleration, is a maximum allowed value of acceleration; for the full dynamics model, the unmanned aerial vehicle thrust constraint is . is a minimum allowed value of thrust, is a maximum allowed value of thrust. All future states of prediction need to gradually satisfy the recursive relationship of the dynamics model, support expansion of obstacles, space, formation and other constraints, and can be directly embedded in the form of equations or inequalities.

[0115] In an implementation manner, in the model predictive control method, each unmanned aerial vehicle adopts objectification and arrayed batch management, and each unmanned aerial vehicle independently runs the model predictive control method.

[0116] Specifically, the embodiment can also implement batch scheduling and synchronization of a multivariable system, that is, each unmanned aerial vehicle adopts objectification and arrayed batch management, and each unmanned aerial vehicle independently runs the model predictive control method for optimization. The information such as the state, control input, and error of the unmanned aerial vehicle is uniformly recorded and synchronized, and the formation information sharing and cooperative mechanism is supported.

[0117] Further, in order to realize efficient optimization and real-time guarantee, the optimization problem modeling and solving of the embodiment adopt cvxpy (a modeling tool), and the actual solver is OSQP (Operator Splitting Quadratic Program, a quadratic programming solver). Each cycle automatically models and directly calls OSQP to complete quadratic programming solving, so as to meet the real-time requirement of high-frequency closed loop.

[0118] Further, in order to realize abnormality and safety processing, when one-step optimization does not converge or OSQP is abnormal, for a linear state space model, the system immediately outputs zero acceleration; or for a full dynamics model, the system immediately inputs safe hovering. In addition, the control input can be forced to be located in a physical safety range, so as to guarantee operation safety.

[0119] In step S300, the iterative learning control method is used to cooperatively control all unmanned aerial vehicles based on the task parameters, to obtain an instant control input of each unmanned aerial vehicle. The iterative learning control method includes: correcting a next round of expected trajectory based on full-time domain historical error, and generating an instant control input according to an error between a current state of the system and the expected trajectory; wherein the current state of the system is used to reflect an internal variable set of the system at the current time.

[0120] Specifically, the embodiment uses a hybrid control strategy to cooperatively control multiple unmanned aerial vehicles, which also includes the iterative learning control method. The controller of the iterative learning control method adopts a double-layer structure to realize the advantage complementation between task-level experience learning and control-level real-time feedback. The task-level experience learning refers to correcting a next round of expected trajectory based on full-time domain historical error, and the control-level real-time feedback refers to generating a control input according to an error between a current state of the system and the expected trajectory. The embodiment defines the control input obtained based on the iterative learning control method as an instant control input.

[0121] In an implementation manner, correcting a next round of expected trajectory based on full-time domain historical error, and generating an instant control input according to an error between a current state of the system and the expected trajectory, includes:

[0122] Through the upper-level trajectory correction learning unit, after each round of task execution, the expected trajectory for the next round is corrected based on the full-time domain historical error, and the corrected trajectory is used as the expected trajectory for the next round; the correction scope includes: each trajectory point and / or each spatial dimension;

[0123] The underlying real-time tracking controller generates the instantaneous control input at each moment based on the error between the current state of the system and the desired trajectory.

[0124] Specifically, in the two-layer structure, the upper layer is the trajectory correction learning unit, and the lower layer is the real-time tracking controller, which can be a proportional-integral-differential (PID) controller.

[0125] For the upper-level trajectory correction learning unit: after each round of task execution, the expected trajectory for the next round is corrected point by point based on the historical error in the entire time domain.

[0126] Assume the first The expected trajectory of the wheel is The system status is The corresponding historical error for:

[0127] ;

[0128] The formula for correcting the expected trajectory is:

[0129] ;

[0130] in, The learning rate is dynamic and can be adaptively adjusted according to the number of iterations and the error distribution.

[0131] The above corrections can be performed independently at each trajectory point and each spatial dimension, supporting segmented weighting and local adaptive operations, such as assigning greater weight to the start-up and acceleration phases.

[0132] Corrected Expected Trajectory This serves as the expected trajectory for the next round, which can then be invoked by the lower-level control.

[0133] For the lower-level real-time tracking controller: The lower-level controller is responsible for, at each moment, adjusting the system's current state. Compared with the corrected expected trajectory Error between ,generate Time control input The control input generated based on the iterative learning control method is an instantaneous control input.

[0134] Taking the three-axis acceleration control of a linear state-space model as an example, the control law of the real-time tracking controller is:

[0135] ;

[0136] in, As a proportion; It is a differential; For integral gain; For integration variables; For integration variables The differential.

[0137] The principle of iterative learning control methods is as follows: Figure 3 As shown, the upper-layer trajectory correction learning unit continuously corrects the desired trajectory through historical error accumulation, achieving long-term performance improvement under model uncertainty and system lag. The lower-layer real-time tracking controller and other feedback controls enable immediate suppression of sudden disturbances and short-term anomalies. This embodiment's hierarchical structure supports multi-UAV collaboration, with each UAV independently learning and correcting its own trajectory while maintaining formation consistency through information exchange. A system consisting of five quadcopter UAVs can be used, with simulation verification based on both a linear state-space model and a full dynamics model.

[0138] Step S400: Synthesize the target control input for each UAV based on the predicted control input and the real-time control input.

[0139] Specifically, a hybrid control strategy is generated using both model predictive control and iterative learning control methods. The system in the first... Task rotation Target control input at any time The final control input is synthesized from both predictive and real-time control inputs. The predictive control input is output by a model predictive control method, while the real-time control input is output by an iterative learning control method. In practical applications, the method of this embodiment is used to coordinate the control of multiple UAV systems. The obtained control input sequence is sent to each UAV, and the actual trajectory, expected trajectory, optimized trajectory, and error data of the equipment during operation are collected for data analysis.

[0140] In one implementation, synthesizing the target control input for each UAV based on the predicted control input and the real-time control input includes:

[0141] The predicted control input and the real-time control input are weighted and fused to synthesize the target control input for each UAV.

[0142] Specifically, the predictive control input and the real-time control input are weighted and synthesized:

[0143] ;

[0144] in, For the first Task rotation The predictive control input for the output at any given time is obtained based on the current state and short-term time domain prediction; For the first Task rotation The real-time control input output is obtained based on the accumulation of historical errors. for The weighting coefficients, for The weighting coefficients satisfy + =1.

[0145] In one implementation, the weight allocation method corresponding to weighted fusion includes: static weight allocation method and / or dynamic weight adaptive allocation method.

[0146] Specifically, the weight allocation method in this embodiment may include one or more of a static weight allocation method and a dynamic adaptive weight allocation method. Static allocation uses preset fixed weights, for example... =0.7, =0.3, suitable for scenarios with stable environments, tasks, and dynamic characteristics, ensuring that model predictive control dominates global dynamic control, while iterative learning control is used for long-term fine-tuning. Dynamic weight adaptive allocation means that in practical applications, weights can be adjusted in real time based on current error, error convergence speed, external disturbances, etc., to maximize system performance.

[0147] Furthermore, the dynamic weight adaptive allocation method includes:

[0148] In the initial stage, the weight of the predictive control input is greater than the weight of the instantaneous control input;

[0149] During the error convergence phase, the weight of the real-time control input is gradually increased;

[0150] During the disturbance or mutation phase, the weight of the predictive control input is increased.

[0151] Specifically, the dynamic weight adaptive allocation method in this embodiment adopts a hybrid weight mechanism. In the initial stage, the weight of the predictive control input is large to prioritize the dynamic response of the system and the safety of the global formation. In the error convergence stage, the weight of the real-time control input is gradually increased, the correction effect of historical experience is enhanced, and the elimination of systematic errors is accelerated. In the disturbance or mutation stage, the weight of the predictive control input is temporarily increased to ensure the timely response and robustness of the system to new disturbances, trajectory mutations and other events.

[0152] The dynamic weight adaptive allocation mode can be described by the following formula:

[0153] ;

[0154] ;

[0155] wherein, is a historical error; is an error change rate; is a dynamic weight mapping function; is an external disturbance.

[0156] In actual flight, due to the influence of external disturbance and model error, it is difficult for the unmanned aerial vehicle to directly track the expected trajectory. Therefore, the model predictive control method is used in the embodiment to roll optimization control input in each cycle to offset real-time disturbance; at the same time, the iterative learning control method is used to iteratively correct the historical error of multiple tasks to compensate for systematic deviation. The hybrid weight mechanism can dynamically adjust the contribution proportion of the two methods, so that the system reaches a balance between dynamic response and long-term accuracy. The simulation experiment shows that the embodiment can realize high-precision cooperative tracking of multiple unmanned aerial vehicles in a complex environment, significantly improve the robustness of the system to disturbance and model error, and has good engineering application value.

[0157] Actual application scenario examples:

[0158] The complete process of the multi-unmanned aerial vehicle control method based on the hybrid control strategy is shown in Figure 4 The control input obtained is transmitted to the simulation system, and the real-time tracking error, average tracking error and other data of the multi-unmanned aerial vehicle formation are collected, and the data is analyzed to obtain the following conclusions:

[0159] Under the linear state space model, five unmanned aerial vehicles track the sinusoidal expected trajectory at an average speed of 0.8 m / s. Due to the influence of external disturbance and model simplification error, there is a slight deviation between the initial expected trajectory and the optimized trajectory (i.e. the corrected expected trajectory), but the actual trajectory follows the optimized trajectory very well, and the spatial position error is controlled within 0.1 m, and the formation spacing (1.5 m) remains stable. This shows that in the linear scene, the model predictive control method can offset the disturbance effect through real-time optimization, and the iterative learning control method can compensate for the systematic deviation of the model simplification through historical correction, ensuring the cooperative accuracy of the formation.

[0160] Under the full dynamics model, the unmanned aerial vehicles need to consider the attitude coupling and thrust constraints simultaneously, and the trajectory presents more complex dynamic characteristics. Although there is a short error fluctuation at the trajectory turning point (the maximum curvature section), the actual trajectory still closely follows the optimized trajectory, and the maximum deviation does not exceed 0.2 m, and the overall structure of the formation is not disrupted. This shows that in a scenario closer to the real physical characteristics, the embodiment can balance the nonlinear dynamics constraints and disturbance suppression, and maintain the cooperative stability of the formation.

[0161] Further, multi-unmanned aerial vehicle trajectory tracking is a typical repetitive task, and the system needs to continuously improve the accuracy through multiple iterations of learning. The average tracking error under the linear state space model gradually converges from 0.08 m in the initial iteration to within 0.05 m, and can quickly suppress error rebound when disturbances are injected. Although the error convergence process under the full dynamics model starts slightly slower (initial error 0.15 m), it stabilizes at about 0.07 m after 5 iterations, which is significantly better than the single model predictive control method or the iterative learning control method.

[0162] Further, in order to verify the effectiveness of the method of the embodiment and ensure the cooperative performance of the multi-unmanned aerial vehicles in complex environments, the embodiment also simulates external wind speed disturbances (random fluctuation range ±0.5 m / s) and model parameter drift (rotational inertia deviation ±10%), inputs dynamic external disturbance terms to the control system to simulate environmental disturbances in actual flight. By introducing model parameter errors, the system characteristics change in long-term operation are simulated.

[0163] As shown in Figure 5 linear 1 refers to executing the model predictive control method under the linear state space model; linear 2 refers to executing the iterative learning control method under the linear state space model; linear mixed refers to executing the method of the embodiment under the linear state space model; full dynamics 1 refers to executing the model predictive control method under the full dynamics model; full dynamics 2 refers to executing the iterative learning control method under the full dynamics model; and full dynamics mixed refers to executing the method of the embodiment under the full dynamics model. It can be seen that the average error of the method of the embodiment under the linear state space model (0.0683 m) is reduced by 18.8% compared with the single model predictive control method (0.0841 m) and by 28.9% compared with the single iterative learning control method (0.0961 m); and the average error of the method of the embodiment under the full dynamics model (0.0898 m) is reduced by 60.0% compared with the single model predictive control method (0.2240 m) and by 29.8% compared with the single iterative learning control method (0.1279 m). This shows that the method of the embodiment can fully integrate the real-time constraint processing capability of the model predictive control method and the experience learning advantage of the iterative learning control method, and can achieve high-precision cooperative trajectory tracking in different complexity scenarios, significantly improving the robustness and adaptability of the system.

[0164] Advantages of the present application

[0165] 1、The present application fully integrates the real-time optimization capability of model predictive control method and the experience learning advantage of iterative learning control method, through a dynamic weight distribution mechanism, taking into account the high-precision trajectory tracking of single task and the adaptive error correction of multi-round task, solving the performance limitation problem of single control method in complex environment.

[0166] 2、The present application constructs two adaptive schemes of linear state space model and full dynamics model, respectively applicable to low-speed small-range and high-speed large-range scenes, through a flexible model switching mechanism, ensuring the calculation efficiency while improving the description accuracy of real physical characteristics, enhancing the adaptability of the system to different task scenes.

[0167] 3、The present application adopts an expected trajectory distribution strategy combining main trajectory and spatial offset, combined with a distributed information interaction mechanism, realizing efficient formation keeping and cooperative control of multiple unmanned aerial vehicles, avoiding the calculation bottleneck of centralized architecture, and solving the problem of disconnection between local and global goals in distributed decision-making.

[0168] 4、The present application effectively suppresses the cumulative effect of model error and external disturbance in long-term operation through historical error iterative correction of iterative learning control method and dynamic parameter adjustment mechanism (such as adaptive change of learning rate and memory factor), significantly improving the robustness and long-term tracking accuracy of the system.

[0169] Based on the above-mentioned embodiments, the present application also provides a terminal, and a principle block diagram thereof can be as shown in Figure 6 The terminal includes a processor, a memory, a network interface, and a display screen connected through a system bus. The processor of the terminal is used to provide calculation and control capability. The memory of the terminal includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium to run. The network interface of the terminal is used to communicate with external terminals through network connection. The computer program is executed by the processor to implement the multi-unmanned aerial vehicle control method based on the hybrid control strategy. The display screen of the terminal can be a liquid crystal display screen or an electronic ink display screen.

[0170] Those skilled in the art can understand, Figure 6 The principle block diagram shown in the above-mentioned embodiments is only a block diagram of part of the structure related to the present application scheme, and does not constitute a limitation on the terminal to which the present application scheme is applied. The specific terminal can include more or fewer components than those shown in the figure, or combine certain components, or have a different component arrangement.

[0171] In one implementation, the memory of the terminal stores one or more programs configured to be executed by one or more processors to include instructions for performing the method of controlling multiple unmanned aerial vehicles based on a hybrid control strategy.

[0172] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing related hardware through a computer program. The computer program can be stored in a non-volatile computer readable storage medium, and when the computer program is executed, the computer program can include the processes of the above-mentioned embodiments. Any reference to memory, storage, database or other medium used in the embodiments of the present application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration but not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0173] In summary, the application discloses a multi-unmanned aerial vehicle control method and terminal based on a hybrid control strategy, and relates to the technical field of unmanned aerial vehicle intelligent control.The method comprises the following steps: a dynamic model of an unmanned aerial vehicle is established in advance, and task parameters are set according to task requirements; the task parameters comprise an expected trajectory; a model predictive control method is used to cooperatively control all unmanned aerial vehicles based on the task parameters, so as to obtain a predicted control input of each unmanned aerial vehicle; the model predictive control method comprises the following steps: a future state is predicted based on a current observation state of a system and the dynamic model, and an optimal control sequence in the future is obtained by solving an optimization problem, and a first control input of the optimal control sequence is taken as the predicted control input; wherein, the optimization problem is constructed based on the task parameters; the current observation state of the system is used to reflect an observation result of a system state; an iterative learning control method is used to cooperatively control all unmanned aerial vehicles based on the task parameters, so as to obtain an instant control input of each unmanned aerial vehicle; the iterative learning control method comprises the following steps: an expected trajectory in the next round is corrected based on a full-time-domain historical error, and an instant control input is generated according to an error between a current state of the system and the expected trajectory; wherein, the current state of the system is used to reflect a set of internal variables of the system at the current time; and a target control input of each unmanned aerial vehicle is synthesized according to the predicted control input and the instant control input.The application fully combines the real-time optimization capability of the model predictive control method and the experience learning advantage of the iterative learning control method, can take into account high-precision trajectory tracking of a single task and adaptive error correction of multiple tasks, and solves the problem that a single control method is limited in performance in a complex environment.

[0174] It should be understood that the application of the present application is not limited to the above examples, and can be improved or changed according to the above description for those skilled in the art, and all these improvements and changes shall belong to the protection scope of the appended claims of the present application.

Claims

1. A multi-unmanned aerial vehicle control method based on a hybrid control strategy, characterized in that, The method includes: A dynamic model of the UAV is pre-established, and mission parameters are set according to mission requirements; the mission parameters include the desired trajectory. A model predictive control method is used to coordinate the control of all UAVs based on the task parameters, thereby obtaining the predictive control input for each UAV. The model predictive control method includes: predicting the future state based on the current observed state of the system and the dynamic model, and obtaining the optimal control sequence for the future by solving an optimization problem, using the first control input of the optimal control sequence as the predictive control input; wherein, the optimization problem is constructed based on the task parameters; the current observed state of the system is used to reflect the observation results of the system state. Through an iterative learning control method, all UAVs are coordinated and controlled based on the task parameters to obtain the instantaneous control input for each UAV. The iterative learning control method includes: correcting the expected trajectory for the next round based on the full-time domain historical error, and generating instantaneous control input based on the error between the current state of the system and the expected trajectory. The current state of the system is used to reflect the set of internal variables of the system at the current moment. The target control input for each UAV is synthesized based on the predicted control input and the real-time control input. The model predictive control method includes the following three execution processes: The process of predicting future states: In each control cycle, starting from the current observed state of the system, the future is recursively predicted based on the dynamic model of the UAV. The state sequence within a step, the prediction process follows discrete dynamics; The optimization decision-making process of the objective function: In each optimization cycle, the system constructs the following general prototype of a weighted, polynomial objective function around the desired trajectory and control objective: ； wherein, is the objective function; is the step number; is the is the three-dimensional spatial position at time instant is the is the reference three-dimensional spatial position at time instant is the is the three-dimensional velocity component at time instant is the is the reference three-dimensional velocity component at time instant is the is the control input at time instant is the weight coefficient for position, is the weight coefficient for position and velocity error; is the weight coefficient for control input energy consumption, is the weight coefficient for input variation; all weight parameters are diagonal matrices; Rolling time-domain optimization and closed-loop feedback process: In each cycle, only the first control input of the optimized control input sequence is applied to the system, the system state is updated accordingly, and the next cycle is entered. The prediction, optimization and application process is repeated to form rolling time-domain optimization and dynamic closed-loop feedback. The desired trajectory for the next round is corrected based on the historical error across the entire time domain. Real-time control inputs are generated based on the error between the system's current state and the desired trajectory, including: Through the upper-level trajectory correction learning unit, after each round of task execution, the expected trajectory for the next round is corrected based on the full-time domain historical error, and the corrected trajectory is used as the expected trajectory for the next round; the correction scope includes: each trajectory point and / or each spatial dimension; The real-time tracking controller at the lower level generates the instantaneous control input at each moment based on the error between the current state of the system and the desired trajectory. For the upper-level trajectory correction learning unit: after each round of task execution, the expected trajectory for the next round is corrected point by point based on the historical error in the entire time domain; The desired trajectory of the wheel is , the system state is , and the corresponding historical error is : ; The correction formula of the desired trajectory is: ; wherein, is a dynamic learning rate; the corrected desired trajectory is used as the desired trajectory of the next round for the lower-level control to call. For the lower-level real-time tracking controller: The lower-level controller is responsible for, at each moment, adjusting the system's current state. Compared with the corrected expected trajectory Error between ,generate Time control input ; For three-axis acceleration control of a linear state space model, the control law of a real-time tracking controller is: ; where, is a proportional term; is a derivative term; is an integral gain; is an integral variable; is a differential of the integral variable .

2. The multi-UAV control method based on a hybrid control strategy according to claim 1, characterized in that, The types of dynamic models include: linear state-space models and full dynamic models. 3.The method of claim 1, wherein, Based on the current observed state of the system and the dynamic model, the future state is predicted, and the optimal control sequence for the future is obtained by solving an optimization problem, including: In each control cycle, starting from the current observed state of the system, the state sequence within a few future time steps is recursively predicted based on the dynamic model. In each optimization cycle, a weighted polynomial objective function is constructed based on the desired trajectory and control objective, and an optimization problem is established based on the objective function. The optimal control sequence is obtained by solving the state sequence and the optimization problem.

4. The multi-UAV control method based on a hybrid control strategy according to claim 3, characterized in that, The constraints corresponding to the optimization problem include: dynamic constraints, input physical boundaries, scalable state constraints, and output constraints.

5. The method of claim 1, wherein, In the model predictive control method, each UAV is managed in batches using object-oriented and array-based methods, and each UAV runs the model predictive control method independently.

6. The method of claim 1, wherein, The target control input for each UAV is synthesized based on the predicted control input and the real-time control input, including: The predicted control input and the real-time control input are weighted and fused to synthesize the target control input for each UAV.

7. The multi-UAV control method based on a hybrid control strategy according to claim 6, characterized in that, The weight allocation methods corresponding to weighted fusion include: static weight allocation method and / or dynamic weight adaptive allocation method.

8. The multi-UAV control method based on a hybrid control strategy according to claim 7, wherein, The dynamic weight adaptive allocation method includes: In the initial stage, the weight of the predictive control input is greater than the weight of the instantaneous control input; During the error convergence phase, the weight of the real-time control input is gradually increased; During the disturbance or mutation phase, the weight of the predictive control input is increased.

9. A terminal, characterized by comprising: The terminal includes a memory and one or more processors; the memory stores one or more programs; the programs contain instructions for executing the multi-UAV control method based on a hybrid control strategy as described in any one of claims 1 to 8; the processors are used to execute the programs.

Images