Unmanned aerial vehicle sliding mode attitude control method and system based on neural network

Abstract

The present disclosure relates to the technical field of unmanned aerial vehicle adaptive attitude control, and proposes a neural network-based unmanned aerial vehicle sliding mode attitude control method and system. By simplifying the virtual control input and attitude dynamics equation of a quadrotor aircraft, an unmanned aerial vehicle motion attitude model equation is obtained. A sliding surface is set, and an adaptive approaching law is set for each attitude angle to control the unmanned aerial vehicle to approach the sliding surface from the current state. A control law is set based on the adaptive approaching law of each attitude angle. The variable parameter term of the control law is dynamically adjusted based on the gradient descent method of the neural network. Based on the obtained dynamic variable parameter term and the control law, the attitude of the unmanned aerial vehicle is controlled. The present disclosure can control the attitude angle of the quadrotor unmanned aerial vehicle in a complex environment such as a wind field by using the neural network method combined with the improved adaptive approaching law, and can solve the robustness and anti-interference problems of the attitude control of the rotor unmanned aerial vehicle.

Description

Technical Field

[0001] This disclosure relates to the technical field of adaptive attitude control for rotary-wing unmanned aerial vehicles (UAVs), specifically, to a sliding mode attitude control method and system for UAVs based on neural networks. Background Technology

[0002] The statements in this section are merely background information relating to this disclosure and do not necessarily constitute prior art.

[0003] With the development of science and technology, drones have emerged to address the high-difficulty and high-risk tasks that are beyond the capabilities of humans, replacing manned aircraft in these missions. Rotary-wing drones, in particular, capable of carrying various types of gimbals for flight missions, have played a significant role in various fields. For example, in the civilian sector, multi-rotor drones are used for aerial photography; in the industrial sector, they are used for power line inspection, construction site supervision, and crop health monitoring; and in the military sector, drones are used for information reconnaissance, fire support, and material transport.

[0004] Quadrotor drones are a typical type of rotorcraft drone. In recent years, due to their simplicity, maneuverability, and vertical takeoff and landing capabilities, they have been widely used in military and civilian fields. Flight control of drones has always been a research hotspot in the drone field. However, the dynamic system of quadrotor drones is characterized by nonlinearity, underactuation, strong coupling, and multiple inputs and multiple outputs. Furthermore, their dynamic models are relatively complex and susceptible to external disturbances. In complex wind fields, system parameters are difficult to measure accurately, posing a significant challenge to controller design. Attitude control, as a crucial component, further demands that the controller design possess excellent dynamic response and robustness, placing even higher requirements on drone controller design.

[0005] The inventors discovered in their research that PID control is currently the most common method in flight control. However, PID control is easily affected by external environmental interference, leading to poor control performance. Other control methods used for quadrotor UAVs, such as LQR (Linear Quadratic Regulator), Model Reference Adaptive Control, and Backstepping, also have their own problems, such as integral explosion, high-frequency oscillation, limited anti-interference ability, and limited system robustness. Summary of the Invention

[0006] To address the aforementioned issues, this disclosure proposes a sliding mode attitude control method and system for unmanned aerial vehicles (UAVs) based on neural networks. By combining the neural network method with an improved adaptive approach law, the attitude angle control of quadrotor UAVs can be achieved in complex environments such as wind fields, thus solving the robustness and anti-interference problems of attitude control for rotary-wing UAVs.

[0007] To achieve the above objectives, the present disclosure adopts the following technical solution:

[0008] One or more embodiments provide a neural network-based sliding mode attitude control method for unmanned aerial vehicles (UAVs), comprising the following steps:

[0009] By simplifying the virtual control input and attitude dynamics equations of the quadrotor aircraft, the motion attitude model equations of the UAV are obtained.

[0010] Set a sliding surface, and for each attitude angle, set an adaptive approach law to control the UAV to approach the sliding surface from the current state. Set a control law based on the adaptive approach law for each attitude angle.

[0011] The gradient descent method based on neural networks dynamically adjusts the variable parameter terms of the control law;

[0012] Based on the obtained dynamic variable parameter terms and control law, the input values ​​of each control channel are calculated through the simplified UAV attitude model equations to achieve UAV attitude control.

[0013] One or more embodiments provide a neural network-based sliding mode attitude control system for unmanned aerial vehicles, including:

[0014] The model equation building module is configured to simplify the virtual control input and attitude dynamics equations of the quadcopter to obtain the motion attitude model equations of the UAV.

[0015] The adaptive approach law and control law determination module is configured to set the sliding surface, set an adaptive approach law for each attitude angle to control the UAV to approach the sliding surface from the current state, and set a control law based on the adaptive approach law for each attitude angle.

[0016] Variable parameter update module: configured to dynamically adjust the variable parameter terms of the control law based on gradient descent using a neural network;

[0017] Control output value determination module: It is configured to calculate the input values ​​of each control channel based on the obtained dynamic variable parameter terms and control law, and realize the control of UAV attitude through the simplified UAV attitude model equation.

[0018] An electronic device includes a memory and a processor, as well as computer instructions stored in the memory and running on the processor, wherein the computer instructions, when executed by the processor, perform the steps described in the above method.

[0019] A computer-readable storage medium for storing computer instructions, which, when executed by a processor, perform the steps described in the above method.

[0020] Compared with the prior art, the beneficial effects of this disclosure are as follows:

[0021] This invention enables the dynamic adjustment of variable parameters in the control law to achieve the optimal state in complex environments such as wind fields, based on a neural network, and combines this with a sliding mode adaptive law to adjust the control input. It effectively addresses chattering issues in sliding mode control, offers higher response speed and robustness for attitude control, provides greater resistance to interference in windy environments, and significantly improves the stability of quadcopter UAVs.

[0022] The advantages of this disclosure, as well as its additional advantages, will be described in detail in the following specific embodiments. Attached Figure Description

[0023] The accompanying drawings, which form part of this disclosure, are used to provide a further understanding of this disclosure. The illustrative embodiments of this disclosure and their descriptions are used to explain this disclosure and do not constitute a limitation thereof.

[0024] Figure 1 This is a model diagram of a quadcopter drone according to Embodiment 1 of this disclosure;

[0025] Figure 2 This is the tracking response of the input step signal in the simulation example of Embodiment 1 of this disclosure;

[0026] Figure 3 This is the tracking response of the input sinusoidal signal in the simulation example of Embodiment 1 of this disclosure;

[0027] Figure 4 This is the wind field turbulence interference signal in the simulation example of Embodiment 1 of this disclosure;

[0028] Figure 5 This is a sinusoidal response under wind field disturbance in the simulation example of Embodiment 1 of this disclosure;

[0029] Figure 6 A comparison of the responses to the input quantity u using different methods in the simulation example of Embodiment 1 of this disclosure;

[0030] Figure 7 This is a flowchart of the control method of Embodiment 1 of this disclosure. Detailed Implementation

[0031] The present disclosure will be further described below with reference to the accompanying drawings and embodiments.

[0032] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of this disclosure. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure pertains.

[0033] It should be noted that the terminology used herein is for descriptive purposes only and is not intended to limit the exemplary embodiments according to this disclosure. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof. It should be noted that, without conflict, the various embodiments and features within those embodiments can be combined with each other. The embodiments will now be described in detail with reference to the accompanying drawings.

[0034] Sliding mode control, by designing a sliding surface and corresponding reaching law, ensures that the sliding trajectory starting from any state reaches the designed sliding surface, exhibiting good robustness. However, traditional sliding mode control also suffers from control variable jitter. Neural networks, with their powerful self-learning capabilities, are increasingly being applied as a control method by dynamically adjusting control law parameters to achieve better results. This disclosure improves the adaptive reaching law and integrates gradient descent with a neural network, effectively addressing the jitter problem in sliding mode control, providing higher response speed and robustness for attitude control, enhancing anti-interference capabilities in windy environments, and significantly improving the stability of quadcopter UAVs. Specific embodiments are described below.

[0035] Example 1

[0036] In one or more of the technical solutions disclosed in the embodiments, such as Figures 1 to 7 As shown, a sliding mode attitude control method for unmanned aerial vehicles based on neural networks includes the following steps:

[0037] Step 1: Simplify the virtual control input and attitude dynamics equations of the quadcopter to obtain the motion attitude model equations of the UAV;

[0038] Step 2: Set the corresponding adaptive approach law for each attitude angle, and set the control law based on the adaptive approach law for each attitude angle;

[0039] Step 3: Dynamically adjust the variable parameter terms of the control law using gradient descent based on neural networks;

[0040] Step 4: Based on the obtained dynamic variable parameter terms and control law, calculate the input values ​​of each control channel through the simplified UAV attitude model equation to achieve UAV attitude control.

[0041] In this embodiment, the present disclosure can effectively handle the chattering problem in sliding mode control by improving the adaptive approach law; the gradient descent method based on neural network dynamically adjusts the variable parameter terms of the control law, realizing rapid dynamic adjustment of parameters, which has higher response speed and robustness for attitude control, higher anti-interference ability in wind field environment, and greatly improves the stability of quadcopter UAV.

[0042] In step 1, a force analysis is performed on the quadcopter UAV. Under reasonable assumptions, a simplified UAV motion attitude model equation is established by combining the coordinate system rotation matrix.

[0043] Optional, the following assumptions are made:

[0044] (1) The structure of the UAV is a rigid body, and the effect of the elastic deformation of the body on the model is not considered.

[0045] (2) The UAV has uniform mass, symmetrical structure, and the geometric origin of the body coincides with the center of mass.

[0046] This embodiment makes reasonable assumptions to eliminate the influence of other factors and to model the stress analysis of the UAV.

[0047] Taking the common "X"-shaped quadcopter drone as an example, such as Figure 1 The model diagram shown illustrates the force analysis performed on the UAV, where [F1, F2, F3, F4] are defined. T The virtual control input and attitude dynamics equations of the quadrotor aircraft, based on Newtonian mechanics and the Newton-Lagrange equations, are expressed as follows:

[0048]

[0049]

[0050] Where U1 represents the input value for the roll channel, U2 represents the input value for the pitch channel, and U3 represents the input value for the yaw channel. Where b represents the reverse torque coefficient of the propeller motor, and k... i (i = 1, 2, 3) represent the air resistance coefficients for rotational motion around the corresponding coordinate axes in the carrier coordinate system. L represents the distance between the arm and the center of mass of the UAV at its furthest point. J r This represents the moment of inertia of the rotor as it rotates about its axis. P = [φ, θ, ψ] T Let θ represent the attitude angular components of the UAV in the inertial coordinate system, where θ is the pitch angle, φ is the roll angle, and ψ is the yaw angle. and These are the angular velocities, denoted as pitch angle θ, roll angle φ, and yaw angle ψ, respectively. and Let θ, φ, and ψ represent the angular accelerations of the pitch angle, roll angle, and yaw angle, respectively; V = [p, q, r] T Let I be the angular velocity component of the UAV in the carrier coordinate system. x ,I y ,I z d1, d2, and d3 are the moments of inertia along the x, y, and z axes, respectively. The uncertain disturbance terms are d1, d2, and d3.

[0051] Ω = -Ω1 - Ω2 + Ω3 + Ω4

[0052] Among them, Ω1 to Ω4 are the rotational speeds of the four rotors.

[0053] The differential equations of the kinematic attitude model of a quadrotor can be expressed as the following state-space expression:

[0054]

[0055]

[0056] Where x represents the state variable, x1 represents the attitude angle, x2 represents its first derivative, F(x) is the uncertainty related to the state variable, including unknown parameters and some coupled uncertainties, and D(t) represents some possible time-varying uncertainties and random disturbances experienced by the UAV during flight. It is assumed that the time-varying uncertain disturbance D(t) and its derivative are bounded at any given time, i.e.:

[0057]

[0058] Where Δ represents the L of D(t) ∞ Norm.

[0059] Since the motion equations for attitude angles are similar, we will now design an adaptive sliding mode controller based on a neural network, taking the pitch angle θ as an example.

[0060] First, the sliding surface is set as: a multiple of μ of the difference between the ideal input attitude angle and the current attitude angle, and the sum of the difference between the derivative of the ideal input attitude angle and the derivative of the current attitude angle.

[0061] Taking the pitch angle θ as an example, the following slide surface is designed:

[0062]

[0063] Where μ is a positive constant to be designed, and θ d This is the ideal input value for the pitch angle.

[0064] From formula (5), the derivative of the sliding surface can be obtained as:

[0065]

[0066] When designing the controller, the nonlinear and coupling elements in the attitude model equations are simplified as disturbance terms. These disturbance terms are ignored in the controller design because they are not dynamic terms in the model. Substituting the pitch angle equation into formula (6) yields:

[0067]

[0068] To mitigate chattering in sliding mode control, an improvement is made to the traditional reaching law by employing an adaptive reaching law to design the controller. The adaptive reaching law in step 2 is constructed using a hyperbolic tangent function, with the addition of natural number coefficient terms. The adaptive reaching law is as follows:

[0069]

[0070] Where δ is the positive number to be designed, α is the adaptive gain coefficient greater than zero, and e is the base of the natural logarithm, which is a constant in mathematics.

[0071] The adaptive reaching process of the above adaptive reaching law is as follows: when θ is large, the right-hand side term is small, and the reaching law is determined by δs. 2 tanh(s) dominates, causing the attitude angle to converge rapidly to the sliding surface. When θ approaches zero, δs... 2 The tanh(s) exponent decreases, due to the natural number coefficient term. leading, Equivalent to The approach speed is also decreasing, thus achieving self-adaptation.

[0072] Traditional sliding mode control uses a sign function, while this embodiment uses a hyperbolic tangent function to avoid signal jumps. Combined with the natural number terms in the fraction, it can achieve adaptive reaching law and solve the chattering problem in sliding mode control.

[0073] Based on the control law designed according to the sliding mode adaptive reaching law, substitute Equation 7 into Equation 8 as follows:

[0074]

[0075] The adaptive approach law and control law for other attitude angles follow the same process as above, and will not be repeated here.

[0076] For the roll angle v, the designed adaptive approach law and control law are as follows:

[0077]

[0078] Where s is the sliding surface, v d This is the ideal input value for the roll angle;

[0079] For the yaw angle ψ, the designed adaptive approach law and control law are as follows:

[0080]

[0081] Where s is the sliding surface, ψ d The ideal input value for the yaw angle;

[0082] In summary, ω can represent the attitude angle, which can be the pitch angle θ, roll angle φ, or yaw angle ψ, n is a positive integer (1, 2, 3), m is the coefficient to be adjusted (1 or 1 / L), and p is the letter of the axis to be selected (x, y, z). The adaptive approach law and control law are then determined.

[0083] The adaptive reaching law is:

[0084]

[0085] Where s is the sliding surface, δ is the positive number to be designed, α is the adaptive gain greater than zero, and ω represents the attitude angle.

[0086] The control law is:

[0087]

[0088] Where m is the coefficient to be adjusted, p is the axis to be selected (x-axis, y-axis, z-axis), and I x I y and I z These represent the moments of inertia along the x, y, and z axes, respectively; ω represents the attitude angle; and μ is a constant to be designed. d is the ideal input value for the attitude angle, L represents the distance between the arm and the center of mass of the UAV at its farthest point, s is the sliding surface, δ is the positive number to be designed, and α is the adaptive gain greater than zero.

[0089] In step 3, for the set control law, the variable parameters μ and δ of the control law are dynamically adjusted using the gradient descent method of a neural network, including the following steps:

[0090] Step 31: Construct the performance error function based on the perfect squared difference between the ideal input value and the current value of the attitude angle;

[0091] Taking the pitch angle θ as an example, the performance error function is constructed as follows:

[0092]

[0093] Step 32: Integrate the partial derivatives of the performance error function based on the partial derivative chain rule, and use the convergence speed of the neural network as the variable parameter change to construct the adaptation equation.

[0094] Taking the pitch angle θ as an example, considering the gradient descent method, we have the following fitness equation:

[0095]

[0096] Where ρ is the learning rate of the neural network convergence speed, and μ0 and δ0 are the initial values ​​of μ and δ, respectively.

[0097] According to the partial derivative chain rule, we can obtain:

[0098]

[0099]

[0100] Step 33: Use the Nabla operator to transform the fitness equation to obtain the formula for calculating the variable parameters.

[0101] For formulas (10) and (11), assume that the sign function of the ratio of the difference between the attitude angle and the difference between the ideal control quantity is the partial derivative of the attitude angle with respect to the ideal control quantity, as follows:

[0102]

[0103] in For the Nabla operator, satisfying:

[0104]

[0105] The expressions for μ and δ can then be obtained as follows:

[0106]

[0107]

[0108] Step 34: Update the gradient according to the set neural network, calculate the variable parameter terms corresponding to the gradient based on the obtained variable parameter calculation formula, and calculate the control output value based on the variable parameter terms to perform attitude control on the UAV. In this step, the parameters are obtained through direct integration and continuously adjusted to achieve optimal parameters for UAV control.

[0109] This embodiment uses neural network-based parameter recognition, which is robust, has high anti-interference ability, and is more resistant to interference in environments such as wind fields.

[0110] Further technical solutions involve stability analysis of the controller designed using the neural network-based adaptive sliding mode control method, as detailed below:

[0111] To prove the stability of the closed-loop system, the following Lyapunov function is chosen:

[0112]

[0113] Taking the derivative of the function, we get:

[0114]

[0115] Let t(s) = s 3 tanh(s);

[0116]

[0117] When s > 0 When s < 0 Therefore, t(s)≥0;

[0118] Let r(s) = s tanh(s);

[0119]

[0120] Similarly, when s > 0, When s < 0 r(s) min =r(0)=0, therefore r(s)≥0;

[0121]

[0122] According to Lyapunov's stability theorem, we know that... Since it is negative definite, the system can be guaranteed to be stable when equation (20) is satisfied.

[0123] The control method in this embodiment can achieve the optimal state by dynamically adjusting the variable parameters of the control law according to the neural network in complex environments such as wind fields, and adjust the control input by combining the sliding mode adaptive law. It effectively handles the chattering problem in sliding mode control, has higher response speed and robustness for attitude control, has higher anti-interference ability in wind field environments, and significantly improves the stability of quadcopter UAVs.

[0124] To illustrate the beneficial effects of the control method in this embodiment, a controller was built using the Matlab / Simulink simulation platform based on the control method of this embodiment, and simulations and analyses were performed on the established attitude motion model. Simulation experiments were conducted from two aspects: dynamic response and anti-interference performance. Simultaneously, experiments were conducted to investigate the chattering phenomenon that occurs in sliding mode control, to verify the actual effect of eliminating chattering.

[0125] Simulation comparison experiments were conducted with this controller, a PID controller, and a traditional sliding mode controller under step and sinusoidal input signals, respectively. The dynamic response simulation comparison results under different input conditions are as follows: Figures 2-3As shown, the next step is to verify the controller's anti-interference capability and robustness. A sinusoidal signal and a wind field turbulence interference signal are selected for simulation. The simulated wind speed of the wind field turbulence is shown in the figure. Figure 4 As shown, the simulation results of the attitude angle under the condition of adding relevant disturbances are as follows. Figure 5 As shown. To address the "chattering" phenomenon that occurs in traditional sliding mode control, a comparative simulation of this controller with the traditional sliding mode control method is performed. Figure 6 The results are the corresponding simulation results.

[0126] Depend on Figure 2 Simulation results using a fixed signal as input show that PID control has a certain overshoot and reaches stability in about 1 second, while traditional sliding mode control reaches stability in about 0.5 seconds. This controller tracks the fixed attitude input in about 0.4 seconds and maintains a stable tracking effect, with faster response speed and better dynamic response performance. Figure 3 Simulation results for signal tracking using a sinusoidal signal show that the PID control exhibits a 1% steady-state error at its maximum amplitude, while the traditional sliding mode control and the controller designed in this paper show little difference, both exhibiting smaller steady-state errors for sinusoidal signals and demonstrating good performance. To verify anti-interference capabilities, a signal was selected... Figure 4 The wind field turbulence signal shown is simulated by... Figure 5 It can be seen that when faced with interference signals such as wind fields, and the amplitude of external interference is large, PID control and sliding mode control will have large fluctuations and errors, and the tracking signal will be greatly distorted. However, this controller has good robustness to interference signals and can maintain the stability of the controlled signal. Figure 6 By comparing the control quantity u of the traditional sliding mode controller and this controller, it can be seen that the "chattering" problem of the control quantity of the traditional sliding mode controller can be effectively solved.

[0127] Example 2

[0128] Based on Embodiment 1, this embodiment provides a neural network-based UAV sliding mode attitude control system, including:

[0129] The model equation building module is configured to simplify the virtual control input and attitude dynamics equations of the quadcopter to obtain the motion attitude model equations of the UAV.

[0130] The adaptive approach law and control law determination module is configured to set the sliding surface, set an adaptive approach law for each attitude angle to control the UAV to approach the sliding surface from the current state, and set a control law based on the adaptive approach law for each attitude angle.

[0131] Variable parameter update module: configured to dynamically adjust the variable parameter terms of the control law based on gradient descent using a neural network;

[0132] Control output value determination module: It is configured to calculate the input values ​​of each control channel based on the obtained dynamic variable parameter terms and control law, and realize the control of UAV attitude through the simplified UAV attitude model equation.

[0133] It should be noted that each module in this embodiment corresponds one-to-one with each step in embodiment 1, and their specific implementation process is the same, so it will not be repeated here.

[0134] Example 3

[0135] This embodiment provides an electronic device, including a memory and a processor, as well as computer instructions stored in the memory and running on the processor. When the processor executes the computer instructions, it performs the steps described in the method of Embodiment 1.

[0136] Example 4

[0137] This embodiment provides a computer-readable storage medium for storing computer instructions, which, when executed by a processor, complete the steps described in the method of Embodiment 1.

[0138] The above description is merely a preferred embodiment of this disclosure and is not intended to limit this disclosure. Various modifications and variations can be made to this disclosure by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this disclosure should be included within the protection scope of this disclosure.

[0139] While the specific embodiments of this disclosure have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of this disclosure. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of this disclosure are still within the scope of protection of this disclosure.

Claims

1. A sliding mode attitude control method for unmanned aerial vehicles based on neural networks, characterized in that, The steps include the following: By simplifying the virtual control input and attitude dynamics equations of the quadrotor aircraft, the motion attitude model equations of the UAV are obtained. Set a sliding surface, and for each attitude angle, set an adaptive approach law to control the UAV to approach the sliding surface from the current state. Set a control law based on the adaptive approach law for each attitude angle. The gradient descent method based on neural networks dynamically adjusts the variable parameter terms of the control law; Based on the obtained dynamic variable parameter terms and control law, the input values ​​of each control channel are calculated through the simplified UAV attitude model equation to achieve UAV attitude control. The adaptive reaching law is constructed using the hyperbolic tangent function, and a natural number coefficient term is added; The adaptive reaching law is: Where s is the sliding surface. For positive numbers to be designed, An adaptive gain greater than zero. Represents attitude angle; The control law designed based on the sliding mode adaptive reaching law is as follows: Where m is the coefficient to be adjusted, and p represents the x-axis, y-axis, or z-axis to be selected. These are the moments of inertia along the x-axis, y-axis, and z-axis, respectively. Represents attitude angle, It is a positive constant to be designed. The ideal input value for the attitude angle. This represents the distance from the arm to the center of mass of the drone, where s is the sliding surface. For positive numbers to be designed, An adaptive gain greater than zero. denoted by , respectively, represents the air resistance coefficients for the carrier's rotational motion around the corresponding coordinate axes in the carrier's coordinate system. ; For the set control law, a method for dynamically adjusting the variable parameters of the control law using the gradient descent method of a neural network includes the following steps: Construct a performance error function based on the perfect squared difference between the ideal input value and the current value of the attitude angle; Based on the partial derivative chain rule, the partial derivative integral of the performance error function is used as the change of the variable parameter to construct the adaptation equation. Using the Nabla operator, the partial derivative of the attitude angle with respect to the ideal control quantity is used as the sign function of the ratio of the difference between the attitude angle and the ideal control quantity. The adaptive equation is then transformed to obtain the formula for calculating the variable parameters. The gradient is updated based on the set neural network, and the variable parameter terms corresponding to the gradient are calculated based on the obtained variable parameter calculation formula.

2. The UAV sliding mode attitude control method based on neural networks as described in claim 1, characterized in that, Set the sliding surface as: the multiple of the difference between the ideal input attitude angle and the current attitude angle, and the sum of the difference between the derivative of the ideal input attitude angle and the derivative of the current attitude angle; Alternatively, the simplified assumptions for the virtual control inputs and attitude dynamics equations of a quadcopter are as follows: The drone structure is a rigid body, and the effect of elastic deformation of the body on the model is not considered. The drone has a uniform mass, a symmetrical structure, and its geometric origin coincides with its center of mass.

3. The UAV sliding mode attitude control method based on neural networks as described in claim 1, characterized in that: The stability of the controller is analyzed using Lyapunov's stability theorem.

4. A sliding mode attitude control system for unmanned aerial vehicles based on neural networks, characterized in that, Implementing the neural network-based UAV sliding mode attitude control method as described in any one of claims 1-3, comprising: The model equation building module is configured to simplify the virtual control input and attitude dynamics equations of the quadcopter to obtain the motion attitude model equations of the UAV. The adaptive approach law and control law determination module is configured to set the sliding surface, set an adaptive approach law for each attitude angle to control the UAV to approach the sliding surface from the current state, and set a control law based on the adaptive approach law for each attitude angle. Variable parameter update module: configured to dynamically adjust the variable parameter terms of the control law based on gradient descent using a neural network; Control output value determination module: It is configured to calculate the input values ​​of each control channel based on the obtained dynamic variable parameter terms and control law, and realize the control of UAV attitude through the simplified UAV attitude model equation.

5. An electronic device, characterized in that, It includes a memory and a processor, as well as computer instructions stored in the memory and running on the processor, which, when executed by the processor, perform the steps of any one of claims 1-3.

6. A computer-readable storage medium, characterized in that, Used to store computer instructions, which, when executed by a processor, perform the steps described in any one of claims 1-3.

Images