A Robust Fixed-Wing UAV Flight Control System and Method

Abstract

The present invention relates to the technical field of fixed-wing unmanned aerial vehicles, and proposes a robust fixed-wing unmanned aerial vehicle flight control system and method. The system includes: a data acquisition module that acquires the basic motion parameters of a target fixed-wing unmanned aerial vehicle; a motion calculation module that performs motion parameter resolution on the basic motion parameters to obtain extended motion parameters; a control module that constructs a state vector, inputs the trained LSTM network model, and obtains an action vector group; and an execution module that respectively controls the corresponding control components of the target fixed-wing unmanned aerial vehicle to execute corresponding flight control actions. By acquiring various parameters of the fixed-wing unmanned aerial vehicle flight control, constructing an attitude transformation rate model and a navigation model, calculating some parameters that cannot be collected by sensors, constructing a state vector, and completing the output of control instructions through a generator and a regulator independently composed of two LSTMs based on reinforcement learning, the present invention takes into account the influence of the actual environment on the state of the aircraft and improves the robustness.

Description

Technical Field

[0001] The present invention relates to the technical field of fixed-wing unmanned aerial vehicles, and in particular to a robust fixed-wing unmanned aerial vehicle flight control system and method. Background Art

[0002] A fixed-wing unmanned aerial vehicle is an aircraft without a crew on board, with a fixed wing structure on its wing surface, similar to the wings of a traditional aircraft or an aircraft model. Different from multi-rotor unmanned aerial vehicles, fixed-wing unmanned aerial vehicles mainly generate lift through the aerodynamic force of the aircraft wing to achieve flight. Fixed-wing unmanned aerial vehicles usually have a longer endurance and a higher flight speed, and are commonly used in fields such as aerial photography, agricultural plant protection, and geological exploration. Flight control, also known as the flight control system, is a key device for controlling the attitude, position, and speed of an unmanned aerial vehicle. It measures the state of the aircraft through sensors, calculates the control commands of the aircraft, and controls the unmanned aerial vehicle through actuators (such as electric servos). However, in the face of complex meteorological conditions, communication interference, sensor failures, and system deviations, there is a risk that the flight control may fail. Therefore, developing a robust unmanned aerial vehicle flight control system can improve flight safety, reduce accident risks, and ensure the reliable operation of unmanned aerial vehicles in various complex environments.

[0003] Most traditional flight controls are based on PID, LQR, and LQGD. These flight control systems are all linear. The main disadvantage of linear flight control is its low resistance to external interference and its susceptibility to the influence of wind, airflows, and other environmental interferences. Under complex flight conditions, a linear flight control system may not be able to maintain the stability of the aircraft. Although some non-linear flight controls have emerged in recent years, the non-linearity of these methods must be linearized around certain specific operating points. Therefore, these flight controls rely heavily on an accurate propulsion model at the trim point and have poor robustness. Summary of the Invention

[0004] To solve the above-mentioned problems in the prior art, the present invention provides a robust fixed-wing unmanned aerial vehicle flight control system and method, aiming to solve the problems existing in the prior art.

[0005] In the first aspect of the present invention, a robust fixed-wing unmanned aerial vehicle flight control system is provided, including:

[0006] A data acquisition module configured to obtain the basic motion parameters of a target fixed-wing unmanned aerial vehicle;

[0007] A motion calculation module configured to perform motion parameter resolution on the basic motion parameters of the target fixed-wing unmanned aerial vehicle to obtain extended motion parameters;

[0008] A control module, configured to construct a state vector of the target fixed-wing unmanned aerial vehicle according to the basic motion parameters and the extended motion parameters, and input the state vector into a trained target LSTM network model to obtain an action vector group of the target fixed-wing unmanned aerial vehicle; wherein, the action vector group includes a plurality of control action parameters for controlling the target fixed-wing unmanned aerial vehicle.

[0009] An execution module, configured to respectively control corresponding control components of the target fixed-wing unmanned aerial vehicle to execute corresponding flight control actions according to a plurality of control action parameters in the action vector group.

[0010] Optionally, the data acquisition module specifically includes:

[0011] A motion parameter acquisition unit, configured to acquire basic motion parameters collected by a motion parameter acquisition component in the target fixed-wing unmanned aerial vehicle.

[0012] Wherein, the motion parameter acquisition component includes a gyroscope and an accelerometer, and the basic motion parameters include Euler angle attitude, main coordinate system translation speed, angular velocity, and horizontal acceleration.

[0013] Optionally, the data acquisition module specifically includes:

[0014] A tensor construction unit, configured to construct a tensor array including the basic motion parameters of the target fixed-wing unmanned aerial vehicle and an unknown inertial coordinate system translation speed.

[0015] Wherein, the expression of the tensor array is specifically:

[0016] [p, q, ζ, ψ, a] T ∈R 16 ;

[0017] Wherein, p = [p x , p y , p z is the unknown inertial coordinate system translation speed, q = [α, β, γ] is the Euler angle attitude of the target fixed-wing unmanned aerial vehicle, ζ = [u, v, w] is the main coordinate system translation speed of the target fixed-wing unmanned aerial vehicle, ψ = [P, Q, R] is the angular velocity of the target fixed-wing unmanned aerial vehicle, and a = [a x , a y , a z is the horizontal acceleration of the target fixed-wing unmanned aerial vehicle.

[0018] Optionally, the extended motion parameters include the Euler angle attitude change rate and the inertial coordinate system translation speed of the target fixed-wing unmanned aerial vehicle; the motion calculation module specifically includes:

[0019] An attitude change rate calculation unit, which is configured to construct an Euler angle attitude change rate equation based on the basic motion parameters and calculate the Euler angle attitude change rate of the target fixed-wing UAV;

[0020] Among them, the expression of the Euler angle attitude change rate equation is specifically:

[0021]

[0022] Among them, α′, β′, γ′ are the Euler angle attitude change rates of the target fixed-wing UAV;

[0023] An inertial coordinate system translation speed calculation unit, which is configured to construct a translation speed navigation equation based on the basic motion parameters and calculate the inertial coordinate system translation speed of the target fixed-wing UAV;

[0024] Among them, the expression of the translation speed navigation equation is specifically:

[0025]

[0026] Among them, ε is the rotation matrix from the main coordinate system to the inertial coordinate system.

[0027] Optionally, the control module specifically includes:

[0028] A state vector construction unit, which is configured to construct the state vector of the target fixed-wing UAV according to the basic motion parameters and the extended motion parameters; among them, the expression of the state vector is specifically:

[0029]

[0030] Among them, V a represents the velocity component in the direction of the UAV's nose, is a reference quantity;

[0031] An action vector group generation unit, which is configured to normalize the state vector, input the processing result into the trained target LSTM network model, and obtain the action vector group u of the target fixed-wing UAV.

[0032] Optionally, the target LSTM network model specifically includes:

[0033] A first LSTM network, which is configured as a model generator and outputs the action vector group of the target fixed-wing UAV when receiving the normalized state vector of the target fixed-wing UAV;

[0034] The second LSTM network, which is configured as a model regulator, takes the action vector group output by the model generator as input and outputs an adjustment amount for feedback correction of the action vector group.

[0035] Optionally, in the target LSTM network model:

[0036] The expression of the first LSTM network is specifically:

[0037]

[0038] The expression of the second LSTM network is specifically:

[0039]

[0040] The expression of the feedback-corrected action vector group is specifically:

[0041] A(s t+l ) = A(s t ) - U(u t );

[0042] Where T represents a preset duration; l represents an intermediate variable; u t is the action vector group at time t; s t is the state vector at time t; M represents the execution policy; r represents the ratio of the current stage policy to the previous stage policy; represents each action under the policy M, τ represents the action in u t , represents the expected value; R(·) represents the reward function, and the expression is specifically:

[0043]

[0044] Where (·) e represents the measured value of each parameter; f(·) is the scale scaling function; W are all weights; |·| represents the difference between the observed value and the measured value, the observed value refers to the parameter in the s t+l state, and the measured value is the parameter in the state s t .

[0045] Optionally, inputting the processing result into the trained target LSTM network model to obtain the action vector group u of the target fixed-wing UAV specifically includes:

[0046] Performing feedback correction on the action vector group A(s t+l ) = A(s t ) - U(u t) When performing iterative calculation, when the adjustment amount U(u obtained by calculation for feedback correction of the action vector group is the smallest, the model training of the target LSTM network model is completed; t )

[0047] Input the state vector s at the current moment into the trained target LSTM network model to obtain the action vector group u of the target fixed-wing UAV.

[0048] Optionally, the expression of the action vector group u is specifically:

[0049] u = [δ t , δ e , δ a , δ r ;

[0050] Among them, u = [δ t , δ e , δ a , δ r respectively correspond to the control action parameters of the throttle, elevator, aileron and rudder of the target fixed-wing UAV; the execution module specifically includes:

[0051] A control action parameter conversion unit, which is configured to extract each control action parameter in the action vector group u, obtain a control coefficient according to the position of the control action parameter in the parameter range, and generate a flight control signal based on the product of the control coefficient and the maximum value of the corresponding type of control parameter;

[0052] A flight control unit, which is configured to send each flight control signal to the control component of the corresponding type of control parameter to control the target fixed-wing UAV to perform the corresponding flight control action.

[0053] In the second aspect of the present invention, a robust fixed-wing UAV flight control method is provided, including:

[0054] S1: Obtain the basic motion parameters of the target fixed-wing UAV;

[0055] S2: Perform motion parameter calculation on the basic motion parameters of the target fixed-wing UAV to obtain extended motion parameters;

[0056] S3: According to the basic motion parameters and the extended motion parameters, construct the state vector of the target fixed-wing UAV, and input the state vector into the trained target LSTM network model to obtain the action vector group of the target fixed-wing UAV; wherein, the action vector group includes several control action parameters for controlling the target fixed-wing UAV;

[0057] S4: According to several control action parameters in the action vector group, respectively control the corresponding control components of the target fixed-wing UAV to perform corresponding flight control actions.

[0058] The beneficial effects of the present invention are as follows: A robust fixed-wing UAV flight control system and method are proposed. By obtaining various parameters of the fixed-wing UAV flight control at the current moment and the past moment, an attitude transformation rate model and a navigation model are constructed to calculate some parameters that cannot be collected by sensors. Subsequently, a state vector is constructed, and the output of the control command is completed through a generator and a regulator independently composed of two LSTMs based on the reinforcement learning method, considering the influence of the actual environment on the aircraft state, and improving the robustness of the fixed-wing UAV flight control system. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 is a schematic structural diagram of the robust fixed-wing UAV flight control system provided by the present invention;

[0060] Figure 2 is a schematic flow diagram of the robust fixed-wing UAV flight control method provided by the present invention.

[0061] Reference Signs:

[0062] 10 - Data acquisition module; 20 - Motion calculation module; 30 - Control module; 40 - Execution module. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0063] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0064] Embodiment 1:

[0065] Refer to Figure 1 , Figure 1 which is a schematic structural diagram of a robust fixed-wing UAV flight control system provided by an embodiment of the present invention.

[0066] As Figure 1As shown in the figure, a robust fixed-wing UAV flight control system includes: a data acquisition module 10 configured to obtain the basic motion parameters of a target fixed-wing UAV; a motion calculation module 20 configured to perform motion parameter resolution on the basic motion parameters of the target fixed-wing UAV to obtain extended motion parameters; a control module 30 configured to construct a state vector of the target fixed-wing UAV according to the basic motion parameters and the extended motion parameters, and input the state vector into a trained target LSTM network model to obtain an action vector group of the target fixed-wing UAV; wherein, the action vector group includes a plurality of control action parameters for controlling the target fixed-wing UAV; an execution module 40 configured to respectively control corresponding control components of the target fixed-wing UAV to execute corresponding flight control actions according to the plurality of control action parameters in the action vector group.

[0067] It should be noted that most traditional flight controls are based on PID, LQR, and LQGD. These flight control systems are all linear. The most important disadvantage of linear flight controls is their low resistance to external interference and susceptibility to the influence of wind, air currents, and other environmental interferences. Under complex flight conditions, linear flight control systems may not be able to maintain the stability of the aircraft. Although some non-linear flight controls have emerged in recent years, the non-linearity of these methods must be linearized around certain specific operating points. Therefore, these flight controls rely heavily on an accurate propulsion model at the trim point and have poor robustness. To solve the above problems, in this embodiment, various parameters at the current moment and past moments are obtained through the sensors of the UAV, including acceleration, speed, etc., to construct an attitude transformation rate model and a navigation model, calculate some parameters that cannot be collected by the sensors, and then construct a state vector s. The output of the control command is completed through a generator and a regulator independently composed of two LSTMs based on the method of reinforcement learning. This method takes into account the influence of the actual environment on the state of the aircraft and improves the robustness of the fixed-wing UAV flight control system.

[0068] In a preferred embodiment, the data acquisition module specifically includes: a motion parameter acquisition unit configured to obtain the basic motion parameters collected by a motion parameter acquisition component in the target fixed-wing UAV; wherein, the motion parameter acquisition component includes a gyroscope and an accelerometer, and the basic motion parameters include Euler angle attitude, main coordinate system translation speed, angular velocity, and horizontal acceleration.

[0069] On this basis, the data acquisition module specifically includes: a tensor construction unit configured to construct a tensor array including the basic motion parameters of the target fixed-wing UAV and the unknown inertial coordinate system translation speed.

[0070] Among them, the expression of the tensor array is specifically:

[0071] [p, q, ζ, ψ, a] T ∈R 16 ;

[0072] Among them, p = [p x , p y , p z is the unknown translational velocity of the inertial coordinate system, q = [α, β, γ] is the Euler angle attitude of the target fixed-wing UAV, ζ = [u, v, w] is the translational velocity of the main coordinate system of the target fixed-wing UAV, ψ = [P, Q, R] is the angular velocity of the target fixed-wing UAV, and a = [a x , a y , a z is the horizontal acceleration of the target fixed-wing UAV.

[0073] In this embodiment, the acquisition of the basic motion parameters of the target fixed-wing UAV is realized through the motion parameter acquisition components (such as gyroscopes, accelerometers, etc.) provided on the target fixed-wing UAV; specifically, the basic motion parameters include: Euler angle attitude (which can be measured by gyroscopes and accelerometers and solved), translational velocity of the main coordinate system (which can be measured by a speedometer), angular velocity (angular velocity), and horizontal acceleration (measured by an accelerometer). After that, the collected basic motion parameters and the translational velocity of the inertial coordinate system (the position of this parameter needs to be solved by the subsequent motion calculation module) are constructed into a tensor form for storage.

[0074] In a preferred embodiment, the extended motion parameters include the Euler angle attitude change rate of the target fixed-wing UAV and the translational velocity of the inertial coordinate system; the motion calculation module specifically includes: an attitude change rate calculation unit, which is configured to construct an Euler angle attitude change rate equation according to the basic motion parameters and calculate the Euler angle attitude change rate of the target fixed-wing UAV; an inertial coordinate system translational velocity solving unit, which is configured to construct a translational velocity navigation equation according to the basic motion parameters and calculate the translational velocity of the inertial coordinate system of the target fixed-wing UAV. Among them, the expression of the Euler angle attitude change rate equation is specifically:

[0075]

[0076] Among them, α′, β′, γ′ are the Euler angle attitude change rates of the target fixed-wing UAV;

[0077] Among them, the expression of the translational velocity navigation equation is specifically:

[0078]

[0079] Among them, ε is the rotation matrix from the main coordinate system to the inertial coordinate system.

[0080] In this embodiment, after obtaining the basic motion parameters of the target fixed-wing UAV, the extended motion parameters (i.e., some parameters that cannot be collected by sensors) can be solved according to the basic motion parameters; specifically, for the attitude change rate, based on the collected basic motion parameters of the target fixed-wing UAV, it is calculated by constructing a UAV attitude change rate equation; for the translational velocity in the inertial coordinate system, based on the collected basic motion parameters of the target fixed-wing UAV, by constructing a translational velocity navigation equation, the coordinates are converted from the main coordinate system to the inertial coordinate system using the rotation matrix. Thus, using the collected basic motion parameters of the target fixed-wing UAV, by constructing an attitude transformation rate model and a navigation model, some parameters that cannot be collected by sensors are calculated, so as to expand the basic motion parameters for controlling the flight control of the fixed-wing UAV into a set of basic motion parameters and extended motion parameters, and use more parameters as reference data for the UAV flight control. When constructing a state vector using the motion parameters and outputting an operation instruction using the state vector, it can make the flight control accuracy and accuracy of the fixed-wing UAV higher, and the environmental adaptability better, improving the system robustness.

[0081] In a preferred embodiment, the control module specifically includes: a state vector construction unit configured to construct a state vector of the target fixed-wing UAV according to the basic motion parameters and the extended motion parameters; an action vector group generation unit configured to perform normalization processing on the state vector, input the processing result into a trained target LSTM network model, and obtain an action vector group u of the target fixed-wing UAV.

[0082] Among them, the expression of the state vector is specifically:

[0083]

[0084] Among them, V a represents the velocity component in the direction of the UAV's nose, is a reference quantity, that is, the theoretical value (used for model initialization);

[0085] In a preferred embodiment, the target LSTM network model specifically includes: a first LSTM network configured as a model generator, which outputs an action vector group of the target fixed-wing unmanned aerial vehicle when receiving the state vector of the target fixed-wing unmanned aerial vehicle after normalization processing; a second LSTM network configured as a model regulator, which takes the action vector group output by the model generator as input and outputs an adjustment amount for feedback correction of the action vector group.

[0086] In a preferred embodiment, in the target LSTM network model:

[0087] The expression of the first LSTM network is specifically:

[0088]

[0089] The expression of the second LSTM network is specifically:

[0090]

[0091] The expression of the feedback-corrected action vector group is specifically:

[0092] A(s t+1 ) = A(s t ) - U(u t );

[0093] Where T represents a preset duration; l represents an intermediate variable; u t is the action vector group at time t; s t is the state vector at time t; M represents the execution policy; r represents the ratio of the current stage policy to the previous stage policy; represents each action under the policy M, τ represents the action in u t , represents the expected value; R(·) represents the reward function, and the expression is specifically:

[0094]

[0095] Where (·) e represents the measured value of each parameter; f(·) is a scale scaling function; W are all weights; |·| represents the difference between the observed value and the measured value, the observed value refers to the parameter in the state of s t+l , and the measured value is the parameter in the state s t .

[0096] It should be noted that the lateral operation of a fixed-wing UAV refers to the left-right movement of the UAV on a plane, similar to the roll rotation of an aircraft. By changing the action of the ailerons of the UAV, the yaw in the left-right direction of the UAV can be controlled. The lateral operation can also include side flight (i.e., lateral taxiing) and adjustment of the lateral flight attitude. The longitudinal operation of a fixed-wing UAV refers to the movement of the UAV in the vertical direction, including ascending and descending. By changing the action of the elevator of the UAV, the pitch of the UAV can be controlled, thereby realizing the operations of ascending and descending. In addition, the longitudinal operation can also include adjustment of the pitch angle and smooth climbing or descending. In this embodiment, the lateral operation and longitudinal operation of the fixed-wing UAV during flight are considered to construct a corresponding reward function to achieve the output of control commands based on the reinforcement learning method.

[0097] In a preferred embodiment, the processing result is input into the trained target LSTM network model to obtain the action vector group u of the target fixed-wing UAV, which specifically includes: when performing the iterative calculation of the feedback correction action vector group A(s t+1 ) = A(s t ) - U(u t ), when the adjustment amount U(u t ) for the feedback correction action vector group obtained by the calculation is the smallest, the model training of the target LSTM network model is completed; the state vector s at the current moment is input into the trained target LSTM network model to obtain the action vector group u of the target fixed-wing UAV.

[0098] In this embodiment, a generator and a regulator independently composed of two LSTM networks with 128 hidden layers are provided. The generator A(s t ) is responsible for outputting the vector u, and the regulator U(u t ) is responsible for outputting the adjustment amount according to the vector u output by the generator for feedback correction of the vector u. Based on the basic motion parameters collected by the sensor and the extended motion parameters obtained by the calculation, the flight state of the current target fixed-wing UAV is analyzed through the trained target LSTM network model composed of the generator and the regulator, so as to output the action vector group for the flight control of the target fixed-wing UAV. Among them, when training the target LSTM network model, the feedback correction action vector group A(s t+l ) = A(s t ) - U(u t ) is iteratively calculated by obtaining the parameters at the current moment and the past moment until the adjustment amount U(u t ) is the smallest, and the model training is completed. Thus, in this embodiment, the output of the control command is completed through the generator and regulator independently composed of two LSTMs and based on the reinforcement learning method. This method can consider the influence of the actual environment on the state of the aircraft and improve the system robustness.

[0099] In a preferred embodiment, the expression of the action vector group u is specifically: u = [δ t , δ e , δ a , δ r ; where [δ t , δ e , δ a , δ r respectively correspond to the control action parameters of the throttle, elevator, aileron and rudder of the target fixed-wing UAV; the execution module specifically includes: a control action parameter conversion unit configured to extract each control action parameter in the action vector group u, obtain a control coefficient according to the position of the control action parameter in the parameter range, and generate a flight control signal based on the product of the control coefficient and the maximum value of the corresponding type of control parameter; a flight control unit configured to send each flight control signal to the control component of the corresponding type of control parameter to control the target fixed-wing UAV to perform the corresponding flight control action.

[0100] In this embodiment, considering that the control components of the target fixed-wing UAV are the throttle, elevator, aileron and rudder of the aircraft, when obtaining the action vector group u of the target fixed-wing UAV at the current moment, each control action parameter in the action vector group u is used to determine the flight control signal of the corresponding type of control parameter; specifically, since the value range of each quantity in u is [-1, 1], a control coefficient can be obtained according to the position of the control action parameter in the parameter range, and a flight control signal is generated based on the product of the control coefficient and the maximum value of the corresponding type of control parameter (for example, the maximum throttle thrust is F, 0 represents 50%F, and -1 represents 0%F). Thus, flight control for different control components of the target fixed-wing UAV is achieved, considering the influence of the actual environment on the aircraft state, and the flight control accuracy of the target fixed-wing UAV is improved by using a non-linear flight control method, having high system robustness.

[0101] Refer to Figure 2 , Figure 2 which is a schematic flow diagram of a robust fixed-wing UAV flight control method provided by an embodiment of the present invention.

[0102] As Figure 2 shown, a robust fixed-wing UAV flight control method includes the steps of:

[0103] S1: Obtain the basic motion parameters of the target fixed-wing UAV;

[0104] S2: Perform motion parameter resolution on the basic motion parameters of the target fixed-wing UAV to obtain extended motion parameters;

[0105] S3: Construct the state vector of the target fixed-wing UAV based on the basic motion parameters and the extended motion parameters, and input the state vector into the trained target LSTM network model to obtain the action vector group of the target fixed-wing UAV; wherein, the action vector group includes a number of control action parameters for controlling the target fixed-wing UAV.

[0106] S4: Control the corresponding control components of the target fixed-wing UAV to perform corresponding flight control actions respectively according to the number of control action parameters in the action vector group.

[0107] In this embodiment, by acquiring various parameters of the fixed-wing UAV flight control at the current moment and the past moment, an attitude transformation rate model and a navigation model are constructed to calculate some parameters that cannot be collected by sensors. Subsequently, a state vector is constructed, and the output of the control command is completed in a reinforcement learning manner by a generator and a regulator independently composed of two LSTMs, considering the influence of the actual environment on the aircraft state, and improving the robustness of the fixed-wing UAV flight control system.

[0108] The specific implementation manners of the robust fixed-wing UAV flight control method of this application are basically the same as those of the above-described embodiments of the robust fixed-wing UAV flight control system, and will not be elaborated herein.

[0109] In the description of the embodiments of the present invention, it should be understood that the orientation or positional relationship indicated by the terms "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "center", "top", "bottom", "top part", "bottom part", "inner", "outer", "inner side", "outer side", etc. is the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, and is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation of the present invention. Among them, the "inner side" refers to the internal or enclosed area or space. The "periphery" refers to the area around a specific component or a specific area.

[0110] In the description of the embodiments of the present invention, the terms "first", "second", "third", "fourth" are only used for descriptive purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Thus, the features defined with "first", "second", "third", "fourth" may explicitly or implicitly include one or more of such features. In the description of the present invention, unless otherwise specified, the meaning of "a plurality" is two or more.

[0111] In the description of the embodiments of the present invention, it should be noted that unless otherwise clearly defined and limited, the terms "installation", "connection", "linkage", and "assembly" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a direct connection, or an indirect connection through an intermediate medium, and it can be the communication inside two components. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0112] In the description of the embodiments of the present invention, specific features, structures, materials, or characteristics can be combined in a suitable manner in any one or more embodiments or examples.

[0113] In the description of the embodiments of the present invention, it should be understood that "-" and "~" represent the range between two numerical values, and this range includes the endpoints. For example: "A - B" represents a range greater than or equal to A and less than or equal to B. "A ~ B" represents a range greater than or equal to A and less than or equal to B.

[0114] In the description of the embodiments of the present invention, the term "and / or" herein is merely a description of the association relationship of associated objects, indicating that three relationships can exist. For example, A and / or B can represent: A exists alone, A and B exist simultaneously, and B exists alone. In addition, the character " / " herein generally represents an "or" relationship between the associated objects before and after.

[0115] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirits of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A robust fixed-wing UAV flight control system, characterized in that: include: A data acquisition module, wherein the data acquisition module is configured to obtain basic motion parameters of the target fixed-wing UAV; A motion calculation module, wherein the motion calculation module is configured to perform motion parameter calculation on basic motion parameters of the target fixed-wing UAV to obtain extended motion parameters; A control module, wherein the control module is configured to construct a state vector of the target fixed-wing UAV according to the basic motion parameters and the extended motion parameters, input the state vector into the trained target LSTM network model, and obtain an action vector group of the target fixed-wing UAV; wherein the action vector group includes a plurality of control action parameters for controlling the target fixed-wing UAV; An execution module, wherein the execution module is configured to control the control components corresponding to the target fixed-wing UAV to perform corresponding flight control actions according to a plurality of control action parameters in the action vector group; The target LSTM network model specifically includes: a first LSTM network, which is configured as a model generator and outputs an action vector group of the target fixed-wing UAV when receiving a normalized state vector of the target fixed-wing UAV; a second LSTM network, which is configured as a model regulator and takes the action vector group output by the model generator as input and outputs an adjustment amount for feedback correction of the action vector group; The expression of the first LSTM network is specifically: The expression of the second LSTM network is specifically: The expression of the feedback correction action vector group is specifically: A(s t+l )=A(s t )-U(u t ); Among them, T represents a preset duration; l represents an intermediate variable; u t is the action vector group at time t; s t is the state vector at time t; represents the execution strategy; r represents the ratio of the current strategy to the previous strategy; represents each action of M under the strategy, and τ represents u t The action in represents the expected value; R(·) represents the reward function, and the specific expression is: in,(·) e represents the measured value of each parameter; f(·) is the scale scaling function; W is the weight; |·| represents the difference between the observed value and the measured value, and the observed value refers to s t+1 The parameter in state, the measured value is state s t The following parameters; Where a is the horizontal acceleration of the target fixed-wing UAV; V a The velocity component representing the direction of the drone's nose.

2. The robust fixed-wing UAV flight control system according to claim 1, characterized in that: The data acquisition module specifically includes: A motion parameter acquisition unit, wherein the motion parameter acquisition unit is configured to acquire basic motion parameters acquired by a motion parameter acquisition component in the target fixed-wing UAV; The motion parameter acquisition component includes a gyroscope and an accelerometer, and the basic motion parameters include Euler angle attitude, principal coordinate system translation velocity, angular velocity and horizontal acceleration.

3. The robust fixed-wing UAV flight control system according to claim 2, characterized in that: The data acquisition module specifically includes: A tensor construction unit, wherein the tensor construction unit is configured to construct a tensor array including basic motion parameters of the target fixed-wing UAV and an unknown inertial coordinate system translation velocity; The expression of the tensor array is specifically: [p,q,z,ψ,a] T ∈R 16 ; Where p = [p x , p y , p z ] is the unknown inertial coordinate translation velocity, q = [α, β, γ] is the Euler angle attitude of the target fixed-wing UAV, ζ = [u, v, w] is the principal coordinate translation velocity of the target fixed-wing UAV, ψ = [P, Q, R] is the angular velocity of the target fixed-wing UAV, a = [a x , a y , a z ].

4. The robust fixed-wing UAV flight control system according to claim 3, characterized in that: The extended motion parameters include the Euler angle attitude change rate and the inertial coordinate system translation speed of the target fixed-wing UAV; the motion calculation module specifically includes: An attitude change rate calculation unit, wherein the attitude change rate calculation unit is configured to construct an Euler angle attitude change rate equation according to the basic motion parameters, and calculate the Euler angle attitude change rate of the target fixed-wing UAV; The expression of the Euler angle attitude change rate equation is specifically: Among them, α′, β′, and γ′ are the Euler angle attitude change rates of the target fixed-wing UAV; An inertial coordinate system translation speed solving unit, wherein the inertial coordinate system translation speed solving unit is configured to construct a translation speed navigation equation according to the basic motion parameters, and calculate and obtain the inertial coordinate system translation speed of the target fixed-wing UAV; The expression of the translation velocity navigation equation is specifically: Among them, ε is the rotation matrix from the principal coordinate system to the inertial coordinate system.

5. The robust fixed-wing UAV flight control system according to claim 4, characterized in that: The control module specifically includes: A state vector construction unit, wherein the state vector construction unit is configured to construct a state vector of the target fixed-wing UAV according to the basic motion parameters and the extended motion parameters; wherein the expression of the state vector is specifically: in, is the reference amount; An action vector group generating unit is configured to perform normalization processing on the state vector, input the processing result into the trained target LSTM network model, and obtain the action vector group u of the target fixed-wing UAV.

6. The robust fixed-wing UAV flight control system according to claim 5, characterized in that: The processing results are input into the trained target LSTM network model to obtain the action vector group u of the target fixed-wing UAV, specifically including: Feedback correction action vector group A(s t+l )=A(s t )-U(u t ) is iterated, the adjustment amount u(u t ) completes the model training of the target LSTM network model when the minimum The current state vector s is input into the trained target LSTM network model to obtain the action vector group u of the target fixed-wing UAV.

7. The robust fixed-wing UAV flight control system according to claim 6, characterized in that: The expression of the action vector group u is specifically: u=[δ t d e d a d r ]; Among them, [δ t , δ e , δ a , δ r ] respectively correspond to the control action parameters of the throttle, elevator, aileron and rudder of the target fixed-wing UAV; the execution module specifically includes: a control action parameter conversion unit, the control action parameter conversion unit being configured to extract each control action parameter in the action vector group u, obtain a control coefficient according to a position of the control action parameter in a parameter range, and generate a flight control signal based on a product of the control coefficient and a maximum value of a control parameter of a corresponding type; A flight control unit is configured to send each flight control signal to a control component of a corresponding type of control parameter to control the target fixed-wing UAV to perform a corresponding flight control action.

8. A robust fixed-wing UAV flight control method, characterized in that: include: S1: Obtain the basic motion parameters of the target fixed-wing UAV; S2: performing motion parameter calculation on the basic motion parameters of the target fixed-wing UAV to obtain extended motion parameters; S3: constructing a state vector of the target fixed-wing UAV according to the basic motion parameters and the extended motion parameters, inputting the state vector into the trained target LSTM network model, and obtaining an action vector group of the target fixed-wing UAV; wherein the action vector group includes a plurality of control action parameters for controlling the target fixed-wing UAV; S4: According to a plurality of control action parameters in the action vector group, control components corresponding to the target fixed-wing UAV are controlled to perform corresponding flight control actions; The target LSTM network model specifically includes: a first LSTM network, which is configured as a model generator and outputs an action vector group of the target fixed-wing UAV when receiving a normalized state vector of the target fixed-wing UAV; a second LSTM network, which is configured as a model regulator and takes the action vector group output by the model generator as input and outputs an adjustment amount for feedback correction of the action vector group; The expression of the first LSTM network is specifically: The expression of the second LSTM network is specifically: The expression of the feedback correction action vector group is specifically: A(s t+l )=A(s t )-U(u t ); Among them, T represents a preset duration; l represents an intermediate variable; u t is the action vector group at time t; s t is the state vector at time t; represents the execution strategy; r represents the ratio of the current strategy to the previous strategy; represents each action of M under the strategy, and τ represents u t The action in represents the expected value; R(·) represents the reward function, and the specific expression is: in,(·) e represents the measured value of each parameter; f(·) is the scale scaling function; W is the weight; |·| represents the difference between the observed value and the measured value, and the observed value refers to s t+l The parameter in state, the measured value is state s t The following parameters; Where a is the horizontal acceleration of the target fixed-wing UAV; V a The velocity component representing the direction of the drone's nose.

Images