Discrete incremental sliding mode quadrotor control method based on neural network and disturbance observation

Abstract

The application discloses a kind of discrete incremental sliding mode quadrotor control methods based on neural network and disturbance observation, belong to the field of quadrotor unmanned aerial vehicle, including steps: S1, considering the uncertainty factor of aircraft establishes six degrees of freedom quadrotor unmanned aerial vehicle model;S2, design disturbance observer and under the condition of gust modeling wind, design wind online predictor based on disturbance observer and neural network;S3, design the solution of sensor error, time delay based on neural network fault problem;S4, respectively in attitude loop and position loop design discrete incremental sliding mode fault-tolerant controller.The application has the advantages of strong robustness, high fault tolerance and high-precision control.

Description

Technical Field

[0001] This invention relates to the field of quadrotor unmanned aerial vehicles (UAVs), and more specifically, to a discrete incremental sliding mode quadrotor control method based on neural networks and disturbance observation. Background Technology

[0002] A quadcopter is a multi-rotor aircraft consisting of four symmetrically distributed rotors. It possesses vertical takeoff and landing (VTOL) and hovering capabilities, thus finding widespread application in fields such as drones, aerial photography, and logistics delivery. To achieve precise control of quadcopters, numerous control methods and algorithms have been proposed. However, traditional quadcopter control methods typically employ continuous controllers. These continuous controllers are often based on continuous-time models, neglecting the fact that the controller law is discrete. Discrete controllers are better suited to the discrete nature of real-world systems, making the development of a control method specifically for discrete systems crucial. Discrete incremental sliding mode control (INDI) exhibits good stability in handling uncertainties in aircraft models, sudden actuator failures, and structural damage. While INDI offers advantages such as robustness, high-precision control, simplicity of implementation, high fault tolerance, and applicability to discrete systems, its application in quadcopter control remains immature.

[0003] Meanwhile, the control performance of a quadcopter is also affected by factors such as control sensor errors, time delays, and malfunctions. Sensor errors may be caused by noise, nonlinear characteristics, or inaccurate calibration of the sensor itself, which can affect the accurate perception of the system status. Time delay refers to the time delay in signal transmission or processing; for example, the signal transmission from the sensor to the controller requires a certain amount of time, which can cause the controller response to lag, thereby reducing control performance. In addition, sensor failures or communication failures can further compromise the reliability and stability of the system. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the prior art and provide a discrete incremental sliding mode quadrotor control method based on neural networks and disturbance observation, which has the advantages of strong robustness, high fault tolerance and high precision control.

[0005] The objective of this invention is achieved through the following approach:

[0006] A discrete incremental sliding mode quadrotor control method based on neural networks and disturbance observation includes the following steps:

[0007] S1. Considering the uncertainties of the aircraft, establish a six-degree-of-freedom quadcopter UAV model;

[0008] S2, Design a disturbance observer and model wind force under gust conditions, and design an online wind force predictor based on the disturbance observer and neural network;

[0009] S3, Design a solution for sensor error and time delay fault problems based on neural networks;

[0010] S4, a discrete incremental sliding mode fault-tolerant controller is designed in the attitude loop and the position loop respectively.

[0011] Furthermore, the process of establishing a six-degree-of-freedom quadcopter UAV model considering aircraft uncertainties specifically includes the following sub-steps:

[0012] Establish the following aircraft dynamics model:

[0013]

[0014] Where x, y, z represent the position coordinates of the quadcopter relative to the inertial frame; θ, φ, ψ represent the Euler angles of the quadcopter, namely pitch, roll, and yaw, respectively; p, q, r represent the rotational angular velocities of the quadcopter in the body coordinate system; J xb J yb J zb The x, y, and z axes represent the inertial constants of the quadrotor, m represents the mass of the quadrotor, and f represents the inertial constants around the x, y, and z axes. r τ is the magnitude of the resultant force of the quadcopter. xr ,τ yr ,τ zr f is the resultant torque generated by the four rotors in the body coordinate system. dx ,f dy ,f dz For the interfering force, τ dx ,τ dy ,τ dz For the disturbance torque; S(·) represents sin(·), C(·) represents cos(·), This represents the acceleration of an aircraft in an inertial frame of reference. This represents the angular acceleration of the quadcopter in the body coordinate system.

[0015] The expressions for the force and torque generated by the rotor are:

[0016]

[0017] in, Ω1, Ω2, Ω3, Ω4 represent the rotational speeds of rotors 1, 2, 3, and 4, respectively; k and b are the thrust and torque constants of the quadcopter; and L is the length of the quadcopter's arm. T This represents the transpose of a matrix.

[0018] Furthermore, in step S2, the neural network is an LSTM network.

[0019] Furthermore, in step S2, the model is trained using historical wind data and relevant parameters, and combined with a disturbance observer, to achieve accurate online prediction of convective wind force.

[0020] Furthermore, step S2 includes the sub-step of optimizing the predictor's network structure and training algorithm to improve prediction performance and robustness.

[0021] Further, in step S2, designing the disturbance observer includes the following sub-steps:

[0022] Consider a MIMO nonlinear control affine system described in the continuous time domain:

[0023]

[0024] Where, x∈R n It is the system state, u∈R m It is the system input, y∈R m This is the system output, d∈R n It is the uncertainty of the model, f1:∈R n →R n G1:∈R m →R n×m G2:∈R m →R n×m and h:∈R n →R m It is a continuous function;

[0025] The nonlinear disturbance observer (NDO) is designed using formula (3) to estimate the model uncertainty and disturbance in the system, as follows:

[0026]

[0027] in, η and l represent the estimates of the perturbation, the internal state of the nonlinear observer, and the design of the observer gain. It represents the derivative of the internal state η of the nonlinear observer.

[0028] Furthermore, in step S2, modeling the wind force under gust conditions includes the following sub-steps: establishing the wind force model as follows:

[0029] The "1-cos" gust model is extended to a symmetrical oscillation described by an equation. and asymmetry Gust field:

[0030]

[0031] Where w m For the maximum gust speed, λ x , λ y For X E Y E The length of the gust in the direction.

[0032] Furthermore, in step S3, the design of a neural network-based solution for sensor error and time delay faults specifically includes the following sub-steps: training a model using historical sensor measurements and related data, and using online data to achieve online compensation for sensor error data and time delay; when a sensor malfunctions, using the neural network model to predict sensor data for a period of time to avoid complete loss of control of the aircraft due to sensor malfunction.

[0033] Furthermore, in step S4, designing the attitude loop discrete incremental sliding mode fault-tolerant controller includes the following sub-steps:

[0034] Design a discrete incremental sliding mode fault-tolerant controller for the attitude loop to enable the aircraft to robustly track roll, yaw, and pitch angle reference values. Select the control variables y = [φ, θ, ψ]. T Let x1 = [p, q, r] T x2 = [φ, θ, ψ] T , x = [x1 T x2 T ] T The nonlinear model corresponds to the rotational motion of a rigid body aircraft, and the control input is represented as u = [τ]. rx ,τ ry ,τ rz ] T This yields the following expression:

[0035]

[0036] y = x2

[0037] in,

[0038]

[0039] Considering actuator failure, the aircraft dynamics model is as follows:

[0040]

[0041] Here, κ∈[0,1] is designed as a unit step function to indicate a sudden actuator failure during flight; y=[φ,θ,ψ] T The relative degree of the vector is ρ = [2, 2, 2]. TThe output dynamics of the aircraft attitude system are:

[0042]

[0043] Therefore:

[0044]

[0045] Equation (6) includes a new variable κ, and y is calculated. (2) First-order Taylor series expansion:

[0046]

[0047] In the formula

[0048] The discrete-time incremental sliding mode control law in equation (8) is as follows:

[0049]

[0050] In the above formula, The control matrix is ​​known; Λ∈R m×m It is a diagonal matrix and an unknown time-varying control degradation matrix, where Λ = diag{w1, w2, ..., w m}; It is a virtual control used for stable, undisturbed systems, K∈R n×n And design K to make A c -B c K is the Herwitz matrix. yes The measured or estimated value, v s (k) is the designed disturbance compensation;

[0051] Design discrete sliding variable σ(k):R n →R m for

[0052] σ(k)=Se(k)-Se(0)+E I (k),E I (k)=E I (k-1)+K e e(k-1) (10)

[0053] In the formula, K e e = -hS(A c -B c K)=diag{[K i,e ,0,[,0]},S=diag{S i},

[0054] Sliding mode virtual control quantity v s(k) is designed as follows:

[0055]

[0056] In the formula, It is the sigmoid function;

[0057] Next, u = u0 + Δu indi-s Based on equation (8), the closed-loop system dynamics (7) of equation (7) under the control input equation (9) yields:

[0058]

[0059] In the formula, Φ s (k) and as follows:

[0060]

[0061]

[0062] Furthermore, in step S4, designing the position loop discrete incremental sliding mode fault-tolerant controller includes the following sub-steps:

[0063] When designing a position controller, the following variables are replaced compared to when designing an attitude controller:

[0064]

[0065] Choose the control variable as u l =[u1,u2,u3] T Choose control variable y l =[x,y,z] T ,use x l2 =[x,y,z] T The nonlinear model corresponds to the translational motion of the rigid body aircraft, and is expressed as follows:

[0066]

[0067] y l =x l2

[0068] in,

[0069] Similarly, the discrete incremental sliding mode control law for the position loop is obtained as follows:

[0070]

[0071] The design process for the parameters in the above formula is the same as that for the parameters in the attitude loop.

[0072] The beneficial effects of this invention include:

[0073] The discrete incremental sliding mode control based on the disturbance observer in this embodiment of the invention has the following advantages: (1) Enhanced robustness: The disturbance observer, incremental dynamic inversion control, and sliding mode control make the control system highly robust. Under the influence of factors such as uncertainty, parameter changes, external disturbances, and model errors, the control system still has good performance. (2) Improved fault tolerance: The discrete incremental sliding mode control method introduces a discretization strategy into the control system, which makes the controller have a certain fault tolerance capability for sensor errors, actuator failures, and time delays. When the system has a fault or abnormal situation, the discrete incremental sliding mode control method can adjust the control command in time to ensure the stability and safety of the system. (3) High-precision control: The discrete incremental sliding mode control method can improve the control accuracy of the system through a suitable discretization strategy. Through the generation and execution of precise discrete control commands, more accurate and stable quadcopter control can be achieved, improving the maneuverability and accuracy of the aircraft.

[0074] The LSTM predictor-based scheme of this invention has the following advantages: (1) Sensor error compensation: LSTM networks can learn and model the patterns and characteristics of sensor errors. By training the LSTM network, the control system can perform error compensation based on real-time sensor data, thereby improving the accuracy of state estimation and control performance. (2) Time delay compensation: LSTM networks can handle the time delay problem of sensor data. Through the ability to learn and memorize historical data, LSTM networks can predict future sensor data, thereby achieving time delay compensation. This helps to improve the real-time performance and response performance of the control system. (3) Fault detection and fault tolerance: LSTM networks can be used to monitor sensor faults and abnormal conditions. By learning the patterns of normal sensor data, LSTM networks can detect abnormal behavior of sensor output. Once a fault is detected, the control system can take corresponding fault-tolerant measures, such as using backup sensors or adaptively adjusting the control strategy, to ensure the stability and safety of the aircraft. Attached Figure Description

[0075] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0076] Figure 1 This is a schematic diagram of a miniature quadcopter.

[0077] Figure 2A schematic diagram showing the relationship between the body coordinate system and the ground coordinate system of a micro quadcopter.

[0078] Figure 3 This describes the internal structure of an LSTM network. Figure 3 The area contained within the middle background color has a specific meaning, representing a computational unit;

[0079] Figure 4 The diagram shows the control block diagram of a micro quadcopter based on the LSTM-D-DOISMC algorithm. The position and attitude controllers are discrete incremental sliding mode controllers based on disturbance observers, and the neural network predictor is based on the LSTM algorithm. Detailed Implementation

[0080] All features disclosed in all embodiments of this specification, or steps in all methods or processes implied in the disclosure, may be combined and / or extended or replaced in any way, except for mutually exclusive features and / or steps.

[0081] In light of the problems described in the background, the inventors of this invention, after creative consideration, recognized that quadrotors may be affected by various disturbances during flight, which can impact their attitude, stability, and control performance. The following are some common disturbances experienced by quadrotors: 1. External wind disturbances; 2. Aerodynamic disturbances; 3. Gravitational disturbances; 4. Load variations. To address these disturbances, appropriate control strategies and algorithms need to be designed to enable the quadrotor to maintain stable attitude and flight performance in unstable environments.

[0082] To address the aforementioned technical problems, this invention proposes a discrete incremental sliding mode quadrotor control method (self-named: LSTM-DOISMC) based on a neural network (LSTM) and a disturbance observer. In this invention, the method considers the discrete nature of the controller law during the controller design phase and fully utilizes the time-series modeling capabilities of the neural network (LSTM). By using LSTM, discrete-time prediction and modeling of the system state can be achieved, resulting in more accurate state estimates. Further, the invention employs an incremental sliding mode control method to design the controller, which enables rapid sliding of the system state on the sliding surface while maintaining stability. Compared to traditional continuous sliding mode control methods, incremental sliding mode control is more suitable for controlling discrete systems and can better handle the nonlinear and uncertain characteristics of discrete systems.

[0083] To improve the robustness of the controller and its ability to suppress external disturbances, this invention also incorporates a disturbance observer. The disturbance observer estimates and compensates for external disturbances in real time by observing and compensating for the difference between the system output and the controller output. The introduction of the disturbance observer effectively reduces the impact of disturbances on the system, thereby improving the controller's performance and robustness.

[0084] In a further inventive concept, and in practical applications, the embodiments of the present invention also consider control reversal and sudden actuator failure. Control reversal refers to the phenomenon in a control system where, when the designed controller law is not applicable to a specific situation or the system dynamics change, the controller's output causes the system to become unstable or respond in the opposite direction; this is relatively common in quadcopter control. Sudden actuator failure refers to the sudden malfunction or failure of one or more actuators (such as motors) in a quadcopter during operation. Such failures may cause sudden changes in actuator output torque or thrust, thus having unpredictable effects on the quadcopter's attitude and motion, potentially leading to flight instability or even accidents.

[0085] In summary, the LSTM-DOISMC method provided in this embodiment of the invention comprehensively considers the discreteness of the controller law, discrete incremental sliding mode control, and the design of the disturbance observer. It also takes into account the time-varying characteristics of the control effectiveness matrix, control inversion, sudden actuator failures, and the handling of sensor errors, time delays, and faults. This method enables precise control of quadcopters and improves the system's performance, robustness, and fault tolerance.

[0086] The specific implementation process includes the following steps:

[0087] (1) Establish a mathematical model of a six-degree-of-freedom quadcopter UAV, taking into account the uncertainty factors of the aircraft;

[0088] (2) Design a disturbance observer to model wind force under gust conditions, and design an online wind force predictor based on the disturbance observer and neural network (LSTM).

[0089] (3) Design a solution for sensor error, delay and fault problems based on LSTM;

[0090] (4) Design discrete incremental sliding mode fault-tolerant controllers in the attitude loop and position loop respectively.

[0091] The aircraft dynamics model in step (1) is as follows:

[0092]

[0093] Where x, y, z represent the position coordinates of the quadcopter relative to the inertial frame; θ, φ, ψ represent the Euler angles of the quadcopter, namely pitch, roll, and yaw, respectively; p, q, r represent the rotational angular velocities of the quadcopter in the body coordinate system; J xb J yb J zb The x, y, and z axes represent the inertial constants of the quadrotor, m represents the mass of the quadrotor, and f represents the inertial constants around the x, y, and z axes. r τ is the magnitude of the resultant force of the quadcopter. xr ,τ yr ,τ zr f is the resultant torque generated by the four rotors in the body coordinate system. dx ,f dy ,f dz For the interfering force, τ dx ,τ dy ,τ dz For the disturbance torque; S(·) represents sin(·), C(·) represents cos(·), This represents the acceleration of an aircraft in an inertial frame of reference. This represents the angular acceleration of the quadcopter in the body coordinate system.

[0094] The expressions for the force and torque generated by the rotor are:

[0095]

[0096] in, Ω1, Ω2, Ω3, Ω4 represent the rotational speeds of rotors 1, 2, 3, and 4, respectively; k and b are the thrust and torque constants of the quadcopter; and L is the length of the quadcopter's arm. T This represents the transpose of a matrix.

[0097] The disturbance observer in step (2) is

[0098] Consider a MIMO nonlinear control affine system described in the continuous time domain:

[0099]

[0100] Where, x∈R n It is the system state, u∈R m It is the system input, y∈R m This is the system output, d∈R n It is the uncertainty of the model, f1:∈R n →R n G1:∈R m →R n×m G2:∈R m →R n×mand h:∈R n →R m It is a continuous function.

[0101] System (3) designs a nonlinear disturbance observer (NDO) to estimate the uncertainties and disturbances in the model of the system, as described below:

[0102]

[0103] in, η and l represent the estimates of the perturbation, the internal state of the nonlinear observer, and the design of the observer gain. It represents the derivative of the internal state η of the nonlinear observer.

[0104] The wind model in step (2) is as follows:

[0105] The “1-cos” gust model can be expanded into a symmetrical oscillation described by an equation. and asymmetry Gusts of wind.

[0106]

[0107] Where w m For the maximum gust speed, λ x , λ y For X E Y E The length of the gust in the direction.

[0108] The internal structure of the LSTM network in step (2) is as follows: Figure 3 By training a model using historical wind data and relevant parameters, and combining it with a disturbance observer, accurate online prediction of convective wind force can be achieved. The network structure and training algorithm of the predictor are optimized to improve prediction performance and robustness.

[0109] The internal structure of the LSTM network in step (3) is as follows: Figure 3 The model is trained using historical sensor measurements and related data, and online compensation for sensor error data and time delay is achieved based on online data. When a sensor fails, the neural network model can predict sensor data for a period of time, thus avoiding complete loss of control of the aircraft due to sensor failure.

[0110] The attitude loop discrete incremental sliding mode fault-tolerant controller designed in step (4) is as follows:

[0111] A discrete incremental sliding mode fault-tolerant attitude loop controller was designed to enable the aircraft to robustly track roll, yaw, and pitch angle reference values. Therefore, the control variables y = [φ, θ, ψ] were chosen. T Let x1 = [p, q, r] Tx2 = [φ, θ, ψ] T , x = [x1 T x2 T ] T The nonlinear model corresponds to the rotational motion of a rigid body aircraft, and the control input is represented as u = [τ]. rx ,τ ry ,τ rz ] T We can obtain the following expression:

[0112]

[0113] y = x2

[0114] in,

[0115]

[0116] Considering actuator failure, the aircraft dynamics can be modeled as follows:

[0117]

[0118] Here, κ∈[0,1] is designed as a unit step function to indicate a sudden actuator failure during flight.

[0119] y = [φ, θ, ψ] T The relative degree of the vector is ρ = [2, 2, 2]. T Therefore, the output dynamics of the aircraft attitude system is:

[0120]

[0121] Therefore, we can conclude that:

[0122]

[0123] Equation (6) includes a new variable κ, and y is calculated. (2) First-order Taylor series expansion:

[0124]

[0125] In the formula

[0126] The discrete-time incremental sliding mode control law in system (8) is as follows:

[0127]

[0128] In the above formula, The control matrix is ​​known; Λ∈R m×mIt is a diagonal matrix and an unknown time-varying control degradation matrix, where Λ = diag{w1, w2, ..., w m} It is a virtual control used for stable, undisturbed systems, K∈R n×n And design K to make A c -B c K is the Herwitz matrix. yes The measured or estimated value, v s (k) is the disturbance compensation designed.

[0129] Design discrete sliding variable σ(k):R n →R m for

[0130] σ(k)=Se(k)-Se(0)+E I (k),E I (k)=E I (k-1)+K e e(k-1)(10)

[0131] In the formula, K e e = -hS(A c -B c K)=diag{[K i,e ,0,...,0]},S=diag{S i},

[0132] Therefore, the sliding mode virtual control quantity v s (k) Designed as

[0133]

[0134] In the formula, This is the sigmoid function.

[0135] Next, u = u0 + Δu indi-s Based on equation (8), the closed-loop system dynamics (7) of equation (7) under the control input equation (9) yields:

[0136]

[0137] In the formula, Φ s (k) and As shown below:

[0138]

[0139]

[0140] The position loop discrete incremental sliding mode fault-tolerant controller designed in step (4) is as follows:

[0141] The design of the position controller is similar to that of the attitude controller, with the following variable substitutions.

[0142]

[0143] Choose the control variable as u l =[u1,u2,u3] T Choose control variable y l =[x,y,z] T ,use x l2 =[x,y,z] T The nonlinear model corresponds to the translational motion of the rigid body aircraft, and can be expressed as follows:

[0144]

[0145] y l =x l2

[0146] in,

[0147] Similarly, the discrete incremental sliding mode control law for the position loop can be obtained as follows:

[0148]

[0149] Note: The design process of the parameters in the above formula is as follows: attitude loop.

[0150] It should be noted that, within the scope of protection defined in the claims of this invention, the following embodiments can be combined and / or extended or replaced in any logical manner from the above specific embodiments, such as the disclosed technical principles, disclosed technical features or implicitly disclosed technical features.

[0151] Example 1

[0152] A discrete incremental sliding mode quadrotor control method based on neural networks and disturbance observation includes the following steps:

[0153] S1. Considering the uncertainties of the aircraft, establish a six-degree-of-freedom quadcopter UAV model;

[0154] S2, Design a disturbance observer and model wind force under gust conditions, and design an online wind force predictor based on the disturbance observer and neural network;

[0155] S3, Design a solution for sensor error and time delay fault problems based on neural networks;

[0156] S4, a discrete incremental sliding mode fault-tolerant controller is designed in the attitude loop and the position loop respectively.

[0157] Example 2

[0158] Based on Example 1, the establishment of a six-degree-of-freedom quadcopter UAV model considering aircraft uncertainties specifically includes the following sub-steps:

[0159] Establish the following aircraft dynamics model:

[0160]

[0161] Where x, y, z represent the position coordinates of the quadcopter relative to the inertial frame; θ, φ, ψ represent the Euler angles of the quadcopter, namely pitch, roll, and yaw, respectively; p, q, r represent the rotational angular velocities of the quadcopter in the body coordinate system; J xb J yb J zb The x, y, and z axes represent the inertial constants of the quadrotor, m represents the mass of the quadrotor, and f represents the inertial constants around the x, y, and z axes. r τ is the magnitude of the resultant force of the quadcopter. xr ,τ yr ,τ zr f is the resultant torque generated by the four rotors in the body coordinate system. dx ,f dy ,f dz For the interfering force, τ dx ,τ dy ,τ dz For the disturbance torque; S(·) represents sin(·), C(·) represents cos(·), This represents the acceleration of an aircraft in an inertial frame of reference. This represents the angular acceleration of the quadcopter in the body coordinate system.

[0162] The expressions for the force and torque generated by the rotor are:

[0163]

[0164] in, Ω1, Ω2, Ω3, Ω4 represent the rotational speeds of rotors 1, 2, 3, and 4, respectively; k and b are the thrust and torque constants of the quadcopter; and L is the length of the quadcopter's arm. T This represents the transpose of a matrix.

[0165] Example 3

[0166] Based on Example 1, in step S2, the neural network is an LSTM network.

[0167] Example 4

[0168] Based on Example 1, in step S2, the model is trained using historical wind data and related parameters, and combined with a disturbance observer, to achieve accurate online prediction of gust wind force.

[0169] Example 5

[0170] Based on Example 1, step S2 includes the sub-step of optimizing the predictor's network structure and training algorithm to improve prediction performance and robustness.

[0171] Example 6

[0172] Based on Example 1, step S2, designing the disturbance observer includes the following sub-steps:

[0173] Consider a MIMO nonlinear control affine system described in the continuous time domain:

[0174]

[0175] Where, x∈R n It is the system state, u∈R m It is the system input, y∈R m This is the system output, d∈R n It is the uncertainty of the model, f1:∈R n →R n G1:∈R m →R n×m G2:∈R m →R n×m and h:∈R n →R m It is a continuous function;

[0176] The nonlinear disturbance observer (NDO) is designed using formula (3) to estimate the model uncertainty and disturbance in the system, as follows:

[0177]

[0178] in, η and l represent the estimates of the perturbation, the internal state of the nonlinear observer, and the design of the observer gain. It represents the derivative of the internal state η of the nonlinear observer.

[0179] Example 7

[0180] Based on Example 1, in step S2, the modeling of wind force under gust conditions includes the following sub-steps: establishing the wind force model as follows:

[0181] The "1-cos" gust model is extended to a symmetrical oscillation described by an equation. and asymmetry Gust field:

[0182]

[0183] Where w m For the maximum gust speed, λ x , λ y For X E Y E The length of the gust in the direction.

[0184] Example 8

[0185] Based on Example 1, in step S3, the design of a neural network-based solution to sensor error and time delay problems specifically includes the following sub-steps: training a model using historical sensor measurements and related data, and using online data to achieve online compensation for sensor error data and time delay; when a sensor malfunctions, using the neural network model to predict sensor data for a period of time to avoid complete loss of control of the aircraft due to sensor malfunction.

[0186] Example 9

[0187] Based on Example 1, in step S4, designing the attitude loop discrete incremental sliding mode fault-tolerant controller includes the following sub-steps:

[0188] Design a discrete incremental sliding mode fault-tolerant controller for the attitude loop to enable the aircraft to robustly track roll, yaw, and pitch angle reference values. Select the control variables y = [φ, θ, ψ]. T Let x1 = [p, q, r] T x2 = [φ, θ, ψ] T , x = [x1 T x2 T ] T The nonlinear model corresponds to the rotational motion of a rigid body aircraft, and the control input is represented as u = [τ]. rx ,τ ry ,τ rz ] T This yields the following expression:

[0189]

[0190] y = x2

[0191] in,

[0192]

[0193] Considering actuator failure, the aircraft dynamics model is as follows:

[0194]

[0195] Here, κ∈[0,1] is designed as a unit step function to indicate a sudden actuator failure during flight; y=[φ,θ,ψ] T The relative degree of the vector is ρ = [2, 2, 2]. T The output dynamics of the aircraft attitude system are:

[0196]

[0197] Therefore:

[0198]

[0199] Equation (6) includes a new variable κ, and y is calculated. (2) First-order Taylor series expansion:

[0200]

[0201] In the formula

[0202] The discrete-time incremental sliding mode control law in equation (8) is as follows:

[0203]

[0204] In the above formula, The control matrix is ​​known; Λ∈R m×m It is a diagonal matrix and an unknown time-varying control degradation matrix, where Λ = diag{w1, w2, ..., w m}; It is a virtual control used for stable, undisturbed systems, K∈R n×n And design K to make A c -B c K is the Herwitz matrix. yes The measured or estimated value, v s (k) is the designed disturbance compensation;

[0205] Design discrete sliding variable σ(k):R n →R m for

[0206] σ(k)=Se(k)-Se(0)+E I (k),E I (k)=E I (k-1)+K e e(k-1) (10)

[0207] In the formula, K e e = -hS(A c -B c K)=diag{[K i,e ,0,...,0]},S=diag{S i},

[0208] Sliding mode virtual control quantity v s (k) is designed as follows:

[0209]

[0210] In the formula, It is the sigmoid function;

[0211] Next, u = u0 + Δu indi-s Based on equation (8), the closed-loop system dynamics (7) of equation (7) under the control input equation (9) yields:

[0212]

[0213] In the formula, Φ s (k) and as follows:

[0214]

[0215]

[0216] Example 10

[0217] Based on Example 1, in step S4, designing the position loop discrete incremental sliding mode fault-tolerant controller includes the following sub-steps:

[0218] When designing a position controller, the following variables are replaced compared to when designing an attitude controller:

[0219]

[0220] Choose the control variable as u l =[u1,u2,u3] T Choose control variable y l =[x,y,z] T ,use x l2 =[x,y,z] T The nonlinear model corresponds to the translational motion of the rigid body aircraft, and is expressed as follows:

[0221]

[0222] yl =x l2

[0223] in,

[0224] Similarly, the discrete incremental sliding mode control law for the position loop is obtained as follows:

[0225]

[0226] The design process for the parameters in the above formula is the same as that for the parameters in the attitude loop.

[0227] The units described in the embodiments of the present invention can be implemented in software or hardware, and the described units can also be located in a processor. The names of these units do not necessarily limit the specific unit itself.

[0228] According to one aspect of the present invention, a computer program product or computer program is provided, the computer program product or computer program including computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium, and executes the computer instructions, causing the computer device to perform the methods provided in the various optional implementations described above.

[0229] In another aspect, embodiments of the present invention also provide a computer-readable medium, which may be included in the electronic device described in the above embodiments; or it may exist independently and not assembled into the electronic device. The computer-readable medium carries one or more programs, which, when executed by the electronic device, cause the electronic device to perform the methods described in the above embodiments.

Claims

1. A discrete incremental sliding mode quadrotor control method based on neural networks and disturbance observation, characterized in that, Includes the following steps: S1. Considering the uncertainties of the aircraft, establish a six-degree-of-freedom quadcopter UAV model; S2, design a disturbance observer and model the wind force under gust conditions, and design an online wind force predictor based on the disturbance observer and neural network; in step S2, the modeling of wind force under gust conditions includes the following sub-steps: establishing the wind force model as follows: Extending the "1-cos" gust model to the symmetric oscillation described by the equation and the asymmetric gust field: wherein is the maximum gust speed, , is , the length of the gust in the direction S3, Design a solution for sensor error and time delay fault problems based on neural networks; S4, discrete incremental sliding mode fault-tolerant controllers are designed in the attitude loop and position loop respectively; among which... The following discrete-time incremental sliding mode control law is used in the design of the attitude loop discrete incremental sliding mode fault-tolerant controller: ； In the formula, It consists of a diagonal matrix in the attitude loop and an unknown time-varying control degradation matrix. Given the known control matrix in the attitude loop. As variables, It is a virtual control in the attitude loop used to stabilize an undisturbed system. It is a disturbance compensation designed in the attitude loop. These are measured or estimated values ​​in the attitude loop; The following discrete incremental sliding mode fault-tolerant controller is used in the position loop design: wherein is a diagonal matrix and an unknown time-varying control degradation matrix in the attitude loop, is a known control matrix in the attitude loop, is a variable, is a virtual control in the attitude loop for stabilizing the undisturbed system, is a disturbance compensation designed in the attitude loop, is a measurement or an estimate in the attitude loop.

2. The discrete incremental sliding mode quadrotor control method based on neural networks and disturbance observation according to claim 1, characterized in that, The process of establishing a six-degree-of-freedom quadcopter UAV model, taking into account the uncertainties of the aircraft, specifically includes the following sub-steps: The following aircraft dynamics model is established: (1) in, Represents the position coordinates of the quadcopter relative to the inertial frame; The Euler angles representing quadcopter aircraft are pitch angle, roll angle, and yaw angle. This represents the angular velocity of rotation of a quadcopter in the body coordinate system; Represents a quadcopter axis, axis, Axial inertia constant, Indicates the mass of a quadcopter. The magnitude of the resultant force of the quadcopter. The resultant torque generated by the four rotors in the airframe coordinate system. For interference force, For disturbance torque; express , express , This represents the acceleration of an aircraft in an inertial frame of reference. This represents the angular acceleration of the quadcopter in the body coordinate system; The expressions for the force and torque generated by the rotor are: (2) in, , These represent the rotational speeds of rotors 1, 2, 3, and 4. Let be the thrust and torque constants of the quadcopter. It is the length of the arm of a quadcopter. This represents the transpose of a matrix.

3. The discrete incremental sliding mode quadrotor control method based on neural networks and disturbance observation according to claim 1, characterized in that, In step S2, the neural network is an LSTM network.

4. The discrete incremental sliding mode quadrotor control method based on neural networks and disturbance observation according to claim 1, characterized in that, In step S2, a model is trained using historical wind data and relevant parameters, and combined with a disturbance observer, to achieve accurate online prediction of convective wind force.

5. The discrete incremental sliding mode quadrotor control method based on neural networks and disturbance observation according to claim 1, characterized in that, Step S2 includes the sub-step of optimizing the predictor's network structure and training algorithm to improve prediction performance and robustness.

6. The discrete incremental sliding mode quadrotor control method based on neural networks and disturbance observation according to claim 1, characterized in that, In step S2, designing the disturbance observer includes the following sub-steps: Consider a MIMO nonlinear control affine system described in the continuous time domain: (3) in, It is the system status. It is system input. It is system output. It is the uncertainty of the model. , , as well as It is a continuous function; The nonlinear disturbance observer (NDO) is designed using formula (3) to estimate the model uncertainty and disturbance in the system, as follows: (4) in, , , These include estimating the perturbation, the internal state of the nonlinear observer, and designing the observer gain. Represents the internal state of the nonlinear observer The derivative of .

7. The discrete incremental sliding mode quadrotor control method based on neural networks and disturbance observation according to claim 1, characterized in that, In step S3, the design of a neural network-based solution to sensor error and time delay problems specifically includes the following sub-steps: training a model using historical sensor measurements and related data, and using online data to achieve online compensation for sensor error data and time delay; when a sensor malfunctions, using the neural network model to predict sensor data for a period of time to avoid complete loss of control of the aircraft due to sensor malfunction.

8. The discrete incremental sliding mode quadrotor control method based on neural networks and disturbance observation according to claim 1, characterized in that, In step S4, designing the attitude loop discrete incremental sliding mode fault-tolerant controller includes the following sub-steps: Design an attitude loop discrete incremental sliding mode fault-tolerant controller to enable the aircraft to robustly track roll, yaw, and pitch angle reference values, and select control variables. ,use , , The nonlinear model corresponds to the rotational motion of a rigid body aircraft, and the control input is represented as... This yields the following expression: (5) in, Considering actuator failure, the aircraft dynamics model is as follows: (6) in, It is designed as a unit step function to indicate a sudden actuator failure during flight; The relative degree of the vector is The output dynamics of the aircraft attitude system are: Therefore: (7) Equation (6) includes a new variable. ,calculate First-order Taylor series expansion: (8) In the formula , The discrete-time incremental sliding mode control law in equation (8) is as follows: (9) In the above formula, The control matrix is ​​known. It is a diagonal matrix and an unknown time-varying control degradation matrix, while ; It is a virtual control system used for stabilizing undisturbed systems. and design make It is the Herwitz matrix. yes The measured or estimated value, It is a disturbance compensation designed in; Design discrete sliding variable for (10) In the formula, , , Sliding Module Virtual Control Quantity Designed as follows: In the formula, It is the sigmoid function; then, Based on equation (8), the closed-loop system dynamics (7) of equation (7) under the control input equation (9) yields: In the formula, and as follows: 。 9. The discrete incremental sliding mode quadrotor control method based on neural networks and disturbance observation according to claim 1, characterized in that, In step S4, designing the position loop discrete incremental sliding mode fault-tolerant controller includes the following sub-steps: When designing a position controller, the following variables are replaced compared to when designing an attitude controller: Select control variables as Select control variables ,use , The nonlinear model corresponds to the translational motion of the rigid body aircraft, and is expressed as follows: in, ; Similarly, the discrete incremental sliding mode control law for the position loop is obtained as follows: The design process for the parameters in the above formula is the same as that for the parameters in the attitude loop.

Images