Model prediction control method and system for unmanned aerial vehicle with uncertain load mass
By constructing a model predictive controller with multiple local linear models, the problems of flight stability and trajectory tracking performance of quadrotor UAVs caused by load mass uncertainty were solved. The controller achieved online learning and adaptation to dynamic changes in the system, and realized robust trajectory tracking control.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- CHANGSHU INSTITUTE OF TECHNOLOGY
- Filing Date
- 2026-02-09
- Publication Date
- 2026-05-05
AI Technical Summary
When a quadcopter drone carries a suspended load, the dynamic characteristics change due to the uncertainty of the load mass, which affects flight stability and trajectory tracking performance. Traditional control methods are difficult to effectively cope with the sudden changes and time-varying dynamics of the load mass.
A model predictive controller is constructed using multiple local linear models. By learning online and adapting to changes in load quality, a general discrete state-space system model is established to achieve the prediction and superposition of system observations and control variables, thus proactively responding to uncertainties.
Robust and accurate trajectory tracking control of UAVs under varying load conditions is achieved, maintaining system stability and excellent trajectory tracking performance, outperforming traditional PD and conventional model predictive control.
Smart Images
Figure CN121978952A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of unmanned aerial vehicle (UAV) control technology. It relates to a model predictive control method and system for UAVs with uncertain payload mass. Background Technology
[0002] Unmanned aerial vehicles (UAVs), especially quadcopter drones, have broad application prospects in logistics, search and rescue, and agriculture due to their simple mechanical structure and vertical takeoff and landing (VTOL) and hovering capabilities. However, quadcopter drones are inherently underactuated systems, requiring control of six degrees of freedom with only four inputs, making stability and control inherently challenging. When carrying suspended loads, the swaying of the load introduces complex external disturbances, significantly altering the quadcopter drone's dynamic characteristics and severely impacting its flight stability and trajectory tracking performance. In practical applications, the mass of the load is often unknown or may suddenly change during transport; this uncertainty translates into time-varying external disturbances.
[0003] To address the problem of load mass uncertainty, traditional control methods mostly rely on the robustness of the controller itself to passively cope with uncertainty. When the load mass changes significantly or undergoes drastic dynamic changes, the control performance deteriorates sharply, even leading to system instability. While model predictive control methods have some feedforward optimization capabilities, they are usually based on a fixed, pre-calibrated system model. When the actual system dynamics become mismatched with the model due to changes in load mass, the control performance deteriorates significantly, lacking the ability to adapt to time-varying dynamics. Therefore, there is an urgent need in this field for a control method that can actively and online learn and adapt to changes in task parameters such as load mass, thereby ensuring stable and precise flight of the quadrotor even in uncertain environments. Summary of the Invention
[0004] The purpose of this invention is to provide a model predictive control method and system for unmanned aerial vehicles with uncertain load mass. This method can learn online and adapt to the dynamic changes of the system caused by changes in load mass, actively respond to uncertainties, and thus achieve robust and accurate trajectory tracking control.
[0005] The technical solution to achieve the purpose of this invention is as follows: A model predictive control method for unmanned aerial vehicles with uncertain payload mass includes the following steps: S01: Establish a dynamic model of the UAV that includes load disturbances, and convert it into a general discrete state-space system model; S02: Construct multiple model prediction controllers, each model prediction controller including a local linear model. Each model prediction controller is based on the corresponding local linear model and predicts the predicted value of the observation based on the current system state. The predicted values of all the calculated observations are superimposed to obtain the predicted value of the system observation. S03: Each model predictive controller calculates the predicted control quantity based on the corresponding local linear model and the current system state. The predicted control quantity is obtained by superimposing all the calculated predicted control quantity values.
[0006] In the preferred technical solution, the UAV dynamics model incorporating load interference is established as follows:
[0007] in, These represent the translational displacements in the x, y, and z directions, respectively. These represent the roll angle, pitch angle, and yaw angle, respectively. This represents the fixed moments of inertia of the organism in the x, y, and z directions. , and These represent the relative positions of the quadcopter and the payload in the x, y, and z directions, respectively. The length of the rope. This represents the distance from the rotor center to the center of mass of the quadcopter UAV. The dot symbol and the double dot symbol represent the first and second derivatives, respectively. For system input, This indicates the thrust generated by each of the four rotors. Refers to the mass of the load. For the quality of drones.
[0008] In the preferred technical solutions, the conversion to a general discrete state-space system model includes: The UAV dynamics model is discretized into translational and rotational subsystems. The translation subsystem is as follows:
[0009] The rotating subsystem is:
[0010] in, , k For the control index after model discretization, , , These represent the velocities of the drone in the x, y, and z directions, respectively. , , denoted as the velocities of the UAV in the roll, pitch, and yaw directions, respectively, and g is the gravitational acceleration constant. Indicates the time interval used for discretization; , , , This represents the torque of each rotor blade; The two systems described above can be converted into a general discrete state-space system as follows:
[0011] in, Indicates the system status. Indicates system output, Indicates system input, Let A represent the disturbance vector caused by the load, B represent the state matrix, C represent the control matrix, and C represent the observation matrix. Indicates state noise. This indicates observation noise.
[0012] In the preferred technical solution, the method for obtaining the predicted values of the system observations in step S02 includes: The predicted value of the observation obtained by the model prediction controller is:
[0013] in, For system status, To predict control quantities, , For coefficients; The predicted values of the system observations are:
[0014] Where W is the sum of all weights of the normalization factor. Local linear model i Input weights, I The number of local linear models, This refers to the actual control quantity.
[0015] In the preferred technical solution, each local linear model i Input weights The Gaussian kernel function is calculated as follows:
[0016] in, For the input vector, For the first i The center of a local linear model kernel Determine the shape and size of the Gaussian model.
[0017] In the preferred technical solution, step S03 yields the predicted value of the system control quantity:
[0018] in, Local linear model i The predicted control quantity for the next time step.
[0019] In the preferred technical solution, the method for calculating the predicted value of the control quantity includes: Optimal prediction of the next system state Obtain it as follows:
[0020] Where Z is and Cross covariance; The optimal predicted value of the next system state is obtained by solving the problem using a solver. After that, from that moment k+1 arrive k+N The optimal estimates of the state and / or output over the entire prediction time domain are obtained as follows:
[0021] in, i = 1,…,N , N To predict the total number of steps; The predicted control quantity is obtained by minimizing the control objective function in the following equation: , :
[0022] Where n is the number of prediction steps, For a moment k The trajectory S It is a weight matrix. S It is a symmetric and positive definite quadratic form , .
[0023] This invention also discloses a model predictive control system for a drone with uncertain payload mass, used to implement the aforementioned model predictive control method for a drone with uncertain payload mass, comprising: The model building module establishes a dynamic model of the UAV that includes load disturbances and converts it into a general discrete state-space system model. The system observation prediction module constructs multiple model prediction controllers. Each model prediction controller includes a local linear model. Each model prediction controller is based on the corresponding local linear model and predicts the observation value according to the current system state. The predicted values of all the calculated observation values are superimposed to obtain the predicted value of the system observation. The system control quantity prediction module calculates the predicted control quantity value based on the corresponding local linear model and the current system state for each model predictive controller. The predicted control quantity value is obtained by superimposing all the calculated predicted control quantity values.
[0024] The present invention also discloses a computer storage medium storing a computer program, which, when executed, implements the above-described model predictive control method for unmanned aerial vehicles with uncertain load mass.
[0025] The present invention also discloses an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor runs the computer program stored in the memory. When the computer program is executed, it implements the above-described model predictive control method for unmanned aerial vehicles with uncertain load mass.
[0026] Compared with the prior art, the significant advantages of this invention are: This invention can proactively identify and adapt to changes in system dynamics such as load quality. It can learn online and adapt to system dynamic changes caused by load quality variations, proactively responding to uncertainties, thereby achieving robust and accurate trajectory tracking control. Even when faced with sudden changes and uncertainties in load quality, it maintains excellent trajectory tracking performance and system stability, far outperforming traditional PD and conventional model predictive control. Attached Figure Description
[0027] Figure 1 This is a flowchart of the model predictive control method for a drone with uncertain payload quality in this embodiment; Figure 2 A schematic diagram of the construction of a drone model with a cable load; Figure 3 This is a schematic diagram illustrating the principle of the model predictive control method in this embodiment; Figure 4 This is a comparison chart of control performance. Detailed Implementation
[0028] The principle of this invention is to use multiple locally linear models to update the dynamic model of a time-varying nonlinear system related to changes in load quality online. This allows the system to learn and adapt online to the dynamic changes caused by variations in load quality, proactively addressing uncertainties and thus achieving robust and accurate trajectory tracking control.
[0029] Example 1: like Figure 1 As shown, a model predictive control method for a UAV with uncertain load mass includes the following steps: S01: Establish a dynamic model of the UAV that includes load disturbances, and convert it into a general discrete state-space system model; S02: Construct multiple model prediction controllers, each model prediction controller including a local linear model. Each model prediction controller is based on the corresponding local linear model and predicts the predicted value of the observation based on the current system state. The predicted values of all the calculated observations are superimposed to obtain the predicted value of the system observation. S03: Each model predictive controller calculates the predicted control quantity based on the corresponding local linear model and the current system state. The predicted control quantity is obtained by superimposing all the calculated predicted control quantity values.
[0030] In a preferred embodiment, the UAV dynamics model incorporating load disturbances is established as follows:
[0031] in, These represent the translational displacements in the x, y, and z directions, respectively. These represent the roll angle, pitch angle, and yaw angle, respectively. This represents the fixed moments of inertia of the organism in the x, y, and z directions. , and These represent the relative positions of the quadcopter and the payload in the x, y, and z directions, respectively. The length of the rope. This represents the distance from the rotor center to the center of mass of the quadcopter UAV. The dot symbol and the double dot symbol represent the first and second derivatives, respectively. For system input, This indicates the thrust generated by each of the four rotors. Refers to the mass of the load. For the quality of drones.
[0032] A preferred embodiment, converting to a general discrete state-space system model includes: The UAV dynamics model is discretized into translational and rotational subsystems. The translation subsystem is as follows:
[0033] The rotating subsystem is:
[0034] in, ,k For the control index after model discretization, , , These represent the velocities of the drone in the x, y, and z directions, respectively. , , denoted as the velocities of the UAV in the roll, pitch, and yaw directions, respectively, and g is the gravitational acceleration constant. Indicates the time interval used for discretization; , , , This represents the torque of each rotor blade; The two systems described above can be converted into a general discrete state-space system as follows:
[0035] in, Indicates the system status. Indicates system output, Indicates system input, Let A represent the disturbance vector caused by the load, B represent the state matrix, C represent the control matrix, and C represent the observation matrix. Indicates state noise. This indicates observation noise.
[0036] In a preferred embodiment, the method for obtaining the predicted values of system observations in step S02 includes: The predicted value of the observation obtained by the model prediction controller is:
[0037] in, For system status, To predict control quantities, , For coefficients; The predicted values of the system observations are:
[0038] Where W is the sum of all weights of the normalization factor. Local linear model i Input weights, I The number of local linear models, This refers to the actual control quantity.
[0039] In a preferred embodiment, each local linear model i Input weights The Gaussian kernel function is calculated as follows:
[0040] in, For the input vector, For the first i The center of a local linear model kernel Determine the shape and size of the Gaussian model.
[0041] In a preferred embodiment, step S03 yields the predicted value of the system control quantity:
[0042] in, Local linear model i The predicted control quantity for the next time step.
[0043] In a preferred embodiment, the method for calculating the predicted control quantity includes: Optimal prediction of the next system state Obtain it as follows:
[0044] Where Z is and Cross covariance; The optimal predicted value of the next system state is obtained by solving the problem using a solver. After that, from that moment k+1 arrive k+N The optimal estimates of the state and / or output over the entire prediction time domain are obtained as follows:
[0045] in, i = 1,…,N , N To predict the total number of steps; The predicted control quantity is obtained by minimizing the control objective function in the following equation: , :
[0046] Where n is the number of prediction steps, For a moment k The trajectory S It is a weight matrix. S It is a symmetric and positive definite quadratic form , .
[0047] In another embodiment, a computer storage medium stores a computer program that, when executed, implements the above-described model predictive control method for a drone with uncertain load mass.
[0048] The specific implementation uses the method described above, which will not be elaborated here.
[0049] In another embodiment, an electronic device includes a memory and a processor, wherein the memory stores a computer program, and the processor runs the computer program stored in the memory, wherein the computer program, when executed, implements the above-described model predictive control method for a drone with uncertain load mass.
[0050] The specific implementation uses the method described above, which will not be elaborated here.
[0051] In another embodiment, a model predictive control system for a drone with uncertain payload mass, used to implement the above-described model predictive control method for a drone with uncertain payload mass, includes: The model building module establishes a dynamic model of the UAV that includes load disturbances and converts it into a general discrete state-space system model. The system observation prediction module constructs multiple model prediction controllers. Each model prediction controller includes a local linear model. Each model prediction controller is based on the corresponding local linear model and predicts the observation value according to the current system state. The predicted values of all the calculated observation values are superimposed to obtain the predicted value of the system observation. The system control quantity prediction module calculates the predicted control quantity value based on the corresponding local linear model and the current system state for each model predictive controller. The predicted control quantity value is obtained by superimposing all the calculated predicted control quantity values.
[0052] Specifically, the following describes the workflow of a model predictive control system for a drone with uncertain load mass, using a preferred embodiment as an example, including the following steps: S1: Establish a quadrotor dynamic model that includes suspension load disturbances, and convert it into a general discrete state-space system model.
[0053] S2: Use multiple local linear models to update the dynamic model of the time-varying nonlinear system related to load quality changes online.
[0054] S3: Based on the observations in S2, the predicted value of the control variable for the next time step is set as follows:
[0055] in, It is the sum of all weights of the normalization factor.
[0056] S4: Repeat S2-S3 until the task is completed.
[0057] Before constructing the dynamic model of a quadcopter UAV using a cable-driven load, the following assumptions are made: (1) The fuselage of the quadcopter UAV is rigid and symmetrical.
[0058] (2) The center of gravity of the quadcopter UAV coincides with the origin of the fixed coordinate system of the fuselage.
[0059] (3) The suspension point of the cable is located exactly at the center of gravity of the quadcopter drone.
[0060] (4) The mass of the rope is negligible, and the rope force is always non-negative and cannot be ignored.
[0061] (5) Apart from the load, the quadcopter drone is not subjected to any other external interference or force, such as gusts of wind.
[0062] like Figure 2 As shown, the origin of the load coordinate system is located at the center of gravity of the quadcopter drone, and the coordinate axes are parallel to the axis of the quadcopter drone. In the diagram, Load refers to the load suspended by the cable on the drone. CoG refers to the center of gravity of the drone. The light blue circles represent the quadcopter drone, the dark blue circles represent the center of gravity of the quadcopter drone with the load, and the light blue line segment between the center of gravity and the load represents the cable, the length of which is... . , and These represent the relative positions of the quadcopter and the payload in the x, y, and z directions, respectively.
[0063] The dynamic model of the UAV with cable load is as follows:
[0064] in, This represents the translational displacement in the x, y, and z directions. This indicates the roll angle, pitch angle, and yaw angle. It represents the fixed moment of inertia of the machine body. This indicates the distance from the rotor center to the center of mass of the quadcopter. The dot symbol and double dot symbol represent the corresponding velocity and acceleration, respectively. , , , This is the system input, i.e., the combination of thrust from each rotor. This indicates the thrust generated by each of the four rotors. Both represent rotor torque and are proportional to the square of the angular velocity. Refers to the mass of the load. The mass of the quadcopter drone.
[0065] In order to use model predictive control in the future, the above model is first discretized into translation subsystem and rotation subsystem.
[0066] The translation subsystem is as follows:
[0067] The rotating subsystem is:
[0068] in, k is the control index after model discretization, which can be understood as time, and g is the gravitational acceleration constant. This indicates the time interval used for discretization.
[0069] The two systems described above can be converted into a general discrete state-space system as follows:
[0070] in, Indicates the system status. Indicates system output, Indicates system input (system control quantity). Let A represent the disturbance vector caused by the load. Let A represent the state matrix, B represent the control matrix, and C represent the observation matrix. Indicates state noise. This represents observation noise. Both types of noise are assumed to follow a Gaussian distribution with a mean of zero, and... and Each has its own covariance and Their values are all set to 0.0001. and The cross covariance is Z, which is set to 0 to minimize the interference of noise caused by non-load mass.
[0071] Based on the general discrete state-space system form, the matrices of the translation subsystem are:
[0072] Based on the general discrete state-space system form, the matrices of the rotating subsystem are:
[0073] For any subsystem (translation subsystem and rotation subsystem), such as Figure 3 As shown, a local linear model is paired with a model predictive controller (MPC), which constitutes a set (or group), denoted by index i. Existing linear models can be used for local linear models. Figure 3 middle It is the drone's reference trajectory (i.e., the trajectory it is expected to reach).
[0074] Each system operates independently, meaning each MPC calculates the control input independently based on its corresponding local linear model and the current system state. Finally, the final control quantity of the system is the weighted sum of these calculated control quantities, i.e. .
[0075] Using multiple model predictive controllers, the current observations are predicted using the following formula:
[0076] Among them, each local linear model predicts the current system observations. , , For coefficients; , , It can be equivalently understood as an observation matrix containing N prediction steps (the second row and thereafter need to be multiplied by the system state matrix, which is purely mathematical knowledge). It can be equivalently understood as a control matrix containing N prediction steps (the second row and thereafter need to be multiplied by the control matrix, and the third row and thereafter need to be multiplied by the system state matrix in the middle, which is pure mathematical knowledge).
[0077] W is the sum of all weights of the normalization factor (i.e., ), input weights for each local linear model Calculations are performed using kernel functions, typically the Gaussian kernel function. , For input vectors , The center of the i-th local model kernel, i.e. , The shape and size of the Gaussian model are determined; it is a square matrix of size N. N is selected based on the actual application; in this invention, N can be taken as N. The identity matrix of N. N is the number of prediction steps in the conventional model predictive control method.
[0078] Input vector This is a set of column vectors calculated based on the current system state and the current actual control inputs. The center of the local model kernel. A set of column vectors calculated from the current system state and the predicted control variables.
[0079] The local linear model is a preset partial model. For example, 10 models are preset / pre-provided, such as a standalone drone model (without load), and drones with loads of 0.01kg, 0.05kg, 0.1kg, 0.15kg, 0.2kg, 0.25kg, 0.30kg, 0.35kg, 0.40kg, etc.
[0080] The models provided here vary depending on the application scenario and the robot itself. For example, if the maximum payload of a drone is 0.6 kg, then the local model can provide a maximum payload of 0.6 kg.
[0081] In addition, this example uses a load mass interval of 0.05 kg, which can be further increased or decreased (step size) depending on the control performance requirements.
[0082] Each (Right now The i-th ) and the predicted control values for their future time steps , ... can all be calculated using conventional model predictive control methods. This is a conventional method, and a specific calculation process is given below: and They represent time. Time is based on and includes time. The information (based on past information up to time h (inclusive)) is used to estimate the state and output, where So, the next state The optimal prediction can be obtained as follows:
[0083] Where Z is and Cross covariance; The above regarding P The equations are called discrete-time algebraic Riccati equations, and these equations can be solved using a dedicated solver in MATLAB.
[0084] To obtain the optimal prediction value for the next state Then, the predicted values over the entire prediction time domain can be considered. Model predictive control typically requires starting from time point... k+1 arrive k+N The estimated values of the state and / or output over the entire prediction time domain, and can only be based on the current time. k These predictions are made based on information including [the data provided].k+2 arrive k+N The optimal estimate can be obtained as follows.
[0085]
[0086] in, i = 1,…,N , Used to distinguish at time k+i The actual input and the input used for prediction purposes.
[0087] Time k The trajectory of time is defined as By minimizing the control objective function in the following formula, we can obtain... , , etc., which are the methods of this invention. , , wait.
[0088]
[0089] Where n is the number of prediction steps in the conventional model prediction (MPC) method. S is the weight matrix, used to define and adjust the control objective function J. Its value varies depending on the application scenario. S is assumed to be symmetric and positive definite. Quadratic form A mechanism is provided that allows different weights to be assigned to different outputs. For Similarly, it allows for different penalties to be imposed on different input variations. It is the input applied at the current moment, that is, the input calculated in the previous cycle.
[0090] Based on the above predictions, the actual control quantity adopted in the next moment is:
[0091] in, It is the sum of all weights of the normalization factor.
[0092] The calculation process can learn online and adapt to dynamic changes in the system caused by changes in load quality.
[0093] Based on the above method, the actual control effect performance is verified as follows.
[0094] Assuming the following parameters are used for the quadcopter drone: the drone's mass is 1.4 kg, and its moments of inertia about both the X and Y axes are 0.003. The moment of inertia about the Z-axis is 0.04. The thrust coefficient is 3.13×10-5, the aerodynamic drag coefficient is 7.5×10-7, the distance from the rotor center to the center of the UAV is 0.2m, the rope length is 0.5m, and the maximum weight of the payload it can carry is 0.4kg.
[0095] Assuming the requirement is for the drone to perform planar circular motion while maintaining an altitude of 1m, , , The initial load mass is 0.1 kg, but it suddenly increases to 0.4 kg at approximately one-quarter of the distance traveled. In the method proposed in this invention... Set to 0.001, the comparison method is the traditional PD control method and the conventional model prediction method: In the traditional PD control method proportionality coefficient , , The values are 10, 10, 20, 50, 50, and 30. differential coefficients , , The values are 8, 8, 15, 20, 20, 15 respectively; the prediction step and control step lengths of the conventional model predictive control method are both 25.
[0096] Control effect such as Figure 4 As shown, when the load quality suddenly changes, the PD controller can no longer track the desired trajectory near point (1,0). In fact, the PD controller stops working when the actual trajectory moves a certain distance away from the desired trajectory. However, despite the online adjustment process from point (1,0) to (0,-1), the method proposed in this invention can still follow the desired trajectory, while the performance of the standard model predictive controller is worse than that of this invention, with a larger maximum deviation and a longer time to track the desired trajectory. This confirms that the method proposed in this invention is more adaptable to different tasks or environments than the conventional model predictive controller.
[0097] The above embodiments are preferred embodiments of the present invention, but the embodiments of the present invention are not limited to the above embodiments. Any changes, modifications, substitutions, combinations, or simplifications made without departing from the spirit and principle of the present invention shall be considered equivalent substitutions and shall be included within the protection scope of the present invention.
Claims
1. A model predictive control method for unmanned aerial vehicles with uncertain payload mass, characterized in that, Includes the following steps: S01: Establish a dynamic model of the UAV that includes load disturbances, and convert it into a general discrete state-space system model; S02: Construct multiple model prediction controllers, each model prediction controller including a local linear model. Each model prediction controller is based on the corresponding local linear model and predicts the predicted value of the observation based on the current system state. The predicted values of all the calculated observations are superimposed to obtain the predicted value of the system observation. S03: Each model predictive controller calculates the predicted control quantity based on the corresponding local linear model and the current system state. The predicted control quantity is obtained by superimposing all the calculated predicted control quantity values.
2. The model predictive control method for UAVs with uncertain load mass according to claim 1, characterized in that, The dynamic model of the UAV including load disturbance is established as follows: in, These represent the translational displacements in the x, y, and z directions, respectively. These represent the roll angle, pitch angle, and yaw angle, respectively. This represents the fixed moments of inertia of the organism in the x, y, and z directions. , and These represent the relative positions of the quadcopter and the payload in the x, y, and z directions, respectively. The length of the rope. This represents the distance from the rotor center to the center of mass of the quadcopter UAV. The dot symbol and the double dot symbol represent the first and second derivatives, respectively. For system input, This indicates the thrust generated by each of the four rotors. Refers to the mass of the load. For the quality of drones.
3. The model predictive control method for UAVs with uncertain payload mass according to claim 2, characterized in that, The conversion to a general discrete state-space system model includes: The UAV dynamics model is discretized into translational and rotational subsystems. The translation subsystem is as follows: The rotating subsystem is: in, , k For the control index after model discretization, , , These represent the velocities of the drone in the x, y, and z directions, respectively. , , denoted as the velocities of the UAV in the roll, pitch, and yaw directions, respectively, and g is the gravitational acceleration constant. Indicates the time interval used for discretization; , , , This represents the torque of each rotor blade; The translation and rotation subsystems are converted into a general discrete state-space system as follows: in, Indicates the system status. Indicates system output, Indicates system input, Let A represent the disturbance vector caused by the load, B represent the state matrix, C represent the control matrix, and C represent the observation matrix. Indicates state noise. This indicates observation noise.
4. The model predictive control method for UAVs with uncertain payload mass according to claim 3, characterized in that, The methods for obtaining the predicted values of the system observations in step S02 include: The predicted value of the observation obtained by the model prediction controller is: in, For system status, To predict control quantities, , For coefficients; The predicted values of the system observations are: Where W is the sum of all weights of the normalization factor. Local linear model i Input weights, I The number of local linear models, This refers to the actual control quantity.
5. The model predictive control method for unmanned aerial vehicles with uncertain load mass according to claim 4, characterized in that, Each local linear model i Input weights The Gaussian kernel function is calculated as follows: in, For the input vector, For the first i The center of a local linear model kernel Determine the shape and size of the Gaussian model.
6. The model predictive control method for a UAV with uncertain payload mass according to claim 5, characterized in that, Step S03 yields the predicted value of the system control quantity: in, Local linear model i The predicted control quantity for the next time step.
7. The model predictive control method for unmanned aerial vehicles with uncertain load mass according to claim 6, characterized in that, The methods for calculating the predicted values of control quantities include: Optimal prediction of the next system state Obtain it as follows: Where Z is and Cross covariance; The optimal predicted value of the next system state is obtained by solving the problem using a solver. After that, from that moment k+1 arrive k+N The optimal estimates of the state and / or output over the entire prediction time domain are obtained as follows: in, i = 1,…,N , N To predict the total number of steps; The predicted control quantity is obtained by minimizing the control objective function in the following equation: , : Where n is the number of prediction steps, For a moment k The trajectory S It is a weight matrix. S It is a symmetric and positive definite quadratic form , .
8. A model predictive control system for a UAV with uncertain payload mass, characterized in that, A method for implementing the model predictive control of a UAV with uncertain payload mass as described in any one of claims 1-7, comprising: The model building module establishes a dynamic model of the UAV that includes load disturbances and converts it into a general discrete state-space system model. The system observation prediction module constructs multiple model prediction controllers. Each model prediction controller includes a local linear model. Each model prediction controller is based on the corresponding local linear model and predicts the observation value according to the current system state. The predicted values of all the calculated observation values are superimposed to obtain the predicted value of the system observation. The system control quantity prediction module calculates the predicted control quantity value based on the corresponding local linear model and the current system state for each model predictive controller. The predicted control quantity value is obtained by superimposing all the calculated predicted control quantity values.
9. A computer storage medium having a computer program stored thereon, characterized in that, When the computer program is executed, it implements the model predictive control method for unmanned aerial vehicles with uncertain load mass as described in any one of claims 1-7.
10. An electronic device comprising a memory and a processor, characterized in that, The memory stores a computer program, and the processor runs the computer program stored in the memory. When the computer program is executed, it implements the model predictive control method for a drone with uncertain load mass as described in any one of claims 1-7.