An Attitude Adaptive Control Method for Fixed-Wing UAVs Based on Model Correction

Abstract

The present invention discloses a fixed-wing UAV attitude adaptive control method based on model modification, belonging to the technical field of UAV flight control. The method includes: constructing a fixed-wing UAV attitude nonlinear model, and using the output information of the reference model and the UAV attitude nonlinear model to design a correction term to correct the reference model; taking the UAV desired attitude angle information, the output information of the reference model and the UAV attitude nonlinear model as the input of the radial basis function neural network to obtain an approximation value of the model uncertainty; designing an extended state observer to obtain an interference estimation value; and designing a UAV attitude controller and a neural network adaptive law based on the obtained approximation value of the model uncertainty and the interference estimation value. The present invention can enable the UAV to asymptotically track the desired reference trajectory and have strong anti-interference ability under the consideration of model uncertainty, control input saturation and wind interference effects.

Description

Technical Field

[0001] The present invention belongs to the technical field of unmanned aerial vehicle (UAV) flight control, and particularly relates to an attitude adaptive control method for a fixed-wing UAV based on model correction. Background Technique

[0002] With the wide application of UAVs, their mission requirements are increasing day by day, and the mission execution environment is becoming more and more complex. Larger maneuvering flight actions need to be completed, and at this time, larger control inputs are required. However, due to its own physical limitations, a UAV cannot provide the required control amount, resulting in the occurrence of control input saturation, thereby reducing the flight control performance of the UAV. In addition, external interference and model uncertainty of the UAV will also reduce the flight control performance of the UAV.

[0003] Therefore, in order to ensure the flight quality of the UAV and complete the flight mission, it is very necessary to design an effective UAV control method to enhance the flight control performance of the UAV in the case of model uncertainty, control input saturation, wind interference, etc. Summary of the Invention

[0004] Aiming at the problems of UAV model uncertainty, control input saturation, wind interference, etc., the present invention proposes an attitude adaptive control method for a fixed-wing UAV based on model correction, which enhances the flight control performance of the UAV in the case of model uncertainty, control input saturation, wind interference, etc., and can asymptotically track the desired reference trajectory and has strong anti-interference ability.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions:

[0006] An attitude adaptive control method for a fixed-wing UAV based on model correction, comprising the following steps:

[0007] Step 1: Considering both model uncertainty and unknown external interference, construct a nonlinear attitude model of the fixed-wing UAV, and use the output information of the reference model and the nonlinear attitude model of the UAV to design a correction term to correct the reference model;

[0008] Step 2: Take the desired attitude angle information of the UAV, the output information of the reference model and the nonlinear attitude model of the UAV as the inputs of the RBF (Radial Basis Function) neural network to obtain an approximation value of the model uncertainty;

[0009] Step 3: Based on the approximation value of the model uncertainty obtained in Step 2, use the output information of the nonlinear attitude model of the UAV and the control input information to design an extended state observer and obtain an estimated value of the interference;

[0010] Step 4: Based on the model uncertainty approximation value and the disturbance estimation value obtained in Step 2 and Step 3, design the UAV attitude controller and the neural network adaptation law.

[0011] Preferably, in Step 1, considering model uncertainty and unknown external disturbances, the fixed-wing UAV attitude nonlinear model is as shown in Equation (1):

[0012]

[0013] where X = [γ θ ψ] T is the attitude angle vector, where γ, θ, and ψ are the roll angle, pitch angle, and yaw angle of the UAV respectively, and f X is the known part in the model, Δf X is the model uncertainty, u is the control input, d is the unknown external disturbance, and g X is the control input gain matrix, and its expression is:

[0014]

[0015] where Q is the dynamic pressure of the free stream, S is the wing area of the UAV, L is the wingspan, b A is the mean aerodynamic chord length of the wing, I x , I y , I z are the moments of inertia, I xy is the product of inertia, is the aileron control surface efficiency, is the rudder control surface efficiency, is the elevator control surface efficiency;

[0016] Considering the UAV attitude nonlinear model in Equation (1), define the tracking error:

[0017]

[0018]

[0019] where is the error between the attitude angle vector and the reference model output vector, is the error between the desired attitude angle vector and the reference model output vector; X d is the desired attitude angle vector, X is the attitude angle vector, and X r is the output vector of the reference model;

[0020] Design the following reference model:

[0021]

[0022] where λ > 0, and λ is the design parameter;

[0023] The error function is obtained from the tracking error and its derivative in equations (3) and (4):

[0024]

[0025]

[0026] where are the tracking error derivatives respectively; ξ, ξ rd are the tracking error error functions respectively; According to the error function ξ, a correction term aξ is designed to correct the reference model shown in equation (5). The corrected reference model is:

[0027]

[0028] where a > 0 and a is a design parameter; when control input saturation occurs, the actual control input requirements cannot be met, resulting in an increase in the tracking error becoming larger. At this time, the correction term aξ also increases accordingly, adjusting the reference model so that the reference model output X r changes, reducing the error with the attitude angle vector X, and the required control input also decreases accordingly, being able to meet the actual control input requirements and making the UAV exit the saturation area; it can be seen from equation (8) that when the tracking error disappears, the error function ξ also disappears, and the corrected reference model (8) will become the original form (5). Therefore, the UAV not only asymptotically tracks the corrected reference model but also asymptotically tracks the original reference model.

[0029] Preferably, in step 2, for the uncertainty Δf X in the UAV attitude nonlinear model, an RBF neural network is designed for approximation. The RBF neural network algorithm is:

[0030]

[0031] Δf X = W *T h(Γ)+ε (10);

[0032] where h j is the output of the j th th neuron in the hidden layer, exp represents the logarithm with base e and the content in the brackets is the exponent, Γ = [Γ1,…,Γ n T is the input vector of the network, c j = [c j1 ,…,c jn is the j​th The center vector of the Gaussian function of a neuron, b j is j th The width of the Gaussian function of a neuron; W * is the ideal weight of the neural network, h = [h1(Γ), …, h m (Γ)] T is the output of the hidden layer of the neural network, and ε is the approximation error;

[0033] Select the input as The output of the neural network is:

[0034]

[0035] In the formula, is the estimated weight of the neural network; is the uncertainty Δf X The approximation value of

[0036] Preferably, in step 3, based on the approximation value of the uncertainty Δf X obtained in step 2, the UAV model formula (1) is written as:

[0037]

[0038] In the formula, includes the approximation error of the neural network and the unknown external disturbance; the approximation error ε of the neural network is estimated by the extended state observer, and at the same time the unknown external disturbance d is also estimated by the extended state observer; based on formula (1) and the neural network output formula (11), the extended state observer is designed as:

[0039]

[0040] In the formula, z1, z2, z3 are the output quantities of the extended state observer, which are the estimated values of X, respectively; β1, β2, β3 are the gains of the extended state observer.

[0041] Preferably, in step 4, based on the approximation value of the model uncertainty obtained in step 2 and step 3 and the disturbance estimated value z3, the UAV attitude controller and the neural network adaptive law are designed, and the controller expression is:

[0042]

[0043] In the formula, K X > 0, K X is the controller gain matrix;

[0044] The RBF neural network adaptive law is:

[0045]

[0046] wherein, G = G T > 0, σ X > 0 are design parameters.

[0047] The beneficial technical effects brought by the present invention:

[0048] The method of the present invention considers the nonlinear model of a fixed-wing UAV with model uncertainty, control input saturation and wind disturbance, improves on the basis of model reference adaptive control, corrects the reference model, and when control input saturation occurs, adjusts the reference model according to the error signal to make the UAV exit the saturation area and solve the problem of control input saturation; combines with neural network and disturbance observer, uses the RBF neural network to approximate the model uncertainty, and on this basis, introduces an extended state observer to estimate the unknown external disturbance and the approximation error of the neural network, and adds the outputs of the neural network and the extended state observer to the controller to eliminate the influence of model uncertainty and unknown external disturbance; the present invention can effectively solve the problem of stable flight control of the UAV under the conditions of model uncertainty, control input saturation and wind disturbance. Brief Description of the Drawings

[0049] Figure 1 It is a principle block diagram of an attitude adaptive control method for a fixed-wing UAV based on model correction. Specific Embodiments

[0050] The present invention will be further described in detail below in conjunction with the drawings and specific embodiments:

[0051] As Figure 1 shown, the attitude adaptive control method for a fixed-wing UAV based on model correction of the present invention specifically includes the following steps:

[0052] Step 1: Considering both model uncertainty and unknown external disturbance, construct a nonlinear model of the attitude of a fixed-wing UAV, and use the output information of the reference model and the nonlinear model of the attitude of the UAV to design a correction term to correct the reference model;

[0053] Considering model uncertainty and unknown external disturbance, the nonlinear model of the attitude of a fixed-wing UAV is described as:

[0054]

[0055] wherein, X = [γ θ ψ] T is the attitude angle vector, where γ, θ, and ψ are the roll angle, pitch angle, and yaw angle of the UAV respectively, f X is the known part in the model, Δf Xis the model uncertainty, u is the control input, d is the unknown external disturbance, and g X is the control input gain matrix, and its expression is:

[0056]

[0057] where Q is the dynamic pressure of the free stream, S is the wing area of the UAV, L is the wingspan, and b A is the mean aerodynamic chord of the wing, and I x 、I y 、I z are the moments of inertia, and I xy is the product of inertia, is the aileron control surface efficiency, is the rudder control surface efficiency, is the elevator control surface efficiency;

[0058] Considering the nonlinear attitude model of the UAV in Equation (1), define the tracking error:

[0059]

[0060]

[0061] where is the error between the attitude angle vector and the reference model output vector, is the error between the desired attitude angle vector and the reference model output vector; X d is the desired attitude angle vector, X is the attitude angle vector, and X r is the output vector of the reference model;

[0062] Design the following reference model:

[0063]

[0064] where λ > 0 is the design parameter;

[0065] From the tracking error and its derivative in Equations (3) and (4), the error function can be obtained:

[0066]

[0067]

[0068] where are the derivatives of the tracking error respectively; According to the error function ξ, design a correction term aξ to correct the reference model shown in Equation (5). The corrected reference model is:

[0069]

[0070] In the formula, a > 0, and a is a design parameter; when control input saturation occurs, the actual control input requirements cannot be met, resulting in a tracking error increasing. At this time, the correction term aξ also increases accordingly, adjusting the reference model to make the output X of the reference model r change, reducing the error with the attitude angle vector X, and the required control input also decreases accordingly, which can meet the actual control input requirements and make the UAV exit the saturation area; it can be seen from Equation (8) that when the tracking error disappears, the error function ξ also disappears, and the corrected reference model (8) will become the original form (5). Therefore, the UAV not only asymptotically tracks the modified reference model but also asymptotically tracks the original reference model.

[0071] Step 2: Take the desired attitude angle information of the UAV, the reference model, and the output information of the UAV attitude nonlinear model as the inputs of the RBF (Radial Basis Function) neural network to obtain an approximation value of the model uncertainty;

[0072] Regarding the uncertainty Δf in the UAV attitude nonlinear model X , design to use an RBF neural network for approximation. The RBF neural network algorithm is:

[0073]

[0074] Δf X = W *T h(Γ)+ε (10);

[0075] In the formula, h j is the output of the j th th neuron in the hidden layer. exp represents the logarithm with e as the base and the content in the parentheses as the exponent. Γ = [Γ1,…,Γ n T is the input vector of the network, c j = [c j1 ,…,c jn is the center vector of the Gaussian basis function of the j th th neuron of the network, b j is the width of the Gaussian basis function of the j th th neuron; W * is the ideal weight of the neural network, h = [h1(Γ),…,h m (Γ)] T is the output of the hidden layer of the neural network, and ε is the approximation error.

[0076] Select the input as The output of the neural network is:

[0077] ​

[0078] In the formula, is the neural network estimation weight; is the uncertainty Δf X approximation value.

[0079] Step 3: Based on the approximation value obtained in Step 2, using the output information and control input information of the UAV attitude nonlinear model, design an extended state observer and obtain the disturbance estimation value;

[0080] Based on the approximation value of the uncertainty Δf X obtained in Step 2, the UAV model formula (1) can be written as:

[0081]

[0082] In the formula, includes the neural network approximation error and unknown external disturbance; the neural network approximation error ε is estimated by the extended state observer, and at the same time the unknown external disturbance d is also estimated by the extended state observer; based on formula (1) and the neural network output formula (11), design the extended state observer as:

[0083]

[0084] In the formula, z1, z2, z3 are the output quantities of the extended state observer, which are the estimated values of X, respectively; β1, β2, β3 are the gains of the extended state observer.

[0085] Step 4: Based on the model uncertainty approximation value and disturbance estimation value obtained in Step 2 and Step 3, design the UAV attitude controller and neural network adaptive law;

[0086] Based on the model uncertainty approximation value obtained in Step 2 and Step 3

[0087]

[0088] In the formula, K X > 0 is the controller gain matrix;

[0089] The RBF neural network adaptive law is:

[0090]

[0091] In the formula, G = G T > 0, σ X > 0 are design parameters.

[0092] In summary, when the control input saturation occurs in the present invention, the reference model is adjusted according to the error signal to make the UAV exit the saturation region; combined with the neural network and the extended state observer, the influence of model uncertainty and unknown external disturbances is eliminated, and the stable operation of the UAV can be effectively guaranteed and it has strong robustness under the influence of UAV model uncertainty, control input saturation and wind disturbance.

[0093] Certainly, the above description is not a limitation of the present invention, and the present invention is not limited to the above examples. Changes, modifications, additions or substitutions made by those skilled in the art within the scope of the essence of the present invention should also fall within the protection scope of the present invention.

Claims

1. An attitude adaptive control method for fixed-wing UAVs based on model correction, characterized in that: It includes the following steps: Step 1: Considering both model uncertainty and unknown external disturbances simultaneously, construct a nonlinear attitude model of a fixed-wing UAV. Using the output information of the reference model and the nonlinear attitude model of the UAV, design a correction term to correct the reference model; Step 2: Take the desired attitude angle information of the UAV, the output information of the reference model and the nonlinear attitude model of the UAV as the inputs of the radial basis function neural network to obtain an approximation of the model uncertainty; Step 3: Based on the approximation of the model uncertainty obtained in Step 2, use the output information and control input information of the nonlinear attitude model of the UAV to design an extended state observer and obtain an estimated value of the disturbance; Step 4: Based on the approximation of the model uncertainty and the estimated value of the disturbance obtained in Step 2 and Step 3, design a UAV attitude controller and a neural network adaptation law; In Step 1, considering model uncertainty and unknown external disturbances, the nonlinear attitude model of the fixed-wing UAV is shown in Equation (1): where \(X = [\gamma\ \theta\ \psi]\) T is the attitude angle vector, where \(\gamma\), \(\theta\), and \(\psi\) are the roll angle, pitch angle, and yaw angle of the UAV, respectively, and \(f\) X is the known part in the model, \(\Delta f\) X is the model uncertainty, \(u\) is the control input, \(d\) is the unknown external disturbance, and \(g\) X is the control input gain matrix, and its expression is: Where Q is the dynamic pressure of the free stream, S is the wing area of the UAV, L is the wingspan, and b A is the mean aerodynamic chord of the wing, and I x , I y , I z are the moments of inertia, and I xy is the product of inertia, is the aileron flap efficiency, is the rudder flap efficiency, is the elevator flap efficiency; Considering the nonlinear attitude model of the UAV in Equation (1), define the tracking error: In the formula, is the error between the attitude angle vector and the output vector of the reference model, is the error between the desired attitude angle vector and the output vector of the reference model; X d is the desired attitude angle vector, X is the attitude angle vector, X r is the output vector of the reference model; Design the following reference model: where λ > 0 and λ is a design parameter; From the tracking error and its derivative in Equations (3) and (4), obtain the error function: wherein, are respectively the tracking error derivatives; ξ, ξ rd are respectively the error functions of the tracking error ; according to the error function ξ, a correction term aξ is designed to correct the reference model shown in Equation (5), and the corrected reference model is: where \(a>0\), and \(a\) is a design parameter; when control input saturation occurs, the actual control input requirements cannot be met, resulting in a tracking error increasing. At this time, the correction term \(a\xi\) also increases accordingly, adjusting the reference model to make the output \(X\) of the reference model r change, reducing the error with the attitude angle vector \(X\). The required control input also decreases accordingly, meeting the actual control input requirements and enabling the UAV to exit the saturation region; it can be seen from Equation (8) that when the tracking error disappears, the error function \(\xi\) also disappears, and the corrected reference model (8) will become the original form (5). Therefore, the UAV not only asymptotically tracks the modified reference model but also asymptotically tracks the original reference model; In step 2, for the uncertainty Δf in the nonlinear model of the UAV attitude X , it is designed to use an RBF neural network for approximation. The RBF neural network algorithm is as follows: Δf X = W *T h(Γ)+ε (10); where h j is the output of the j th neurons in the hidden layer, exp represents the logarithm with base e and the content in the parentheses as the exponent, Γ = [Γ1, …, Γ n T is the input vector of the network, c j = [c j1 , …, c jn is the center vector of the Gaussian basis function of the j th neurons in the network, b j is the width of the Gaussian basis function of the j th neurons; W * is the ideal weight of the neural network, h = [h1(Γ), …, h m (Γ)] T is the output of the hidden layer of the neural network, and ε is the approximation error;​ Select the input as The output of the neural network is: In the formula, is the neural network estimation weight; is the approximation value of the uncertainty Δf X ; In step 3, based on the approximation value of the uncertainty Δf obtained in step 2, the UAV model in Equation (1) is written as: X ​ wherein, includes the neural network approximation error and unknown external disturbances; the neural network approximation error ε is estimated by an extended state observer, and at the same time, the unknown external disturbance d is also estimated by the extended state observer; based on Equation (1) and the neural network output Equation (11), the extended state observer is designed as follows: where z1, z2, and z3 are the outputs of the extended state observer, which are the estimated values of X, respectively; β1, β2, and β3 are the gains of the extended state observer.

2. The attitude adaptive control method for fixed-wing UAVs based on model correction according to claim 1, characterized in that: In step 4, based on the model uncertainty approximation values obtained in steps 2 and 3 and the disturbance estimation value z3, design the UAV attitude controller and the neural network adaptation law. The expression of the controller is as follows: where K X > 0, K X is the controller gain matrix; The RBF neural network adaptation law is: where G = G T > 0, σ X > 0 are design parameters.

Images