Variable-sweepback aircraft preset performance control method based on adaptive dynamic programming

Abstract

The embodiment of the invention provides a variable-sweepback aircraft preset performance control method based on adaptive dynamic programming, and belongs to the technical field of control, and the method specifically comprises the steps: constructing a variable-sweepback aircraft motion model for attitude control; a preset performance function is constructed, a converted error system is obtained through homeomorphic mapping, and a virtual controller and a steady-state controller are designed for the two subsystems of the attitude angle loop and the attitude angular velocity loop respectively; a cost function is introduced to convert an aircraft attitude tracking control problem into solution of an optimal control problem, and an Actor-Critic network is constructed based on an adaptive dynamic programming technology to online solve an optimal compensation controller; and a variable-sweepback aircraft controller is obtained by combining a steady-state controller and an optimal compensation controller, and numerical simulation is performed on the performance of the controller based on specific flight conditions. Through the scheme disclosed by the invention, the control system has better control precision and transient performance, and meanwhile, the robustness to an unknown flight environment is improved.

Description

Technical Field

[0001] The disclosed embodiments relate to the field of control technology, and in particular to a method for controlling preset performance of a variable-sweep aircraft based on adaptive dynamic programming. Background Art

[0002] Variable-sweep aircraft are a new type of aircraft that can autonomously change the size of the wing sweep angle according to different flight missions and flight environments to achieve better aerodynamic characteristics and optimal flight performance. They have great development potential and application value in aerospace and other fields. Compared with fixed-wing aircraft, the aerodynamic characteristics, moment of inertia, center of mass position and other parameters of the aircraft change significantly during the deformation process, which poses new challenges and higher requirements for the design of the aircraft control system. Therefore, enabling the aircraft to achieve attitude control during the deformation process is a hot topic in current research.

[0003] At present, there have been some research results on the control methods of variable-sweep aircraft. The main research direction focuses on how to effectively estimate the unknown dynamic interference and strong uncertainty brought by the deformation process, and design corresponding compensation controllers. Classical control methods have the problem of insufficient adaptability and it is difficult to ensure the stability of the aircraft under uncertain factors. Although intelligent control methods can effectively improve the adaptability of aircraft, existing methods mostly rely on large amounts of data collection and offline training. The actual control system has limited state collection capabilities, and it is difficult to obtain enough data to train the network.

[0004] It can be seen that there is an urgent need for a variable-sweep aircraft preset performance control method based on adaptive dynamic programming with high control accuracy, fast response speed and strong robustness. Summary of the invention

[0005] In view of this, an embodiment of the present disclosure provides a variable-sweep aircraft preset performance control based on adaptive dynamic programming, which at least partially solves the problems of poor control accuracy, response speed and robustness in the prior art.

[0006] The embodiment of the present disclosure provides a method for controlling preset performance of a variable-sweep aircraft based on adaptive dynamic programming, comprising:

[0007] Step 1, constructing a variable-sweep aircraft motion model for attitude control;

[0008] Step 2: Based on the preset performance framework, a preset performance function is constructed, and the converted error system is obtained through homeomorphic mapping. The variable-sweep aircraft motion model is divided into two subsystems, namely, the attitude angle loop and the attitude angular velocity loop, and controllers are designed for each subsystem to obtain a virtual controller corresponding to the attitude angle loop and a steady-state controller corresponding to the attitude angular velocity loop.

[0009] Step 3: Introduce the cost function to transform the aircraft attitude tracking control problem into the solution of the optimal control problem. Based on the adaptive dynamic programming technology, the Actor-Critic network is constructed to solve the Bellman equation online to obtain the optimal compensation controller.

[0010] Step 4, obtain the variable sweep aircraft controller based on the steady-state controller and the optimal compensation controller, and perform numerical simulation on the controller performance based on specific flight conditions.

[0011] According to a specific implementation of the embodiment of the present disclosure, step 1 specifically includes:

[0012] Step 1.1: Establish a variable-sweep aircraft motion model for attitude control

[0013]

[0014] In the formula, is the attitude angle vector, Respectively represent the pitch angle, yaw angle and roll angle, w=[w x w y w z ] T is the attitude angular velocity vector, w x ,w y ,w z Denote the roll angular velocity, yaw angular velocity and pitch angular velocity respectively, I b is the moment of inertia matrix without considering deformation, ΔI is the additional moment of inertia matrix caused by deformation, M a is the aerodynamic torque, M f is the additional moment caused by deformation, is the position vector of the wing mounting point in this system, is the rotation vector of the wing center of mass relative to the wing mounting point, d is the lumped disturbance, and the matrix R,w × The expression is as follows:

[0015]

[0016] Step 1.2, define the state variables x1 = Θ, x2 = ω, and define the variable sweep aircraft controller u = [δ x δ y δ z ] T By the steady-state controller u d and the optimal compensation controller u f Composition, δ x ,δ y ,δ z are the equivalent rudder deflection angles of the roll, yaw, and pitch channels, respectively, and the variable-sweep aircraft motion model for attitude control is obtained:

[0017]

[0018] In the formula,

[0019] u=u d +u f

[0020] I=I b +ΔI

[0021] f2=I -1 (-ω × I b ω+M0)

[0022]

[0023] g2=I -1 g

[0024]

[0025] in, is the operating torque coefficient, is the damping torque coefficient, Q is the dynamic pressure, V is the speed of the aircraft, S is the reference area, and L is the reference length.

[0026] According to a specific implementation of the embodiment of the present disclosure, step 2 specifically includes:

[0027] Step 2.1, design a controller for the attitude angle loop subsystem and define the attitude angle tracking instruction as x 1d , then the attitude angle tracking error e1(t)=[e 11 e 12 e 13 ] T =x1-x 1d , design the preset performance function ρ i (t), i = 1, 2, 3 represent three different channels, and the attitude angle tracking error satisfies the constraint: -ρ i (t)<e 1i (t)<ρ i (t);

[0028] Step 2.2, design a homeomorphic mapping function to transform the performance constraint space into the unconstrained space and obtain the converted error system:

[0029]

[0030] In the formula, ε1=[ε 11 ε 12 ε 13 ] Tis the error variable after mapping,

[0031] Step 2.3, design the virtual controller x corresponding to the attitude angle loop 2d for

[0032] x 2d =R -1 [x 2c,1 x 2c,2 x 2c,3 ] T

[0033]

[0034] In the formula, K1=[K 11 K 12 K 13 ] T is the virtual controller gain;

[0035] Step 2.4, define the attitude angular velocity tracking error e2 = x2-x 2d , whose derivative is taken as the error system:

[0036]

[0037] Step 2.5: Design the steady-state controller u corresponding to the attitude angular velocity loop based on the preset performance framework d as follows:

[0038] u d =g2 -1 (-f2(x 2d )-K2e2+ξ2)

[0039] Among them, ξ1 is the value for the virtual controller x 2d The estimated value of the reciprocal, K2 is the steady-state controller gain.

[0040] According to a specific implementation of the embodiment of the present disclosure, step 3 specifically includes:

[0041] Step 3.1, set the steady-state controller u d Substituting into the attitude angular velocity loop, the error system is transformed into:

[0042]

[0043] In the formula, f e =f2-f2(x 2d ), is the estimated error of the virtual controller derivative;

[0044] Step 3.2, design the cost function as:

[0045] J(e2(t))=r(e2(t),u f (t))+γr(e2(t+Δt),u f (t+Δt))+γ 2 r(e2(t+2Δt),u f (t+2Δt))+···

[0046] =r(e2(t),u f (t))+γJ(e2(t+Δt))

[0047] In the formula, r(e2(t),u f (t)) = e2(t) T Qe2(t)+u f (t) T Ru f (t) is the utility function at the current sampling time, Δt is the sampling step, γ is the discount coefficient, Q, R are positive definite matrices;

[0048] Step 3.3, design the optimal cost function J * (e2(t)), satisfies the Bellman equation:

[0049]

[0050] Step 3.4: Construct the Critic network and the Actor network to fit the optimal cost function J * and the optimal compensation controller u f as a solution to the optimal control problem.

[0051] According to a specific implementation of the embodiment of the present disclosure, step 3.4 specifically includes:

[0052] Step 3.4.1, construct a critic network to fit the optimal cost function, wherein the critic network includes an input layer, a hidden layer and an output layer, and its input vector C I =[e2 T x 2d T ] T , the output vector The cost function estimate It is expressed as:

[0053]

[0054] Where W c is the weight parameter of the Critic network, Φ c (·) is the activation function;

[0055] Step 3.4.2: Build the Actor network to fit the optimal compensation controller u f , where the Actor network includes an input layer, a hidden layer and an output layer, and its input vector A I =e2, output vector A O =u f , the optimal compensation controller u f It is expressed as:

[0056] u f =W a T Φ a (A I )

[0057] Where W a is the weight parameter of the Actor network, Φ a (·) is the Actor network activation function.

[0058] According to a specific implementation of the embodiment of the present disclosure, before step 3.4.1, the method further includes:

[0059] The estimated error of the Critic network is defined as:

[0060]

[0061] The objective function of updating the critic network weights is designed as:

[0062]

[0063] Based on the gradient descent strategy, the update strategy for the Critic network weight parameters is obtained as follows:

[0064]

[0065] Where η c is the Critic network learning rate;

[0066] The weight parameters of the Critic network are updated according to the update strategy of the Critic network weight parameters.

[0067] According to a specific implementation of the embodiment of the present disclosure, before step 3.4.2, the method further includes:

[0068] The estimated error of the Actor network is defined as:

[0069]

[0070] The objective function for updating the Actor network weights is designed as:

[0071]

[0072] Based on the gradient descent strategy, the update strategy for the Actor network weight parameters is obtained as follows:

[0073]

[0074] Where η a is the Actor network learning rate, is the gradient of the activation function with respect to the input;

[0075] Update the weight parameters of the Actor network according to the update strategy of the Actor network weight parameters.

[0076] The preset performance control scheme of a variable-sweep aircraft based on adaptive dynamic programming in the disclosed embodiment includes: step 1, constructing a variable-sweep aircraft motion model for attitude control; step 2, constructing a preset performance function based on a preset performance framework, obtaining a converted error system through homeomorphic mapping, dividing the variable-sweep aircraft motion model into two subsystems, an attitude angle loop and an attitude angular velocity loop, and performing controller design respectively to obtain a virtual controller corresponding to the attitude angle loop and a steady-state controller corresponding to the attitude angular velocity loop; step 3, introducing a cost function to transform the aircraft attitude tracking control problem into a solution to an optimal control problem, constructing an Actor-Critic network based on adaptive dynamic programming technology to solve the Bellman equation online to obtain an optimal compensation controller; step 4, obtaining a variable-sweep aircraft controller based on the steady-state controller and the optimal compensation controller, and performing numerical simulation of the controller performance based on specific flight conditions.

[0077] The beneficial effects of the disclosed embodiments are as follows: through the disclosed solution, the basic controller designed based on the preset performance framework and the backstepping control concept can ensure that the attitude angle error is always within the preset performance envelope during the entire flight process, with fast convergence speed, effectively suppressing overshoot, and greatly improving the steady-state performance and transient performance of the control system. At the same time, in order to increase the robustness and adaptability of the controller in different flight environments, the feedback controller is designed using adaptive dynamic programming technology, without offline training, and can be directly learned online, and updated in real time based on data such as the current tracking error to compensate for disturbances caused by uncertain factors such as aerodynamic parameters, thereby improving control accuracy, response speed and robustness. BRIEF DESCRIPTION OF THE DRAWINGS

[0078] In order to more clearly illustrate the technical solutions of the embodiments of the present disclosure, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present disclosure. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0079] Figure 1 A flow chart of a variable-sweep aircraft preset performance control method based on adaptive dynamic programming provided in an embodiment of the present disclosure;

[0080] Figure 2 A system block diagram of a variable-sweep aircraft preset performance control method based on adaptive dynamic programming provided in an embodiment of the present disclosure;

[0081] Figure 3 A schematic diagram of a sweep angle variation curve provided in an embodiment of the present disclosure;

[0082] Figure 4 Schematic diagrams of attitude angle tracking curves under two working conditions provided by the embodiments of the present disclosure, wherein (a) is the pitch angle tracking condition, (b) is the yaw angle tracking condition, and (c) is the roll angle tracking condition;

[0083] Figure 5 A schematic diagram of an attitude angle error curve provided in an embodiment of the present disclosure, wherein (a) is a pitch angle error, (b) is a yaw angle error, and (c) is a roll angle error;

[0084] Figure 6 Schematic diagram of actual rudder deflection angle curves of working condition 1 provided by an embodiment of the present disclosure, (a) is a rudder deflection angle curve No. 1, (b) is a rudder deflection angle curve No. 2, (c) is a rudder deflection angle curve No. 3, and (d) is a rudder deflection angle curve No. 4;

[0085] Figure 7 Schematic diagram of actual rudder deflection angle curves of working condition 2 provided by an embodiment of the present disclosure, (a) is a rudder deflection angle curve No. 1, (b) is a rudder deflection angle curve No. 2, (c) is a rudder deflection angle curve No. 3, and (d) is a rudder deflection angle curve No. 4;

[0086] Figure 8 Schematic diagrams of weight parameter change curves of neural networks under two working conditions provided in the embodiments of the present disclosure, wherein (a) is a weight parameter change curve of a Critic network, and (b) is a weight parameter change curve of an Actor network; DETAILED DESCRIPTION

[0087] The embodiments of the present disclosure are described in detail below with reference to the accompanying drawings.

[0088] The following is a detailed description of the embodiments of the present disclosure, but the present invention can be implemented in many different ways as defined and covered by the claims.

[0089] The embodiment of the present disclosure provides a variable-sweep aircraft preset performance control method based on adaptive dynamic programming, and the method can be applied to aircraft control processes in the aviation field.

[0090] See also Figure 1 , is a flow chart of a variable-sweep aircraft preset performance control method based on adaptive dynamic programming provided by an embodiment of the present disclosure. Figure 1 As shown, the method mainly comprises the following steps:

[0091] The overall system framework is as follows Figure 2 As shown. The controller is expressed as u=u d +u f , where u d is a steady-state controller based on a preset performance framework, u f It is the optimal compensation controller based on the HDP algorithm and is directly output by the Actor network.

[0092] Step 1: Construct a mathematical model of a variable-sweep aircraft for attitude control

[0093] The mathematical model of the variable-sweep aircraft for attitude control is established as follows:

[0094]

[0095] In the formula, is the attitude angle vector, Respectively represent the pitch angle, yaw angle and roll angle. w=[w x w y w z ] T is the attitude angular velocity vector, w x ,w y ,w z Denote the roll angular velocity, yaw angular velocity and pitch angular velocity respectively, I b is the moment of inertia matrix without considering deformation, ΔI is the additional moment of inertia matrix caused by deformation, is the position vector of the wing mounting point in this system, is the rotation vector of the wing center of mass relative to the wing mounting point, d is the lumped disturbance, and the matrix R,w × The expression is as follows:

[0096]

[0097]

[0098] M a is the aerodynamic torque, and its expression is as follows:

[0099]

[0100] In the formula, C mx ,C my ,C mzare the rolling moment, yaw moment, and pitch moment coefficients, respectively. is the control torque coefficient, Q = ρV 2 / 2 is the dynamic pressure, is the atmospheric density, V is the speed of the aircraft, S is the reference area, and L is the reference length.

[0101] M f is the additional moment caused by deformation, and its expression is as follows:

[0102]

[0103] In the formula, w j is the angular velocity of the wing, is the angular acceleration of the wing, R bj is the transformation matrix, which is expressed as follows:

[0104]

[0105] In the formula, R y is the elementary transformation matrix around the y-axis, and θ is the sweep angle.

[0106] Define the state x1 = Θ, x2 = ω, and define the controller u = [δ x δ y δ z ] T By the steady-state controller u d and the optimal compensation controller u f Composition, δ x ,δ y ,δ z are the equivalent rudder deflection angles of the roll, yaw, and pitch channels respectively, then the mathematical model of the variable-sweep aircraft for attitude control can be rewritten as:

[0107]

[0108] In the formula,

[0109] u=u d +u f

[0110] I=I b +ΔI

[0111] f2=I -1 (-ω × I b ω+M0)

[0112]

[0113] g2=I -1 g

[0114]

[0115] definition is the actual rudder deflection angle, and its transformation relationship with the equivalent rudder deflection angle, i.e., the controller u, is as follows:

[0116]

[0117] Step 2: Based on the preset performance framework, construct the preset performance function, obtain the converted new error system through homeomorphic mapping, and combine the backstepping control idea to divide the motion model into two subsystems: attitude angle loop and attitude angular velocity loop, and design controllers for each subsystem.

[0118] First, design a controller for the attitude angle loop subsystem. Define the attitude angle tracking command as x1d, then the attitude angle tracking error e1(t) = [e 11 e 12 e 13 ] T =x1-x 1d , during the entire flight, the tracking error e1(t) should always remain within the designed performance envelope, that is, satisfying:

[0119] -ρ i (t)<e 1i (t)<ρ i (t)

[0120] Among them, ρ i (t) is the preset performance function, i=1, 2, 3 represents three different channels, and its form is as follows:

[0121]

[0122] in, Mainly affects the convergence speed of the preset performance function, ρ 0,i >0 is the preset initial value of the performance function, ρ ∞,i >0 is the preset final value of the performance function, which can ensure the steady-state error of the control system.

[0123] Since the introduction of additional performance constraints increases the difficulty of controller design, in order to reduce the complexity of the corresponding controller, a homeomorphic mapping function is designed to achieve the transformation from performance constraint space to unconstrained space. This paper adopts a logarithmic mapping, and its expression is as follows:

[0124]

[0125] ε1=[ε 11 ε 12 ε 13 ] Tis the error variable after mapping, and taking its derivative, we can get:

[0126]

[0127] In the formula,

[0128] Design virtual controllerx 2d as follows:

[0129]

[0130] x 2d =R -1 [x 2c,1 x 2c,2 x 2c,3 ] T

[0131] In the formula, K1=[K 11 K 12 K 13 ] T is the virtual controller gain.

[0132] Design a controller for the attitude angular velocity loop subsystem. 2d By tracking the differentiator, an estimate of the virtual controller derivative is obtained The design of the finite time tracking differentiator is as follows:

[0133]

[0134] Where H is the gain of the differentiator, λ1,λ2>0 are constants, ξ1,ξ2 are respectively 2d and its derivatives The estimated value of .

[0135] Define the attitude angular velocity tracking error e2 = x2-x 2d , its derivative can be expressed as:

[0136]

[0137] Design a steady-state controller u based on a preset performance framework d as follows:

[0138] u d =g2 -1 (-f2(x 2d )-K2e2+ξ2)

[0139] Where K2 is the steady-state controller gain.

[0140] Step 3: Introduce the cost function to transform the problem into the solution of the optimal control problem. Based on the adaptive dynamic programming technology, construct the Actor-Critic network to solve the Bellman equation online and obtain the optimal compensation controller.

[0141] The steady-state controller u based on the preset performance framework d Substituting into the attitude angular velocity loop, the new error system form is as follows:

[0142]

[0143] In the formula, f e =f2-f2(x 2d ), is the estimated error of the derivative of the virtual controller. Therefore, the problem is transformed into designing the optimal compensation controller for the new error system to minimize the cost function.

[0144] The designed cost function expression is as follows:

[0145] J(e2(t))=r(e2(t),u f (t))+γr(e2(t+Δt),u f (t+Δt))+γ 2 r(e2(t+2Δt),u f (t+2Δt))+···

[0146] =r(e2(t),u f (t))+γJ(e2(t+Δt))

[0147] In the formula, r(e2(t),u f (t)) = e2(t) T Qe2(t)+u f (t) T Ru f (t) is the utility function at the current sampling moment, Δt is the sampling step, γ is the discount coefficient, and Q, R are positive definite matrices.

[0148] Then the optimal cost function J * (e2(t)) should satisfy the following Bellman equation:

[0149]

[0150] Since the Bellman equation itself is not easy to solve, the HDP algorithm is used to approximate the optimal solution of the Bellman equation, that is, to construct a Critic network and an Actor network to fit the optimal cost function J * and the optimal compensation controller u f .

[0151] Construct a Critic network to fit the cost function for strategy evaluation. The constructed Critic network consists of an input layer, a hidden layer, and an output layer. Its input vector C I =[e2 T x 2d T ] T , output vector Therefore, the cost function estimate It can be expressed as:

[0152]

[0153] Where W c is the weight parameter of the Critic network, Φ c (·) is the activation function.

[0154] The estimated error of the Critic network is defined as:

[0155]

[0156] The objective function of updating the critic network weights is designed as:

[0157]

[0158] Based on the gradient descent strategy, the update strategy for the Critic network weight parameters is obtained as follows:

[0159]

[0160] Where η c is the Critic network learning rate.

[0161] Build an Actor network to fit the optimal compensation controller u f For policy update, the Actor network consists of an input layer, a hidden layer, and an output layer. Its input vector A I =e2, output vector A O =u f Therefore, the optimal compensation controller u f It can be expressed as:

[0162] u f =W a T Φ a (A I )

[0163] Where W a is the weight parameter of the Actor network, Φ a (·) is the Actor network activation function.

[0164] The Actor network updates its parameters by minimizing the value function, so the estimated error of the Actor network is defined as:

[0165]

[0166] The objective function for updating the Actor network weights is designed as:

[0167]

[0168] Based on the gradient descent strategy, the update strategy for the Actor network weight parameters is obtained as follows:

[0169]

[0170] Where η a is the Actor network learning rate, is the gradient of the activation function with respect to the input.

[0171] Step 4: In summary, the preset performance control method based on adaptive dynamic programming, the controller expression is u = u d +u f , where u d is a steady-state controller based on a preset performance framework, u f It is an optimal compensation controller based on the HDP algorithm, which is directly output by the Actor network. Numerical simulation is performed based on MATLAB, and a motion model of a variable-sweep aircraft is built in MATLAB. Based on this model, the variable-sweep aircraft preset performance control method based on adaptive dynamic programming proposed in the present invention is simulated.

[0172] In order to test the robustness of the controller, two different pneumatic pulling conditions are set:

[0173] (1) Condition 1: Atmospheric density deviation -20%, aerodynamic coefficient deviation +20%, aerodynamic moment coefficient deviation +20%

[0174] (2) Condition 2: Atmospheric density deviation +20%, aerodynamic coefficient deviation -20%, aerodynamic moment coefficient deviation -20%

[0175] The initial state of the variable-sweep aircraft is set as follows: sweep angle θ(0) = 26°, pitch angle Yaw angle ψ(0) = 1°, roll angle γ(0) = 1°, attitude angular rate w x (0) = w y (0) = w z (0) = 0° / s, speed 1.5 Ma, altitude 15 km.

[0176] Preset performance controller parameter settings: K1 = [15,15,15] T ,K2=diag(25,25,25),ρ0=[5°,5°,5°] T ,ρ ∞ =[0.5°,0.5°,0.5°] T ,l=2,T=3; HDP controller parameter settings: Q=diag(300,300,300), R=diag(10,10,10), γ=0.8, η c =0.6,η a =0.6; Tracking differentiator parameter settings: H=50,λ1=[2,2,2],λ2=[2,2,2].

[0177] Figure 3 This is the sweep angle change curve. The sweep angle maintains an initial state of 26° during 0-10s, and starts to change to 50° at a speed of 2° / s after 10s, and then maintains a sweep angle of 50° to continue flying. Figure 4 These are the attitude angle tracking curves under two working conditions. In the case of initial deviation, the attitude angle can quickly track the target instruction. Figure 5 is the attitude angle error curve. It can be seen that the attitude angle error is always kept within the preset performance envelope, and the terminal tracking error is less than 0.5°. Figure 6 is the actual rudder deflection curve under working condition 1, Figure 7 is the actual rudder deflection curve under working condition 2, Figure 8 The weight parameter change curves of the Critic network and the Actor network show that both the Critic network and the Actor network converge quickly under both working conditions.

[0178] In summary, the variable-sweep aircraft preset performance control method based on adaptive dynamic programming proposed in the present invention can ensure that the aircraft maintains a stable attitude when there is uncertainty in aerodynamic parameters and atmospheric density parameters. The attitude angle during the entire flight process is always within the preset performance envelope, and the terminal error does not exceed 0.5°. The control method has good convergence speed and transient performance.

[0179] The preset performance control method for a variable-sweep aircraft based on adaptive dynamic programming provided in this embodiment can ensure that the attitude angle error is always within the preset performance envelope during the entire flight process through a basic controller designed based on a preset performance framework and backstepping control ideas. It has high control accuracy, fast convergence speed, and effectively suppresses overshoot, greatly improving the steady-state performance and transient performance of the control system. The feedback controller designed using adaptive dynamic programming technology can perform online learning without the need for offline training in advance, and can be updated online in real time based on data such as the current tracking error to compensate for disturbances caused by uncertain factors such as aerodynamic parameters, thereby increasing the robustness and adaptability of the controller in different flight environments.

[0180] It should be understood that various parts of the present disclosure may be implemented in hardware, software, firmware or a combination thereof.

[0181] The above is only a specific implementation of the present disclosure, but the protection scope of the present disclosure is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present disclosure should be included in the protection scope of the present disclosure. Therefore, the protection scope of the present disclosure should be based on the protection scope of the claims.

Claims

1. A variable-sweep aircraft preset performance control method based on adaptive dynamic programming, characterized in that: include: Step 1, constructing a variable-sweep aircraft motion model for attitude control; Step 2: Based on the preset performance framework, a preset performance function is constructed, and the converted error system is obtained through homeomorphic mapping. The variable-sweep aircraft motion model is divided into two subsystems, namely, the attitude angle loop and the attitude angular velocity loop, and controllers are designed for each subsystem to obtain a virtual controller corresponding to the attitude angle loop and a steady-state controller corresponding to the attitude angular velocity loop. Step 3: Introduce the cost function to transform the aircraft attitude tracking control problem into the solution of the optimal control problem. Based on the adaptive dynamic programming technology, the Actor-Critic network is constructed to solve the Bellman equation online to obtain the optimal compensation controller. Step 4, obtain the variable sweep aircraft controller based on the steady-state controller and the optimal compensation controller, and perform numerical simulation on the controller performance based on specific flight conditions.

2. The method according to claim 1, characterized in that , the step 1 specifically includes: Step 1.1: Establish the attitude motion model of the variable-sweep aircraft around the center of mass In the formula, is the attitude angle vector, ψ,γ represent the pitch angle, yaw angle and roll angle respectively, w=[w x w y w z ] T is the attitude angular velocity vector, w x ,w y ,w z Denote the roll angular velocity, yaw angular velocity and pitch angular velocity respectively, I b is the moment of inertia matrix without considering deformation, ΔI is the additional moment of inertia matrix caused by deformation, M a is the aerodynamic torque, M f is the additional moment caused by deformation, is the position vector of the wing mounting point in this system, is the rotation vector of the wing center of mass relative to the wing mounting point, d is the lumped disturbance, and the matrix R,w × The expression is as follows: Step 1.2, define the state variables x1 = Θ, x2 = ω, and define the variable sweep aircraft controller u = [δ x δ y δ z ] T By the steady-state controller u d and the optimal compensation controller u f Composition, δ x ,δ y ,δ z are the equivalent rudder deflection angles of the roll, yaw, and pitch channels, respectively, and the variable-sweep aircraft motion model for attitude control is obtained: In the formula, in=in d +in f I=I b +ΔI f2=I -1 (-ω × I b ω+M0) g2=I -1 g in, is the operating torque coefficient, is the damping torque coefficient, Q is the dynamic pressure, V is the speed of the aircraft, S is the reference area, and L is the reference length.

3. The method according to claim 2, characterized in that , the step 2 specifically includes: Step 2.1, design a controller for the attitude angle loop subsystem and define the attitude angle tracking instruction as x 1d , then the attitude angle tracking error e1(t)=[e 11 e 12 e 13 ] T =x1-x 1d , design the preset performance function ρ i (t), i = 1, 2, 3 represent three different channels, and the attitude angle tracking error satisfies the constraint: -ρ i (t)<e 1i (t)<ρ i (t); Step 2.2, design a homeomorphic mapping function to transform the performance constraint space into the unconstrained space and obtain the converted error system: In the formula, ε1=[ε 11 ε 12 ε 13 ] T is the error variable after mapping, Step 2.3, design the virtual controller x corresponding to the attitude angle loop 2d for x 2d =R -1 [x 2c,1 x 2c,2 x 2c,3 ] T In the formula, K1=[K 11 K 12 K 13 ] T is the virtual controller gain; Step 2.4, define the attitude angular velocity tracking error e2 = x2-x 2d , whose derivative is taken as the error system: Step 2.5, the steady-state controller ud corresponding to the attitude angular velocity loop designed based on the preset performance framework is as follows: u d =g2 -1 (-f2(x 2d )-K2e2+ξ2) Among them, ξ1 is the value for the virtual controller x 2d The estimated value of the reciprocal, K2 is the steady-state controller gain.

4. The method according to claim 3, characterized in that , the step 3 specifically includes: Step 3.1, set the steady-state controller u d Substituting into the attitude angular velocity loop, the error system is transformed into: In the formula, f e =f2-f2(x 2d ), is the estimated error of the virtual controller derivative; Step 3.2, design the cost function as: J(e2(t))=r(e2(t),u f (t))+γr(e2(t+Δt),u f (t+Δt)) +g 2 r(e2(t+2Δt),u f (t+2Δt))+··· =r(e2(t),u f (t))+γJ(e2(t+Δt)) In the formula, r(e2(t),u f (t)) = e2(t) T Qe2(t)+u f (t) T Ru f (t) is the utility function at the current sampling time, Δt is the sampling step, γ is the discount coefficient, Q, R are positive definite matrices; Step 3.3, design the optimal cost function J * (e2(t)), satisfies the Bellman equation: Step 3.4: Construct the Critic network and the Actor network to fit the optimal cost function J * and the optimal compensation controller uf as the solution to the optimal control problem.

5. The method according to claim 4, characterized in that , the step 3.4 specifically includes: Step 3.4.1, construct a critic network to fit the optimal cost function, wherein the critic network includes an input layer, a hidden layer and an output layer, and its input vector C I =[e2 T x 2d T ] T , the output vector The cost function estimate It is expressed as: Where W c is the weight parameter of the Critic network, Φ c (·) is the activation function; Step 3.4.2: Build the Actor network to fit the optimal compensation controller u f , where the Actor network includes an input layer, a hidden layer and an output layer, and its input vector A I =e2, output vector A O =u f , the optimal compensation controller u f It is expressed as: u f =W a T Φ a (A I ) Where W a is the weight parameter of the Actor network, Φ a (·) is the Actor network activation function.

6. The method according to claim 5, characterized in that , before step 3.4.1, the method further includes: The estimated error of the Critic network is defined as: The objective function of updating the critic network weights is designed as: Based on the gradient descent strategy, the update strategy for the Critic network weight parameters is obtained as follows: Where η c is the Critic network learning rate; The weight parameters of the Critic network are updated according to the update strategy of the Critic network weight parameters.

7. The method according to claim 5, characterized in that , before step 3.4.2, the method further includes: The estimated error of the Actor network is defined as: The objective function for updating the Actor network weights is designed as: Based on the gradient descent strategy, the update strategy for the Actor network weight parameters is obtained as follows: Where η a is the Actor network learning rate, is the gradient of the activation function with respect to the input; Update the weight parameters of the Actor network according to the update strategy of the Actor network weight parameters.

Images