Adaptive backstepping control method for morphing aircraft based on RBF neural network

Abstract

This invention discloses an adaptive inversion control method for variant aircraft based on RBF neural networks, comprising the following steps: setting up an RBF neural network to estimate the unknown influence term d of the trajectory inclination angle caused by model uncertainty due to aircraft deformation and external disturbances. γ And the unknown influence term d of pitch angular velocity q This invention discloses an adaptive inverse control method for variant aircraft based on RBF neural networks. The method uses RBF to approximate unknown influence terms and compensates for them in the control law, thereby improving the accuracy and robustness of the controller. The adaptive weights of the RBF neural network for unknown influence terms of ballistic tilt angle and pitch angular velocity are obtained and used to set up an inverse adaptive controller. The adaptive weights of the RBF neural network are then added to the control law of the controller, and the variant aircraft is controlled through the inverse adaptive controller.

Description

Technical Field

[0001] This invention relates to an adaptive inversion control method for variant aircraft based on RBF neural networks, belonging to the field of guidance and control. Background Technology

[0002] Variant aircraft can change their aerodynamic shape to maintain superior flight performance under various complex flight conditions and different flight environments. Variant aircraft control can be roughly divided into: linear control methods, nonlinear control methods, and intelligent control methods.

[0003] Linear control methods were first applied to solve the control problem of variator aircraft, and based on this, methods such as pole placement, scheduling gain, optimal control and robust control were derived. These methods only consider the stability of the state near the equilibrium point and cannot guarantee the global stability of the entire flight process.

[0004] Nonlinear control can be used to design nonlinear controllers to improve control performance by addressing the nonlinear characteristics of the morphing process of a morphing aircraft. While this can improve system stability, it also introduces disturbances due to system uncertainties. Although state observers can be used to counteract these disturbances, traditional state observers have poor robustness, and the combination of state observers and morphing can lead to control jitter during the morphing process.

[0005] Therefore, it is necessary to further study the control methods for variant aircraft in order to solve the above problems. Summary of the Invention

[0006] To overcome the above problems, the inventors conducted in-depth research and designed an adaptive inversion control method for variant aircraft based on RBF neural networks, comprising the following steps:

[0007] An RBF neural network is set up to estimate the unknown influence term d of the ballistic inclination angle caused by model uncertainties due to aircraft deformation and external disturbances. γ And the unknown influence term d of pitch angular velocity q The adaptive weights of the RBF neural network for the unknown influence term of the ballistic inclination angle are obtained. And the adaptive weights of the RBF neural network for the unknown influence term of pitch angular velocity.

[0008] An inversion adaptive controller is set up, and the adaptive weights of the RBF neural network are added to the control law of the controller. The control of the variant aircraft is achieved through the inversion adaptive controller.

[0009] In a preferred embodiment, the RBF neural network is represented as:

[0010]

[0011] Where x1 and x2 are the input vectors of the RBF neural network, W γ W q ε represents the weight parameters of the RBF neural network. γ ε q Let Φ be the approximation error of the RBF neural network, and Φ() be the radial basis function.

[0012] In a preferred embodiment, the control object model of the inversion adaptive controller is set as the longitudinal altitude model of the variant aircraft, represented as:

[0013]

[0014] in,

[0015] θ=α+γ

[0016]

[0017]

[0018]

[0019]

[0020]

[0021]

[0022] h represents the aircraft altitude, γ represents the trajectory angle, θ represents the pitch angle, α represents the angle of attack, q represents the pitch rate, V represents the aircraft velocity; m represents the aircraft mass, g represents the gravitational acceleration, I... y This represents the moment of inertia of the aircraft.

[0023] ξ represents the sweep angle, L0(ξ) represents the lift at 0° angle of attack, and L α (ξ) represents the derivative of lift with respect to angle of attack; T represents the thrust of the aircraft, F IKz Z represents the inertial force generated during the deformation of a morphing aircraft. T M represents the thrust arm of the aircraft. A0 (ξ) represents α=q=δ e The pitching moment at 0°, M Aα (ξ) represents the derivative of the pitching moment with respect to the angle of attack, M Aq (ξ) represents the derivative of the pitch moment with respect to the pitch angular velocity, c A (ξ) represents the mean chord length of the aircraft. This represents the derivative of the pitch moment with respect to the pitch deflection angle.

[0024] In a preferred embodiment, the tracking errors of the inversion adaptive controller are set as altitude tracking error, trajectory tilt angle tracking error, pitch angle tracking error, and pitch velocity tracking error.

[0025] In a preferred embodiment, the setting of the inversion adaptive controller includes the following steps:

[0026] S21. Set the virtual control rate γ for the ballistic inclination angle. r Eliminate high tracking error;

[0027] S22. Set the virtual control law for pitch angle θ r Eliminate the ballistic tilt tracking error introduced by the virtual control rate γr of the ballistic tilt angle;

[0028] S23. Set the virtual control rate of pitch angular velocity q r Eliminate the introduction of the virtual control law θ for pitch angle r Pitch angle tracking error;

[0029] S24. Set the control law δ e Eliminate the introduction of the virtual control law q for pitch angular velocity r The pitch angular velocity tracking error.

[0030] In a preferred embodiment, in S21, the ballistic tilt virtual control rate γ r Set to:

[0031]

[0032] Where k1 is a settable constant, h r Here, e1 represents the altitude command, and e1 represents the altitude tracking error, expressed as e1 = hh. r ;

[0033] Set the virtual control rate γ for the ballistic inclination angle r Then, the derivative of the altitude tracking error is:

[0034]

[0035] Where e2 is the ballistic tilt tracking error, expressed as e2=γ-γ r .

[0036] In a preferred embodiment, in S22, the pitch angle virtual control law θ r Set to:

[0037]

[0038] in, It is a compensation term for the uncertainty of the approximation error of the RBF neural network. For the control compensation of the unknown influence term of the ballistic inclination angle obtained by RBF adaptive approximation, k2 is a settable constant. For γ r The first derivative;

[0039] Preferably, the control compensation for the unknown influence term of the ballistic inclination angle Set as

[0040] Set the virtual control law θ for the pitch angle. r Then, the derivative of the ballistic tilt tracking error is:

[0041]

[0042]

[0043] Where e3 represents the pitch tracking error, The adaptive weights of the RBF neural network are for the unknown influence term of the ballistic inclination angle, and the adaptive law of these adaptive weights is... Set to:

[0044] τ represents the adaptive law coefficient.

[0045] In a preferred embodiment, a second-order command filter for ballistic tilt angle is set to obtain γ. r first derivative The second-order command filter for the ballistic tilt angle is expressed as follows:

[0046]

[0047] in, As an intermediate variable, Indicates the relationship with γ r The first derivative of η2 is a configurable constant, and λ2 is a configurable constant.

[0048] In a preferred embodiment, in S23, the virtual control law of pitch angular velocity q r Set to:

[0049]

[0050] Where k3 is a settable constant. For θ r The first derivative,

[0051] Set the virtual control rate of pitch angular velocity q r Then, the derivative of the ballistic tilt tracking error is:

[0052]

[0053] Where e4 represents the pitch angular velocity tracking error, expressed as e3 = θ - θ r .

[0054] In a preferred embodiment, in S24, the control law δ e Set to:

[0055]

[0056] in, It is a perturbation compensation term for the approximation error of the RBF neural network. For q r The first derivative, For control compensation of the unknown influence term of pitch angular velocity obtained by RBF adaptive approximation,

[0057] Preferably, Set to:

[0058]

[0059] in, The adaptive weights of the RBF neural network for the unknown influence term of pitch angular velocity are defined by the adaptive law of these weights. Set to: σ represents the adaptive law coefficient.

[0060] After setting the control law, the derivative of the pitch angular velocity tracking error is expressed as:

[0061]

[0062]

[0063] The beneficial effects of this invention include:

[0064] (1) Solved the control jitter problem caused by the deformation process of the morphing aircraft;

[0065] (2) By setting a second-order command filter, the “differential explosion” in the inversion control method is avoided, and the large overshoot caused by direct feedback is reduced;

[0066] (3) Using RBF to approximate unknown influence terms and compensate in the control law can improve the accuracy and robustness of the controller and effectively suppress uncertain disturbances in the system. Attached Figure Description

[0067] Figure 1 A schematic flowchart of an adaptive inversion control method for a variant aircraft based on an RBF neural network according to a preferred embodiment of the present invention is shown.

[0068] Figure 2 This diagram illustrates the setup process of the inversion adaptive controller in a variant aircraft adaptive inversion control method based on an RBF neural network according to a preferred embodiment of the present invention.

[0069] Figure 3 The graph showing the sweep angle variation in Example 1 is shown.

[0070] Figure 4 The altitude tracking curves of the variant aircraft in Example 1, Comparative Example 1, and Comparative Example 2 are shown.

[0071] Figure 5 The curves showing the variation of the trajectory inclination angle γ of the variant aircraft in Example 1, Comparative Example 1, and Comparative Example 2 are shown.

[0072] Figure 6 The curves showing the change in angle of attack α of the variant aircraft in Example 1, Comparative Example 1, and Comparative Example 2 are shown.

[0073] Figure 7 The pitch angular velocity q variation curves of the variant aircraft in Example 1, Comparative Example 1, and Comparative Example 2 are shown.

[0074] Figure 8 The control input rudder deflection angle δ of the variant aircraft shown in Example 1, Comparative Example 1, and Comparative Example 2 is illustrated. e Change curve;

[0075] Figure 9 The example shown in Example 1 illustrates the treatment of the uncertainty term d. γ The estimated value versus the actual value;

[0076] Figure 10 The example shown in Example 1 illustrates the treatment of the uncertainty term d. q The estimated value and the actual value. Detailed Implementation

[0077] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Through these descriptions, the features and advantages of the present invention will become clearer and more apparent.

[0078] The term “exemplary” as used herein means “serving as an example, embodiment, or illustration.” Any embodiment illustrated herein as “exemplary” is not necessarily to be construed as superior to or better than other embodiments. Although various aspects of embodiments are shown in the accompanying drawings, the drawings are not necessarily drawn to scale unless specifically indicated otherwise.

[0079] According to the present invention, an adaptive inversion control method for a variant aircraft based on an RBF neural network is provided, such as... Figure 1 As shown, it includes the following steps:

[0080] An RBF neural network is set up to estimate the unknown influence term d of the ballistic inclination angle caused by model uncertainties due to aircraft deformation and external disturbances. γ And the unknown influence term d of pitch angular velocity q The adaptive weights of the RBF neural network for the unknown influence term of the ballistic inclination angle are obtained. And the adaptive weights of the RBF neural network for the unknown influence term of pitch angular velocity.

[0081] An inversion adaptive controller is set up, and the weight update rate of the RBF neural network is added to the control law of the controller. The control of the variant aircraft is achieved through the inversion adaptive controller.

[0082] According to the present invention, the RBF neural network is selected from the RBFNN network. RBFNN is a single-hidden-layer neural network with advantages such as fast convergence speed and strong approximation ability.

[0083] In a preferred embodiment, the RBF neural network is represented as:

[0084]

[0085] Where x1 and x2 are the input vectors of the RBF neural network, W γ W q ε represents the weight parameters of the RBF neural network. γ ε q Let Φ be the approximation error of the RBF neural network, and Φ() be the radial basis function, which is usually a Gaussian function.

[0086] In a preferred embodiment, the radial basis function Φ() is expressed as:

[0087] Φ(x)={Φ1(x), Φ2(x),…,Φ j (x), ..., Φ n (x)}

[0088]

[0089] Among them, c j b represents the coordinate vector of the center point of the Gaussian function of the j-th neuron in the hidden layer. j represents the width of the center point of the Gaussian function of the j-th neuron in the hidden layer, and n represents the number of neurons in the hidden layer.

[0090] According to the present invention, the approximation error of the RBF neural network is bounded, expressed as |ε|≤ε N , ε N It is a positive number.

[0091] In a preferred embodiment, the control object model of the inversion adaptive controller is set as a longitudinal altitude model of the variator aircraft. This altitude model is capable of characterizing the influence of changes in the sweep angle on aerodynamic forces, aerodynamic moment coefficients, and moment of inertia.

[0092] More preferably, the longitudinal altitude model of the variant aircraft is represented as:

[0093]

[0094] According to the present invention, in the longitudinal altitude model of the variant aircraft, the ballistic tilt angle γ is a very small value, sinγ = γ, the velocity V is set to a constant value, the value of the perturbation d is bounded and differentiable, and there exists a positive constant N. d Make |d|≤N d and

[0095] The longitudinal altitude model of the variant aircraft is then represented as:

[0096]

[0097] in,

[0098] θ=α+γ

[0099]

[0100]

[0101]

[0102]

[0103]

[0104]

[0105] h represents the aircraft altitude, γ represents the trajectory angle, α represents the angle of attack, θ represents the pitch angle, q represents the pitch rate, V represents the aircraft velocity; m represents the aircraft mass, g represents the gravitational acceleration, I... y This represents the moment of inertia of the aircraft.

[0106] ξ represents the sweep angle, L0(ξ) represents the lift at 0° angle of attack, and L α (ξ) represents the derivative of lift with respect to angle of attack; T represents the thrust of the aircraft, F IKz Z represents the inertial force generated during the deformation of a morphing aircraft. T M represents the thrust arm of the aircraft. A0 (ξ) represents α=q=δ e The pitching moment at 0°, MAα (ξ) represents the derivative of the pitching moment with respect to the angle of attack, M Aq (ξ) represents the derivative of the pitch moment with respect to the pitch angular velocity, c A (ξ) represents the mean chord length of the aircraft. This represents the derivative of the pitch moment with respect to the pitch deflection angle.

[0107] Furthermore, f γ g γ f q and g q It is a known term containing aerodynamic parameters, d γ For the unknown influence of the ballistic inclination angle, d q The pitch angular velocity is an unknown influence term. Both are unknown influence terms caused by the uncertainty resulting from aircraft deformation and external disturbances.

[0108] In a preferred embodiment, the inertial force generated during the deformation of the morphing aircraft can be expressed as:

[0109]

[0110]

[0111] Among them, S x (ξ) is the static moment of the morphing aircraft, which can be expressed as:

[0112] S x (ξ)≈2m1r 1x +m3r 3x

[0113] m1 and m3 represent the masses of the wing and fuselage of the variant aircraft, respectively; r 1x and r 3x These represent the positions of the wings and fuselage in the missile's coordinate system, respectively.

[0114] Furthermore, the lift L of the aircraft can be expressed as:

[0115]

[0116] The pitching moment M of the aircraft A It can be represented as:

[0117]

[0118] in, and Let represent the fitting polynomials for the aerodynamic derivative and the derivative of the aerodynamic moment coefficient, respectively, as they change with the sweep angle ξ.

[0119] According to the present invention, the tracking errors of the inversion adaptive controller are set as altitude tracking error, ballistic tilt angle tracking error, pitch angle tracking error and pitch velocity tracking error.

[0120] More preferably, the settings of the inversion adaptive controller are as follows: Figure 2 As shown, it includes the following steps:

[0121] S21. Set the virtual control rate γ for the ballistic inclination angle. r Eliminate high tracking error;

[0122] S22. Set the virtual control law for pitch angle θ r Eliminate the ballistic tilt tracking error introduced by the virtual control rate γr of the ballistic tilt angle;

[0123] S23. Set the virtual control rate of pitch angular velocity q r Eliminate the introduction of the virtual control law θ for pitch angle r Pitch angle tracking error;

[0124] S24. Set the control law δ e Eliminate the introduction of the virtual control law q for pitch angular velocity r The pitch angular velocity tracking error.

[0125] In S21, the virtual control rate of the ballistic tilt angle γ r Set to:

[0126]

[0127] Where k1 is a settable constant, h r Here, e1 represents the altitude command, and e1 represents the altitude tracking error, expressed as e1 = hh. r ;

[0128] Furthermore, according to the present invention, the height instruction h r It is a constant, and

[0129] The height tracking error e1 = hh r Taking the time derivative with respect to e1, we get:

[0130]

[0131] That is, setting the virtual control rate γ of the ballistic inclination angle. r Then, the derivative of the altitude tracking error is:

[0132]

[0133] Where e2 is the ballistic tilt tracking error, expressed as e2=γ-γ r .

[0134] In a preferred embodiment, in S22, the pitch angle virtual control law θ r Set to:

[0135]

[0136] in, This is the uncertainty compensation term for the approximation error of the RBF neural network, preferably set as follows: l1 is a designable parameter. For the control compensation of the unknown influence term of the ballistic inclination angle obtained by RBF adaptive approximation, k2 is a settable constant. For γ r The first derivative;

[0137] Preferably, the control compensation for the unknown influence term of the ballistic inclination angle Set as

[0138] The tracking error e2 = γ - γ is derived from the ballistic tilt angle. r Taking the time derivative with respect to e², we get:

[0139]

[0140] That is, set the virtual control law θ for the pitch angle. r Then, the derivative of the ballistic tilt tracking error is:

[0141]

[0142]

[0143] Where e3 represents the pitch tracking error, The adaptive weights of the RBF neural network are for the unknown influence term of the ballistic inclination angle, and the adaptive law of these adaptive weights is... Set to:

[0144] τ represents the adaptive law coefficient, which is preferably set to τ = 0.001.

[0145] The inventors discovered that obtaining gamma using traditional methods... r The first derivative, i.e. the method of continuous differentiation, can lead to the "differential explosion" problem, and due to different initial value settings, the tracking error can be too large, which will lead to the control overshoot problem.

[0146] According to a preferred embodiment of the present invention, a second-order command filter for ballistic tilt angle is set to obtain γ. r first derivative The second-order command filter for the ballistic tilt angle is expressed as follows:

[0147]

[0148] in, As an intermediate variable, Indicates the relationship with γ r The first derivative of η2 is a configurable constant, and λ2 is a configurable constant.

[0149] In this invention, by setting a second-order command filter for the ballistic tilt angle, the "differential explosion" problem can be avoided and the control overshoot problem can be solved.

[0150] In a preferred embodiment, in S23, the virtual control law of pitch angular velocity q r Set to:

[0151]

[0152] Where k3 is a settable constant. For θ r The first derivative;

[0153] The pitch tracking error e3 = θ - θ r Taking the time derivative with respect to e3, we get:

[0154]

[0155] make That is, set the virtual control law q for pitch angular velocity. r Then, the derivative of the ballistic tilt tracking error is:

[0156]

[0157] Where e4 represents the pitch angular velocity tracking error, expressed as e3 = θ - θ r .

[0158] The inventors discovered that obtaining θ using traditional methods... r The first derivative, i.e. the method of continuous differentiation, can lead to the "differential explosion" problem, and due to different initial value settings, the tracking error can be too large, which will lead to the control overshoot problem.

[0159] In a preferred embodiment, a second-order pitch command filter is used to obtain θ. r The first derivative of the pitch angle second-order command filter is expressed as:

[0160]

[0161] in, As an intermediate variable, Indicates the relationship with θr The first derivative of η3 is a configurable constant, and λ3 is a configurable constant.

[0162] In this invention, by setting a second-order command filter for pitch angle, the "differential explosion" problem can be avoided, and the control overshoot problem caused by excessive tracking error due to different initial value settings can be solved.

[0163] In a preferred embodiment, in S24, the control law δ e Set to:

[0164]

[0165] in, This is a perturbation compensation term for the approximation error of the RBF neural network, preferably... l2 is a designable parameter. For q r The first derivative, due to the differential relationship between pitch angle and pitch velocity, For control compensation of the unknown influence term of pitch angular velocity obtained by RBF adaptive approximation,

[0166] Preferably, Set to:

[0167]

[0168] in, For the adaptive weights of the RBF neural network in response to the unknown influence of pitch angular velocity, preferably, the adaptive law of these adaptive weights... Set to:

[0169] σ represents the adaptive law coefficient, and it is preferably set to σ = 0.1.

[0170] The pitch tracking error e4 = qq r Taking the time derivative with respect to e4, we get:

[0171]

[0172] make That is, after setting the control law, the derivative of the pitch angular velocity tracking error is expressed as:

[0173]

[0174]

[0175] In this invention, a stability analysis was also performed on the above-mentioned preferred adaptive inversion control method for variant aircraft based on RBF neural networks, specifically:

[0176] First, define the Lyapunov function. Taking its time derivative yields:

[0177]

[0178] Next, we define the Lyapunov function. Taking its time derivative yields:

[0179]

[0180] Due to the weights of the RBF neural network The adaptive law is:

[0181]

[0182] We can obtain:

[0183]

[0184] Secondly, define the Lyapunov function. Taking its time derivative yields:

[0185]

[0186] Finally, define the Lyapunov function. Taking its time derivative yields:

[0187]

[0188] Due to the weights of the RBF neural network The adaptive law is:

[0189]

[0190] We can obtain:

[0191]

[0192] Since the approximation error of the neural network is bounded, i.e., |ε γ |≤ε N 、|ε q |≤ε N We can obtain:

[0193]

[0194] make By designing suitable parameters l1 and l2 such that |l1+l2|≥|e2+e4|, then we have This ultimately stabilized the control system.

[0195] Example

[0196] Example 1

[0197] Simulation experiments were conducted. The aerodynamic forces, aerodynamic moments, and aircraft model parameters in the simulation were based on data provided in the literature [Wu, Z., Lu, J., Rajput, J., Shi, J., Ma, W.: Adaptive neural control based on high order integral chained differentiator for morphing aircraft. Math. Probl. Eng. (2015).]. The uncertainty of the aerodynamic coefficient was set to 20%. The initial values ​​for the aircraft altitude tracking control simulation were set as: [h0, γ0, θ0, q0] = [1100m, 0°, 1°, 0° / s], and the altitude tracking command was h r =1200m, flight speed set to V=40m / s. The sweep angle of the mutator changes over time. To simulate the sweep angle change during the mutator's flight, the sweep angle is set to change from 0° to 45° in the first 20 seconds of flight time and then remain constant. The sweep angle change curve is shown below. Figure 3 As shown.

[0198] The simulation control is performed using an adaptive inversion control method for a variant aircraft based on an RBF neural network, including the following steps:

[0199] Set the RBF neural network to estimate the unknown influence term d of the ballistic inclination angle. γ And the unknown influence term d of pitch angular velocity q ;

[0200] An inversion adaptive controller is set up, and the weight update rate of the RBF neural network is added to the control law of the controller. The control of the variant aircraft is achieved through the inversion adaptive controller.

[0201] The RBF neural network mentioned above is represented as:

[0202]

[0203] The radial basis function Φ() is expressed as:

[0204] Φ(x)={Φ1(x), Φ2(x),…,Φ j (x), ..., Φ n (x)}

[0205]

[0206] The number of neurons in the hidden layer is set to 5.

[0207] The control object model of the inversion adaptive controller is set as the longitudinal altitude model of the variant aircraft, represented as:

[0208]

[0209] in,

[0210]

[0211]

[0212]

[0213]

[0214]

[0215]

[0216] In the longitudinal dynamics model, the uncertain external disturbance is designed as Δd. γ =2e -0.1t sin(0.1t)°, Δd q =10e -0.05t sin(0.1t)°.

[0217] The tracking errors of the inversion adaptive controller are set as altitude tracking error, trajectory tilt angle tracking error, pitch angle tracking error, and pitch velocity tracking error.

[0218] The setting of the inversion adaptive controller includes the following steps:

[0219] S21. Set the virtual control rate γ for the ballistic inclination angle. r Eliminate high tracking error;

[0220] S22. Set the virtual control law for pitch angle θ r Eliminate the ballistic tilt tracking error introduced by the virtual control rate γr of the ballistic tilt angle;

[0221] S23. Set the virtual control rate of pitch angular velocity q r Eliminate the introduction of the virtual control law θ for pitch angle r Pitch angle tracking error;

[0222] S24. Set the control law δ e Eliminate the introduction of the virtual control law q for pitch angular velocity r The pitch angular velocity tracking error.

[0223] In S21, the virtual control rate of the ballistic tilt angle γ r Set to:

[0224]

[0225] Set the virtual control rate γ for the ballistic inclination angle r Then, the derivative of the altitude tracking error is:

[0226]

[0227] Where k1 = 3, e2 = γ - γ r .

[0228] In S22, the virtual control law of pitch angle θ r Set to:

[0229]

[0230] Where k2 = 5, is the control compensation for the unknown influence term of the ballistic inclination angle. Set as

[0231] Set the virtual control law θ for the pitch angle. r Then, the derivative of the ballistic tilt tracking error is:

[0232]

[0233]

[0234] Among them, the adaptive law of adaptive weights Set to:

[0235] Set a second-order command filter for the ballistic inclination angle to obtain γ. r first derivative The second-order command filter for the ballistic tilt angle is expressed as follows:

[0236]

[0237] Where λ2=10, η2=0.707.

[0238] In S23, the virtual control law of pitch angular velocity q r Set to:

[0239]

[0240] Where k3 = 6, the virtual control law of pitch angular velocity q is set. r Then, the derivative of the ballistic tilt tracking error is:

[0241]

[0242] Set the second-order command filter for pitch angle to obtain θ. r The first derivative of the pitch angle second-order command filter is expressed as:

[0243]

[0244] Where λ3 = 20 and η3 = 0.707.

[0245] In S24, the control law δ e Set to:

[0246]

[0247] in, Set to:

[0248]

[0249] After setting the control law, the derivative of the pitch angular velocity tracking error is expressed as:

[0250]

[0251]

[0252] Comparative Example 1

[0253] The same experiment as in Example 1 was conducted, except that the outer loop controller proposed in the literature Li, G., Q, W.: Switching control of morphing aircraft based on Q-learning, Chin. J. Aeronaut, 33(2), 672-687 (2020) was used.

[0254] Comparative Example 2

[0255] The same experiment as in Example 1 was conducted, except that the RBF neural network was not set up, and the inversion adaptive controller was still used to control the variant aircraft. However, the inversion adaptive controller did not have the weights of the RBF neural network.

[0256] The simulation results obtained in Example 1, Comparative Example 1, and Comparative Example 2 are as follows: Figure 4-10 As shown, m1 represents the simulation curve in Example 1, m2 represents the simulation curve in Comparative Example 1, and m3 represents the simulation curve in Comparative Example 1. Figure 4 The altitude tracking curve of the variant aircraft is shown. Figure 5 The curves showing the variation of the trajectory inclination angle γ of the variant aircraft are shown. Figure 6 The curves showing the change in angle of attack α of the variant aircraft are shown. Figure 7The pitch rate q variation curve of the variant aircraft is shown. Figure 8 The control input rudder deflection angle δ of the variant aircraft is shown. e Change curve.

[0257] from Figure 4 It can be seen that the control methods in Example 1, Comparative Example 1, and Comparative Example 2 can all ensure that the altitude curve effectively tracks the altitude command and can converge the tracking error. Figure 4 The magnified view shows that m1 has a better convergence speed and tracking error accuracy than m2 and m3.

[0258] from Figure 5-8 As can be seen from the data, the control methods in Example 1, Comparative Example 1, and Comparative Example 2 all guarantee that the control methods are bounded and convergent. It can be seen that within t∈[0,20s], due to the change in the sweep angle of the aircraft, the state variables γ, α, q, and control variables δ of m2 change. e Significant shaking occurred. Due to the influence of uncertain disturbances during the deformation of the aircraft, m3 exhibited state and control variables with large fluctuations. In comparison, m1 showed the best control performance.

[0259] Figure 9 The example shown in Example 1 illustrates the treatment of the uncertainty term d. γ The estimated value and the actual value, Figure 10 The example shown in Example 1 illustrates the treatment of the uncertainty term d. q The estimated and actual values ​​show that the error between the estimated and actual values ​​can be controlled within ±0.05%, indicating that the method in Example 1 can effectively suppress uncertain disturbances in the system and has good accuracy.

[0260] In the description of this invention, it should be noted that the terms "upper," "lower," "inner," "outer," "front," and "rear," etc., indicate the orientation or positional relationship based on the orientation or positional relationship in the working state of this invention, and are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of this invention. Furthermore, the terms "first," "second," "third," and "fourth" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0261] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal communication between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0262] The present invention has been described above with reference to preferred embodiments; however, these embodiments are merely exemplary and illustrative. Various substitutions and modifications can be made to the present invention based on these embodiments, all of which fall within the scope of protection of the present invention.

Claims

1. An adaptive inversion control method for a variant aircraft based on an RBF neural network, characterized in that, Includes the following steps: An RBF neural network is set up to estimate the unknown influence term d of the ballistic inclination angle caused by model uncertainties due to aircraft deformation and external disturbances. γ And the unknown influence term d of pitch angular velocity q The adaptive weights of the RBF neural network for the unknown influence term of the ballistic inclination angle are obtained. And the adaptive weights of the RBF neural network for the unknown influence term of pitch angular velocity. An inversion adaptive controller is set up, and the adaptive weights of the RBF neural network are added to the control law of the controller. The control of the variant aircraft is achieved through the inversion adaptive controller. The setting of the inversion adaptive controller includes the following steps: S21. Set the virtual control rate γ for the ballistic inclination angle. r Eliminate high tracking error; S22. Set the virtual control law for pitch angle θ r Eliminate the virtual control rate γ introduced by the ballistic tilt angle r The ballistic tilt angle tracking error; S23. Set the virtual control rate of pitch angular velocity q r Eliminate the introduction of the virtual control law θ for pitch angle r Pitch angle tracking error; S24. Set the control law δ e Eliminate the introduction of the virtual control law q for pitch angular velocity r Pitch angular velocity tracking error; In S21, the virtual control rate of the ballistic tilt angle γ r Set to: Where k1 is a settable constant, h r Here, e1 represents the altitude command, and e1 represents the altitude tracking error, expressed as e1 = hh. r V represents the speed of the aircraft; Set the virtual control rate γ for the ballistic inclination angle r Then, the derivative of the altitude tracking error is: Where e2 is the ballistic tilt tracking error, expressed as e2=γ-γ r ; In S22, the virtual control law of pitch angle θ r Set to: in, It is a compensation term for the uncertainty of the approximation error of the RBF neural network. For the control compensation of the unknown influence term of the ballistic inclination angle obtained by RBF adaptive approximation, k2 is a settable constant. For γ r The first derivative, f γ g γ It includes known items containing aerodynamic parameters; Set a second-order command filter for the ballistic inclination angle to obtain γ. r first derivative The second-order command filter for the ballistic tilt angle is expressed as follows: in, As an intermediate variable, Indicates the relationship with γ r The first derivative of η2, where η2 is a configurable constant and λ2 is a configurable constant; In S23, the virtual control law of pitch angular velocity q r Set to: Where k3 is a settable constant. For θ r The first derivative of , e3 represents the pitch tracking error; In S24, the control law δ e Set to: in, It is a perturbation compensation term for the approximation error of the RBF neural network. For q r The first derivative, For control compensation of the unknown influence term of pitch angular velocity obtained by RBF adaptive approximation, e4 is the pitch tracking error, f q g q It contains known items including aerodynamic parameters.

2. The adaptive inversion control method for variant aircraft based on RBF neural network according to claim 1, characterized in that, The RBF neural network is represented as follows: Where x1 and x2 are the input vectors of the RBF neural network, W γ W q ε represents the weight parameters of the RBF neural network. γ ε q Let Φ be the approximation error of the RBF neural network, and Φ() be the radial basis function.

3. The adaptive inversion control method for variant aircraft based on RBF neural network according to claim 1, characterized in that, The control object model of the inversion adaptive controller is set as the longitudinal altitude model of the variant aircraft, represented as: in, θ=α+γ h represents the aircraft altitude, γ represents the trajectory angle, θ represents the pitch angle, α represents the angle of attack, q represents the pitch rate; m represents the aircraft mass, g represents the gravitational acceleration, I... y This represents the moment of inertia of the aircraft. ξ represents the sweep angle, L0(ξ) represents the lift at 0° angle of attack, and L α (ξ) represents the derivative of lift with respect to angle of attack; T represents the thrust of the aircraft, F IKz Z represents the inertial force generated during the deformation of a morphing aircraft. T M represents the thrust arm of the aircraft. A0 (ξ) represents α=q=δ e The pitching moment at 0°, M Aα (ξ) represents the derivative of the pitching moment with respect to the angle of attack, M Aq (ξ) represents the derivative of the pitch moment with respect to the pitch angular velocity, c A (ξ) represents the mean chord length of the aircraft. This represents the derivative of the pitch moment with respect to the pitch deflection angle.

4. The adaptive inversion control method for variant aircraft based on RBF neural network according to claim 1, characterized in that, The tracking errors of the inversion adaptive controller are set as altitude tracking error, trajectory tilt angle tracking error, pitch angle tracking error, and pitch velocity tracking error.

5. The adaptive inversion control method for variant aircraft based on RBF neural network according to claim 1, characterized in that, Control compensation for unknown influence terms of ballistic inclination angle Set as Set the virtual control law θ for the pitch angle. r Then, the derivative of the ballistic tilt tracking error is: in, The adaptive weights of the RBF neural network are for the unknown influence term of the ballistic inclination angle, and the adaptive law of these adaptive weights is... Set to: τ represents the adaptive law coefficient.

6. The adaptive inversion control method for variant aircraft based on RBF neural network according to claim 1, characterized in that, Set the virtual control rate of pitch angular velocity q r Then, the derivative of the ballistic tilt tracking error is: Here, e4 represents the pitch angular velocity tracking error.

7. The adaptive inversion control method for variant aircraft based on RBF neural network according to claim 1, characterized in that, Set to: in, The adaptive weights of the RBF neural network for the unknown influence term of pitch angular velocity are defined by the adaptive law of these weights. Set to: σ represents the adaptive law coefficient. After setting the control law, the derivative of the pitch angular velocity tracking error is expressed as:

Images