Terminal Sliding Mode Control Method for Variable Aircraft Based on Recurrent Neural Network Observer

Abstract

The present invention discloses a terminal sliding mode control method for a morphing aircraft based on a recurrent neural network observer. The method comprises the following steps: Step 1: Construct a morphing aircraft attitude dynamics and kinematics model, and form an attitude control error model by taking the difference with the attitude command; Step 2: Based on the attitude control error model, construct a terminal sliding mode surface; Step 3: Design an observer based on a recurrent neural network to observe the uncertain disturbance term in the morphing aircraft model; Step 4: Based on the terminal sliding mode surface and the output value of the observer, obtain the control quantity of the morphing aircraft. The recurrent neural network observer of the present invention can utilize the measurement information at the current moment and the past moment simultaneously, and the estimation of the disturbance is more accurate. At the same time, the terminal sliding mode surface of the present invention can significantly improve the response speed and is more suitable for objects such as morphing aircraft that need to quickly achieve tracking.

Description

Technical Field

[0001] The present invention belongs to the technical field of aircraft control, and relates to a terminal sliding mode control method for a morphing aircraft, specifically to a terminal sliding mode control method for a morphing aircraft based on a recurrent neural network observer. Background Art

[0002] As a new concept aircraft, a morphing aircraft can broaden the flight envelope and improve flight performance by actively changing its shape. However, the additional disturbances generated during the morphing process seriously affect the attitude stability of the aircraft, and the state of the aircraft changes greatly before and after morphing, and the attitude control system needs to achieve fast tracking. Summary of the Invention

[0003] In order to solve the above problems in the background art, the present invention provides a terminal sliding mode control method for a morphing aircraft based on a recurrent neural network observer.

[0004] The object of the present invention is achieved by the following technical solutions:

[0005] A terminal sliding mode control method for a morphing aircraft based on a recurrent neural network observer includes the following steps:

[0006] Step 1: Construct a morphing aircraft attitude dynamics and kinematics model, and form an attitude control error model by taking the difference with the attitude command;

[0007] Step 2: Based on the attitude control error model, construct a terminal sliding mode surface;

[0008] Step 3: Design an observer based on a recurrent neural network to observe the uncertain disturbance term in the morphing aircraft model;

[0009] Step 4: Based on the terminal sliding mode surface and the output value of the observer, obtain the control quantity of the morphing aircraft.

[0010] Compared with the prior art, the present invention has the following advantages:

[0011] The traditional disturbance estimation method for a morphing aircraft only uses the measurement information at the current moment, while the recurrent neural network observer of the present invention can simultaneously use the measurement information at the current moment and the past moment, and the estimation of the disturbance is more accurate. At the same time, the terminal sliding mode surface of the present invention can significantly improve the response speed and is more suitable for objects such as morphing aircraft that need to achieve fast tracking. Description of the Drawings

[0012] Figure 1 It is a flow chart of a terminal sliding mode control method for a morphing aircraft based on a recurrent neural network observer. Detailed Embodiment

[0013] The technical solution of the present invention will be further described below in conjunction with the accompanying drawings, but it is not limited thereto. Any modification or equivalent replacement of the technical solution of the present invention without departing from the spirit and scope of the technical solution of the present invention shall be covered by the protection scope of the present invention.

[0014] The present invention provides a variable aircraft terminal sliding mode control method based on a recurrent neural network observer, as Figure 1 shown. The method includes the following steps:

[0015] Step 1: Construct the attitude dynamics and kinematics models of the variable aircraft, and form an attitude control error model by taking the difference with the attitude command. The specific steps are as follows:

[0016] Step 1.1: The attitude dynamics and kinematics models of the variable aircraft are:

[0017]

[0018] In the formula: is the first derivative of Ω with respect to time; is the attitude angle vector of the aircraft; is the pitch angle; ψ is the yaw angle; γ is the roll angle; R is the attitude transformation matrix, is the first derivative of ω with respect to time; ω = [ω x , ω y , ω z is the attitude angular velocity vector of the aircraft; ω x is the roll angular velocity; ω y is the yaw angular velocity; ω z is the pitch angular velocity; J represents the moment of inertia matrix of the aircraft; B1 represents the control torque coefficient matrix; δ = [δ x , δ y , δ z represents the control input quantity; δ x is the deflection angle of the aileron; δ y is the deflection angle of the rudder; δ z is the deflection angle of the elevator; M SD is the variable additional torque; M SG is the additional torque generated due to the change of the center of mass during the variable process; M a is the aerodynamic torque; M d is the external disturbance torque.

[0019] Step 1.2: Set the aircraft attitude angle change command:

[0020] Ω c = [Ω cz , Ω cy , Ω cx (2)

[0021] Where: Ω cx is the roll channel command; Ω cy is the yaw channel command; Ω cz is the pitch channel command.

[0022] Define the attitude angle tracking error as:

[0023]

[0024] Where: is the pitch angle tracking error; x 1ψ is the yaw angle tracking error; x 1γ is the roll angle tracking error.

[0025] Taking the second-order derivative of Equation (3) continuously, the attitude control error model can be obtained as:

[0026]

[0027] Where: represents the first-order derivative of x1 with respect to time; x2 represents the first-order derivative of the attitude angle tracking error; represents the second-order derivative of x1 with respect to time; u is the control quantity, and u = RJ -1 B1δ; H is the uncertain disturbance term, and

[0028] Step 2: Based on the attitude control error model, construct the terminal sliding mode surface.

[0029] The terminal sliding mode surface S is defined as:

[0030]

[0031] Where: λ1 and λ2 are positive gain coefficients; q and p are positive odd numbers, and

[0032] Step 3: Design an observer based on the recurrent neural network to observe the uncertain disturbance term in the variant aircraft model. The specific steps are as follows:

[0033] Step 3.1 Use a three-layer recurrent neural network (RNN) to estimate the uncertain disturbance term H in real time. The RNN consists of an input layer, a hidden layer, and an output layer. The RNN mapping is:

[0034]

[0035] Where: x i represents the input quantity of the RNN; y contains the input variables and output variables of the RNN; N is the number of iteration steps; n is the number of nodes in the hidden layer; v k is the connection weight between the output layer and the hidden layer; sik and δ ik are the center and width of the radial basis function respectively; w i is the recurrent weight coefficient of the output layer neurons; Φ k is the threshold of the k-th neuron in the hidden layer, and m is the dimension of the input quantity.

[0036] Step 3.2: To simplify the notation, define all the parameter vectors δ, s, x, and w of the RNN hidden layer, where: δ is the vector composed of all the widths δ of the radial basis functions in the hidden layer ik s is the vector composed of all the centers s of the radial basis functions in the hidden layer ik x is the vector composed of all the input quantities x i w is the vector composed of all the recurrent weight coefficients w of the output layer neurons i in the vector form.

[0037] Express the output layer of the RNN in vector form:

[0038] y(x, δ, s, w, v) = v T Φ(x, δ, s, w) (7)

[0039] where v = [v1 v2…v n T and Φ = [Φ1 Φ2…Φ n T , v represents the weight of the RNN, v i represents the i-th weight, Φ represents the activation function of the RNN, Φ i represents the i-th activation function, i = 1, 2,..., n.

[0040] Step 3.3: Define the RNN observer as:

[0041]

[0042] In the formula: is the estimated value of the RNN network for the disturbance term H; is the estimated value of the radial basis function; is the estimated value of the width of the radial basis function; is the estimated value of the center of the radial basis function; is the estimated value of the recurrent weight coefficient of the output layer neurons.

[0043] Step 3.4: Design the adaptive law of the RNN observer as:

[0044]

[0045] In the formula: η1, η2, η3, η4 are positive gain coefficients in the adaptive law; A1 is​​ The partial derivative matrix of ; A2 is The partial derivative matrix of ; A3 is The partial derivative matrix of .

[0046] Step 4: Based on the terminal sliding mode surface and the output value of the observer, obtain the control quantity of the variable aircraft.

[0047] The control quantity of the variable aircraft is designed as:

[0048]

[0049] where: Λ is a positive number and is the approaching gain.

Claims

1. A terminal sliding mode control method for a variant aircraft based on a recurrent neural network observer, characterized in that The method includes the following steps: Step 1: Construct the attitude dynamics and kinematics model of the variant aircraft, and form an attitude control error model by taking the difference with the attitude command; Step 2: Based on the attitude control error model, construct the terminal sliding mode surface S, and the terminal sliding mode surface S is defined as: where: λ1 and λ2 are positive gain coefficients; q and p are positive odd numbers, and x1 represents the attitude angle tracking error, and x2 represents the first derivative of the attitude angle tracking error; Step 3: Design an observer based on the recurrent neural network to observe the uncertain disturbance terms in the variant aircraft model. The specific steps are as follows: Step 3.1: Use a three-layer RNN to estimate the uncertain disturbance term H in real time. The RNN consists of an input layer, a hidden layer, and an output layer. The RNN mapping is: where: x i represents the input of the RNN; y contains the input and output variables of the RNN; N is the number of iterative steps; n is the number of nodes in the hidden layer; v k is the connection weight between the output layer and the hidden layer; s ik and δ ik are the center and width of the radial basis function respectively; w i is the recursive weight coefficient of the output layer neuron; Φ k is the threshold of the k-th neuron in the hidden layer, and m is the dimension of the input Step 3.2: To simplify the notation, define all the parameter vectors δ, s, x, and w of the RNN hidden layer, where: δ is the vector composed of all the widths δ of the radial basis functions in the hidden layer ik s is the vector composed of all the centers s of the radial basis functions in the hidden layer ik x is the vector composed of all the input quantities x i w is the vector composed of all the recursive weight coefficients w of the output layer neurons i ; Express the output layer of the RNN in the form of a vector: y(x, δ, s, w, v) = v T Φ(x, δ, s, w) where \(v = [v_1\ v_2\ \cdots\ v n \) T and \(\Phi = [\Phi_1\ \Phi_2\ \cdots\ \Phi n \) T , \(v\) represents the weights of the RNN, \(v i \) represents the \(i\)-th weight, \(\Phi\) represents the activation function of the RNN, \(\Phi i \) represents the \(i\)-th activation function, \(i = 1, 2, \cdots, n\); Step 3.3: Define the RNN observer as: Wherein: is the estimated value of the RNN network for the perturbation term H; is the estimated value of the radial basis function; is the estimated value of the width of the radial basis function; is the estimated value of the center of the radial basis function; is the estimated value of the recursive weight coefficient of the output layer neuron; Step 3.4: Design the adaptive law of the RNN observer as: where: η1, η2, η3, η4 are positive gain coefficients in the adaptive law; A1 is the partial derivative matrix of with respect to ; A2 is the partial derivative matrix of with respect to Step 4: Based on the terminal sliding mode surface and the output value of the observer, obtain the control quantity of the variant aircraft.

2. The variant aircraft terminal sliding mode control method based on a recurrent neural network observer according to claim 1, characterized in that The specific steps of Step 1 are as follows: Step 1.1: The attitude dynamics and kinematics model of the variant aircraft is: In the formula: is the first-order derivative of Ω with respect to time; is the attitude angle vector of the aircraft; is the pitch angle; ψ is the yaw angle; γ is the roll angle; R is the attitude transformation matrix, is the first-order derivative of ω with respect to time; ω = [ω x , ω y , ω z is the attitude angular velocity vector of the aircraft; ω x is the roll angular velocity; ω y is the yaw angular velocity; ω z is the pitch angular velocity; J represents the moment of inertia matrix of the aircraft; B1 represents the control moment coefficient matrix; δ = [δ x , δ y , δ z represents the control input quantity; δ x is the deflection angle of the aileron; δ y is the deflection angle of the rudder; δ z is the deflection angle of the elevator; M SD is the additional moment of the variant; M SG is the additional moment generated due to the change of the center of mass during the variant process; M a is the aerodynamic moment; M d is the external disturbance moment; Step 1.2: Set the aircraft attitude angle change command: Ω c = [Ω cz , Ω cy , Ω cx (2) Where: Ω cx is the roll channel command; Ω cy is the yaw channel command; Ω cz is the pitch channel command; Define the attitude angle tracking error as: Wherein: is the pitch angle tracking error; x 1ψ is the yaw angle tracking error; x 1γ is the roll angle tracking error; Taking the second-order derivative of Equation (3) continuously, the attitude control error model can be obtained as: wherein: represents the first derivative of x1 with respect to time; x2 represents the first derivative of the attitude angle tracking error; represents the second derivative of x1 with respect to time; u is the control variable, and u = RJ -1 B1δ; H is the uncertain disturbance term, and 3. The variable aircraft terminal sliding mode control method based on a recurrent neural network observer according to claim 1, characterized in that In Step 4, the control quantity of the variant aircraft is designed as: where: Λ is a positive number and is the approaching gain.

Images