A Neural Network Identification Method for High-Performance Maneuverability Attitude Control of Variable-Configuration Unmanned Aerial Vehicles

Abstract

This invention discloses a neural network-based, disturbance-resistant, high-maneuverability attitude control method for variable-configuration unmanned aerial vehicles (UAVs). The method includes the following steps: Step 1: Establishing a multi-rigid-body kinematic model and a dynamic model of the variable-configuration UAV; Step 2: Constructing the input vector and sensitive region of the neural network identifier, and building an adaptive learning law for the weights of the neural network identifier; Step 3: Designing a baseline attitude controller for the variable-configuration UAV based on an offline-modeled nonlinear dynamic model; Step 4: Solving the control signal within a single control cycle and driving the servo motor to complete the actuation command; Step 5: In the next control cycle, using online measurement data from sensors to complete the adaptive learning and updating of the neural network identifier's weights. This method can comprehensively identify nonlinear unknown dynamics and external disturbances based on the current state of the variable-configuration UAV, and perform high-maneuverability attitude control based on the identification results, thereby improving the flight quality of the variable-configuration UAV.

Description

Technical Field

[0001] This invention relates to an attitude control method for unmanned aerial vehicles (UAVs), specifically to a neural network-based, disturbance-resistant, and highly maneuverable attitude control method for variable-configuration UAVs. Background Technology

[0002] Variable-configuration unmanned aerial vehicles (UAVs) are aircraft that adaptively change their configuration according to mission requirements and flight environment, offering advantages over traditional fixed-configuration UAVs such as wider flight envelope and stronger mission capabilities. During flight, the attitude control performance (dynamic quality and steady-state accuracy) of variable-configuration UAVs is crucial for the execution efficiency of missions such as ground reconnaissance and ground attack. It is necessary to overcome the deformation-related terms, unmodeled nonlinear dynamics, and external disturbances of variable-configuration UAVs to achieve low-jerkiness, high-precision attitude command tracking. Traditional UAVs typically achieve full-envelope attitude control through offline parameter tuning and gain sequencing, but this is difficult to adapt to the reduced stability margin caused by unmodeled nonlinear dynamics and the loss of control accuracy due to strong external disturbances. Therefore, there is an urgent need for a neural network-based, disturbance-resistant, high-maneuverability attitude control method for variable-configuration UAVs that can adapt to deformation-related terms, unmodeled nonlinear dynamics, and external disturbances. Summary of the Invention

[0003] To overcome the aforementioned shortcomings of existing technologies, this invention considers the impact of nonlinear dynamic terms that are difficult to model offline in the dynamic model of variable-configuration UAVs, deformation-related disturbance terms that are difficult to obtain online, and external wind disturbances on the attitude control of UAVs. It provides a neural network-based method for identifying disturbance-resistant, high-maneuverability attitude control for variable-configuration UAVs. This method can comprehensively identify nonlinear unknown dynamics and external disturbances based on the current state of the variable-configuration UAV, and perform high-maneuverability attitude control based on the identification results, thereby improving the flight quality of variable-configuration UAVs.

[0004] The objective of this invention is achieved through the following technical solution:

[0005] A neural network-based method for controlling the attitude of a variable-configuration unmanned aerial vehicle (UAV) with strong maneuverability and disturbance resistance includes the following steps:

[0006] Step 1: Analyze the structural composition of the variable-configuration UAV and establish a multi-rigid-body kinematic and dynamic model of the variable-configuration UAV;

[0007] Step 2: Consider the deformation-related interference terms and the dynamics of the microsystem of the actuator mechanism during the flight of the variable configuration UAV, analyze the dynamic related physical quantities, construct the input vector and sensitive interval of the neural network identifier, and build the weight adaptive learning law of the neural network identifier.

[0008] Step 3: Considering the requirements for fast response and high control precision of the variable configuration UAV attitude within the flight envelope, design the baseline attitude controller of the variable configuration UAV based on the offline modeling nonlinear dynamic model. It should have dynamic and steady-state qualities such as low overshoot, short rise time and small steady-state error.

[0009] Step 4: During the flight of the variable configuration UAV, the control signal is calculated within a single control cycle based on the online measurement data of the sensors, and the servo motor is driven to complete the actuation command.

[0010] Step 5: In the next control cycle, based on the design parameter update law of the neural network identifier, the weight adaptive learning and update of the neural network identifier is completed using the online measurement data of the sensor, and then the process returns to step 4.

[0011] Compared with the prior art, the present invention has the following advantages:

[0012] (1) Design a neural network identifier to accurately identify the deformation-added interference terms and unmodeled nonlinear terms of the dynamics of variable configuration UAVs. The identifier has a simple structure and good engineering applicability.

[0013] (2) Combining the neural network identifier with the preset time controller effectively improves the nonlinear adaptability and convergence speed of the attitude control system of the variable configuration UAV, thereby enhancing the flight performance of the UAV.

[0014] (3) The upper limit of the convergence time of the designed preset time controller is only related to two parameters. The parameter tuning process is simple and the convergence time can be dynamically adjusted online.

[0015] (4) The preset time controller parameters can be adaptively adjusted according to the control error signal, thereby realizing high-gain control under large disturbance conditions and low-gain control under large noise conditions, effectively improving the practicality of the control system. Attached Figure Description

[0016] Figure 1 A flowchart for a neural network-based method for identifying disturbance-resistant, high-maneuverability attitude control for variable-configuration unmanned aerial vehicles;

[0017] Figure 2 The image shows the simulation results. Detailed Implementation

[0018] The technical solution of the present invention will be further described below with reference to the accompanying drawings, but it is not limited thereto. Any modifications or equivalent substitutions to the technical solution of the present invention that do not depart from the spirit and scope of the technical solution of the present invention should be covered within the protection scope of the present invention.

[0019] This invention provides a neural network-based method for identifying and controlling the highly maneuverable attitude of a variable-configuration unmanned aerial vehicle (UAV). The variable-configuration UAV features both folding and retractable wing surfaces on its left and right wings, which deform symmetrically during flight. Figure 1 As shown, the specific steps of UAV kinematics and dynamics modeling, neural network identifier design, baseline controller construction, control command calculation, and neural network identifier weight update are as follows:

[0020] Step 1: Analyze the structural composition of the variable-configuration UAV and establish its multi-rigid-body kinematic and dynamic models. The specific steps are as follows:

[0021] Step 1.1: Based on the deformation mode of the variable configuration UAV, decompose the variable configuration UAV into multiple rigid bodies (deformable body part and fixed body part). The specific steps are as follows:

[0022] Based on the deformation method of the variable configuration UAV, the fixed parts of the UAV fuselage and wings are regarded as the fixed body parts, and the relevant parameters are represented by the subscript "0". The deformable body parts of the UAV include the retractable left wing part, the retractable right wing part, the folding left wing part, and the folding right wing part. The relevant parameters of the retractable left wing part are represented by the subscript "1", the relevant parameters of the retractable right wing part are represented by the subscript "2", the relevant parameters of the folding left wing part are represented by the subscript "3", and the relevant parameters of the folding right wing part are represented by the subscript "4".

[0023] Step 1.2: Based on the equation of conservation of angular momentum Dynamic modeling of rotation about the center of mass is performed for different rigid bodies, among which, Indicates external torque. Represents angular momentum. To express time, the specific steps are as follows:

[0024] Based on the equation of conservation of angular momentum, the rotational dynamics of the stationary and deformable parts about the center of mass are modeled separately. The rotational dynamics models of all parts about the center of mass are then superimposed to obtain:

[0025]

[0026] in, The control torque coefficient representing rudder deflection. Indicates the rudder vector. This is the aerodynamic stabilizing moment vector acting on the entire variable-configuration UAV. This represents the gravitational moment vector generated by each part of the deformable body relative to the stationary body. This indicates the mass of each part of the transforming machine body. This represents the relative position vector of the center of mass of each part of the deformable mecha compared to the center of mass of the stationary mecha. This represents the absolute velocity vector of the fixed body part. This represents the moment of inertia matrix of the fixed body parts. This represents the angular velocity vector of the stationary body relative to the ground inertial coordinate system.

[0027] Step 1.3: Derive the kinematic equations of rotation about the center of mass of the variable configuration UAV based on the geometric relationship between the ground inertial coordinate system and the body coordinate system. The specific steps are as follows:

[0028] Set the attitude angles of the variable configuration UAV relative to the ground inertial coordinate system. The rotational angular velocity vector of the variable configuration UAV is set as Based on the transformation relationship between the ground inertial coordinate system and the body coordinate system, we can obtain:

[0029]

[0030] in, , Indicates pitch angle, Indicates the yaw angle. Indicates the roll angle. Indicates the roll angular velocity. Indicates yaw rate, This indicates the pitch angular velocity.

[0031] This completes the construction of the kinematic and dynamic model of the variable configuration UAV.

[0032] Step 2: Considering the deformation-related disturbances and the dynamics of the actuator microsystem during the flight of the variable-configuration UAV, analyze the dynamically related physical quantities, construct the input vector and sensitive interval of the neural network identifier, and build the weight adaptive learning law of the neural network identifier. The specific steps are as follows:

[0033] Step 2.1: Consider the deformation dynamics of the variable configuration UAV Dynamics of microsystems of actuators Based on the dynamic equation of rotation about the center of mass (1), the input quantities of the neural network identifier are selected to include the deformation vector of the variable configuration UAV. Deformation derivative vector Rudder angle vector rudder angle derivative vector Attitude angle vector and attitude angular velocity vector , Includes the wing's extension / retraction rate (0~100%) and folding angle (0~70°). Including the pitch rudder angle from the pitch channel, the yaw rudder angle from the yaw channel, and the roll rudder angle from the roll channel, the total input vector of the neural network identifier can be expressed as: ;

[0034] Step 2.2: Define a continuously monotonically increasing barrier function. ,in The independent variable of the barrier function and , as independent variable The upper bound of the absolute value, Represent the independent variable The boundary function and satisfying at the initial time Place Then solve for the monotonically increasing barrier function. Regarding the independent variable The lower bound of the absolute value of the partial derivative ,Right now ;

[0035] Step 2.3: Define the neural network discriminator The output state quantity estimate is ,in Indicates the input vector. Represents the input vector The estimated value, Let represent the identifier design parameter vector, then the estimated error is set to . ,in That is, the order of the input vector of the neural network identifier, and the estimation error for each term. Set boundary functions and satisfy Then set the function. And solve them separately. , and Specific form;

[0036] Step 2.4: Definition , , , Let be the activation function vector, where , , express norm, Represent natural numbers of Power of 1 It is a constant; For another activation function matrix, , It is a constant;

[0037] Step 2.5: Define the neural network discriminator as follows:

[0038]

[0039] in, and Let represent the weight vectors respectively, and the update law for the weight vectors is designed as follows:

[0040]

[0041] in, All of these are constants, and the neural network identifier design is now complete.

[0042] Step 3: Considering the requirements for fast attitude response and high control precision within the flight envelope of the variable-configuration UAV, a baseline attitude controller for the variable-configuration UAV is designed based on an offline-modeled nonlinear dynamic model. This controller must possess dynamic and steady-state qualities such as low overshoot, short rise time, and small steady-state error. The specific steps are as follows:

[0043] Step 3.1: Based on the kinematic and dynamic equations of the variable-configuration UAV, establish a control-oriented attitude tracking error model for the variable-configuration UAV, where the state variables are attitude angle error. and attitude angular velocity error The specific steps are as follows:

[0044] Define attitude angle command vector The attitude angle command tracking error is The attitude angle tracking error model of the variable configuration UAV can then be expressed as:

[0045]

[0046] in, , This indicates uncertain aerodynamic torque and external disturbances. It can be obtained from the guidance circuit;

[0047] Step 3.2: Based on the time-invariant preset time convergence theory, design a preset time convergence controller for variable configuration UAVs. ,in The specific steps to express a control law are as follows:

[0048] Step 3.2.1: Define the tuning function , ,in It is a constant;

[0049] Step 3.2.2: Introduce variables And the control law is set to the following form:

[0050]

[0051] Wherein, time-varying gain function and Proportional, and The attitude angle tracking error model under the action of this control law is... It converges to the neighborhood near the origin within a certain time.

[0052] Step 4: During the flight of the variable configuration UAV, based on the online measurement data from the sensors, the control signal is calculated within a single control cycle, and the servo motors are driven to complete the actuation commands. The specific steps are as follows:

[0053] Step 4.1: Obtain real-time attitude angles using attitude angle sensors and angular velocity sensors. and attitude angular velocity Real-time deformation amplitude is obtained through deformation sensors and deformation rate sensors. and deformation rate The rudder angle is obtained through the rudder sensor. and rudder angular velocity Based on real-time attitude angle commands, the attitude angle error is obtained through a fixed-time differentiator. and attitude angular velocity error The specific steps are as follows:

[0054] Step 4.1.1: Initialize the neural network discriminator: weights Each is initially a unit vector and a matrix of all ones in its corresponding dimension. Simply set it to a positive number, based on... The distribution range of each scalar in the vector, from step 2.4 and It can be set to the distribution range of the corresponding scalar; based on The expected distribution of each scalar in the equation, and the corresponding boundary function. , where the boundary function The initial value can be non-zero;

[0055] Step 4.1.2: Initialize the control law: Set the time constant according to engineering requirements. and Set the time-varying gain function This makes the time-varying gain function and Proportional, and ;set up It can be any positive number;

[0056] Step 4.1.3: Obtain real-time attitude angles using attitude angle sensors and angular velocity sensors. and attitude angular velocity Real-time deformation amplitude is obtained through deformation sensors and deformation rate sensors. and deformation rate The rudder angle is obtained through the rudder sensor. and rudder angular velocity Based on real-time attitude angle commands and guidance loop output values, the attitude angle error is obtained through a fixed-time differentiator. and tracking error derivative ;

[0057] Step 4.2: Neural Network-Based Recognizer The identification results and the preset time convergence controller The specific steps to complete the control variable calculation are as follows:

[0058] Step 4.2.1: Based on attitude angle error and tracking error derivative Update the gain coefficient ;

[0059] Step 4.2.2: Solve for the control quantity in formula (6) The signal is transmitted to the servo system to achieve aerodynamic rudder deflection.

[0060] Step 5: In the next control cycle, based on the design parameter update law (4) of the neural network identifier, the weight adaptive learning and update of the neural network identifier is completed using the online measurement data of the sensor, and then the process returns to step 4. The specific steps are as follows:

[0061] Step 5.1: Real-time quantities obtained from sensors Using formula (4), solve for the weight update value. and ,in The online calculation cycle for variable configuration UAVs;

[0062] Step 5.2: Weight Update Value and Real-time Weight Value and The weights are added together to obtain the updated weight values, which are then substituted into the recognizer to form the recognition result. and ;

[0063] Step 5.3: If the UAV has not yet reached the target point, return to step 4.3 based on the identification results to complete the control law calculation for the next cycle.

[0064] Example:

[0065] Simulation conditions:

[0066] (1) Overall parameters of the UAV: ​​The wing sweep angle is 0°, including the telescopic wing section, the fixed wing section and the folding wing section. The total mass of the UAV fuselage and the fixed wing section is 4200 kg, the mass of a single telescopic wing is 220 kg, and the mass of a single folding wing is 90 kg. The range of variation of a single telescopic wing is [0, 400] mm. The folding wing can be in two states: "at 70° with the fixed wing" and "on the same horizontal plane as the fixed wing". The length of the folding wing is 200 mm.

[0067] (2) The position of the center of mass of the deformable wing portion of the variable configuration UAV in the body coordinate system is:

[0068]

[0069] (3) Simulation settings:

[0070] The simulation considers that the variable-configuration UAV undergoes two configuration changes during flight, with each configuration change being as follows:

[0071] 1) The first change configuration folding wing maintains a 20° angle, and the second telescopic wing extends by 300mm.

[0072] 2) The first change configuration folding wing maintains a 30° angle, and the second telescopic wing extends by 250mm.

[0073] 3) The first change configuration folding wing maintains a 40° angle, and the second telescopic wing extends by 200mm.

[0074] 4) The first change configuration folding wing maintains a 50° angle, and the second telescopic wing extends by 150mm.

[0075] 5) The first time the folding wing changes configuration, it remains at 60°. The second time the folding wing extends by 100mm.

[0076] The controller parameters are set as follows: The identifier parameters are set as follows: ; , , Activation function parameters ,and Set to the following order . Weight Each is initially a unit vector and a matrix of all ones of the corresponding dimension.

[0077] (4) Simulation results:

[0078] Simulation results are as follows Figure 2As shown, "adaptive" represents the simulation result with the addition of a neural network identifier, and "non-adaptive" represents the simulation result without the addition of a neural network identifier. The simulation results show that, under the control system, the attitude angle error of the variable configuration UAV converges rapidly within a preset time, and the introduction of the neural network identifier significantly reduces the attitude tracking error.

Claims

1. A method for anti-disturbance strong maneuvering attitude control of a morphing unmanned aerial vehicle (UAV) neural network identification, characterized in that The method comprises the following steps: Step 1: analyzing the structural composition of the morphing unmanned aerial vehicle, and establishing a multi-rigid-body kinematics model and a dynamics model of the morphing unmanned aerial vehicle; Step 2: considering the deformation additional disturbance term and the microsystem dynamics of the actuating mechanism existing in the flight process of the morphing unmanned aerial vehicle, analyzing dynamic related physical quantities, constructing an input vector and a sensitive interval of the neural network identifier, and building a weight self-adaptive learning law of the neural network identifier; Step 3: considering the response rapidity requirement and the high-precision control requirement of the attitude in the flight envelope of the morphing unmanned aerial vehicle, designing a baseline attitude controller of the morphing unmanned aerial vehicle based on the offline modeling nonlinear dynamics model, and the specific steps are as follows: Step 3.1: based on the kinematics and dynamics equations of the morphing unmanned aerial vehicle, an attitude tracking error model of the morphing unmanned aerial vehicle facing control is established; Step 3.2: Designing the morphing UAV pre-set time convergence controller based on the non-time-varying pre-set time convergence theory wherein denotes the control law; Step 4: in the flight process of the morphing unmanned aerial vehicle, the control signal is calculated in a single control cycle according to the online measurement data of the sensor, and the actuating mechanism is driven to complete the actuating instruction; Step 5: in the next control cycle, the weight self-adaptive learning and updating of the neural network identifier are completed based on the design parameter updating law of the neural network identifier, and then the step 4 is returned.

2. The alloform unmanned aerial vehicle neural network identification anti-disturbance strong maneuvering attitude control method according to claim 1, characterized in that The specific steps of the step 1 are as follows: Step 1.1: according to the deformation mode of the morphing unmanned aerial vehicle, the morphing unmanned aerial vehicle is decomposed into multiple rigid bodies; Step 1.2: Based on the angular momentum conservation equation Modeling the dynamics of rotation around the center of mass for different rigid bodies, where, Mextdenotes the external moment, Ldenotes the angular momentum, t denotes the time; Step 1.3: the kinematics equation of the morphing unmanned aerial vehicle around the center of mass is derived based on the geometric relationship between the ground inertial coordinate system and the body coordinate system.

3. The alloform unmanned aerial vehicle neural network identification anti-disturbance strong maneuvering attitude control method according to claim 2, characterized in that The specific steps of the step 1.1 are as follows: According to the deformation mode of the morphing unmanned aerial vehicle, the unmanned aerial vehicle body and the fixed wing surface part are regarded as the fixed body part, and the related parameters are represented by subscript "0"; the deformation body part of the unmanned aerial vehicle includes the telescopic left wing surface part, the telescopic right wing surface part, the folding left wing surface part and the folding right wing surface part, the related parameters of the telescopic left wing surface part are represented by subscript "1", the related parameters of the telescopic right wing surface part are represented by subscript "2", the related parameters of the folding left wing surface part are represented by subscript "3", and the related parameters of the folding right wing surface part are represented by subscript "4"; The specific steps of the step 1.2 are as follows: Based on the angular momentum conservation equation, the fixed body part and the deformation body part are modeled for rotation around the center of mass respectively, and the rotation dynamics models of all parts around the center of mass are superimposed to obtain: wherein, is the control moment coefficient representing rudder deflection, is the rudder swing vector, is the aerodynamic stability moment vector acting on the whole morphing UAV, is the gravity moment vector generated by each part of the morphing body to the fixed body part, is the mass of each part of the morphing body, is the relative position vector of the center of mass of each part of the morphing body compared to the center of mass of the fixed body part, is the absolute velocity vector of the fixed body part, is the rotational inertia matrix of the fixed body part, is the rotational angular velocity vector of the fixed body part relative to the ground inertial coordinate system; The specific steps of the step 1.3 are as follows: The attitude angle of the set variable configuration unmanned aerial vehicle in the ground inertial coordinate system is The rotation angular velocity vector of the set variable configuration unmanned aerial vehicle is According to the conversion relationship between the ground inertial coordinate system and the body coordinate system, wherein , denotes the pitch angle, denotes the yaw angle, denotes the roll angle, denotes the roll angle velocity, denotes the yaw angle velocity, denotes the pitch angle velocity.

4. The alloform unmanned aerial vehicle neural network identification anti-disturbance strong maneuvering attitude control method according to claim 3, characterized in that The specific steps of the step 2 are as follows: Step 2.1: Considering the morphing dynamics of the morphing UAV and the actuation mechanism microsystem dynamics , the input quantities of the neural network identifier are selected based on the rotation dynamics equation around the center of mass, including the morphing quantity vector of the morphing UAV , the derivative vector of the morphing quantity , the rudder swing angle vector , the derivative vector of the rudder swing angle , the attitude angle vector , and the attitude angular velocity vector , so that the total input quantity vector of the neural network identifier is represented as ; Step 2.2: Define a continuously monotonically increasing barrier function. ,in The independent variable of the barrier function and , as independent variable The upper bound of the absolute value, Represent the independent variable The boundary function and satisfying at the initial time Place Then solve for the monotonically increasing barrier function. Regarding the independent variable The lower bound of the absolute value of the partial derivative ,Right now ; Step 2.3: Define the neural network identifier The output state estimation value is wherein represents the input vector, represents the input vector of the estimation value, and the estimation error is set as wherein is the order of the input vector of the neural network identifier, and each estimation error is set as a boundary function and satisfies ; then the function is set, and the specific forms of , and are solved respectively; Step 2.4: Definition , , , is a vector of activation functions, where , , denotes the norm of , denotes the natural number power of is a constant; is another matrix of activation functions, , is a constant;​ Step 2.5: the neural network identifier is defined as: where and denote the weight vectors, respectively, and the update law for the weight vectors is designed as wherein are constants, at which point the neural network discriminator design is complete.

5. The alloform drone neural network identification anti-disturbance strong maneuvering attitude control method according to claim 4, characterized in that The specific steps of the step 3.1 are as follows: Defining the attitude angle command vector , the attitude angle command tracking error is The attitude angle tracking error model of the variable configuration UAV is represented as wherein , denotes uncertain aerodynamic moments and external disturbances; The specific steps of the step 3.2 are as follows: Step 3.2.1 : Defining the tuning function , where is a constant, and are time constants; Step 3.2.2: Introduce variables and set the control law form as: where the time-varying gain function is proportional to and .

6. The alloform drone neural network identification anti-disturbance strong maneuvering attitude control method according to claim 5, characterized in that The specific steps of the step 4 are as follows: Step 4.1: Obtain real-time attitude angle through attitude angle sensor and angular velocity sensor and attitude angular velocity , obtain real-time deformation amplitude through deformation sensor and deformation rate sensor and deformation rate , obtain rudder swing angle through rudder sensor and rudder swing angular velocity ; based on real-time attitude angle instruction, obtain attitude angle error through fixed time differentiator and attitude angular velocity error ; Step 4.2: Recognition result based on neural network recognizer and preset time convergence controller The control amount calculation is completed.

7. The alloform drone neural network identification anti-disturbance strong maneuvering attitude control method according to claim 6, characterized in that The specific steps of the step 4.1 are as follows: Step 4.1.1: Initialization of the neural network discriminator: weights each initially a unit vector of the corresponding dimension and an all-ones matrix, set to a normal number, depending on the distribution range of each scalar in the vector, set the weights in step 2.4 and to a value within the distribution range of the corresponding scalar; depending on the expected distribution of each scalar in the vector, set the corresponding boundary function where the initial value of the boundary function is non-zero; Step 4.1.2: Initialization of the control law: Set time constant according to engineering requirements and ; Set time-varying gain function such that the time-varying gain function is proportional to and ; Set to be any positive constant; Step 4.1.3: Real-time attitude angle is obtained through the attitude angle sensor and the angular velocity sensor and attitude angular velocity Real-time deformation amplitude is obtained through the deformation sensor and the deformation rate sensor and deformation rate Rudder swing angle is obtained through the rudder sensor and rudder swing angular velocity Based on the real-time attitude angle instruction and the guidance loop output value, the attitude angle error is obtained through the fixed-time differentiator and tracking error derivative ; The specific steps of the step 4.2 are as follows: Step 4.2.1 : Update of the pose angle error based on the tracking error derivative and the tracking error derivative , the gain coefficient ; Step 4.2.2: Solve the control variable in equation (6) and transmitted to the rudder system to realize the deflection of the aerodynamic rudder.

8. The alloform drone neural network identification anti-disturbance strong maneuvering attitude control method according to claim 7, characterized in that The specific steps of the step 5 are as follows: Step 5.1: Obtaining real-time quantities based on sensors and equation (4), solve the weight update value and where is the online solution period of the morphing UAV; Step 5.2: Weight update value is added to real-time weight value and to obtain updated weight value, and the updated weight value is substituted into the recognizer to form recognition result and ; Step 5.3: if the unmanned aerial vehicle has not reached the target point at this time, return to step 4.3 based on the identification result to complete the control law calculation of the next cycle.

Images