Gliding aircraft intelligent adaptive control method based on dynamic model update driving

Abstract

The invention discloses a gliding aircraft intelligent adaptive control method based on dynamic model update driving, and belongs to the technical field of aircraft control. The method comprises the following steps: constructing an offline gliding aircraft proxy model; designing a gliding aircraft controller; obtaining an actual control instruction; and dynamically updating and adaptively controlling and adjusting the gliding aircraft agent model. According to the method, the gliding aircraft agent model is constructed by adopting the neural network, so that modeling errors are avoided, and a good basis is provided for online dynamic updating of the aircraft model; a sliding mechanism and an incremental learning method based on memory perception synapse are adopted for online dynamic updating of the aircraft agent model, the adaptability of an aircraft control system to the environment and tasks is enhanced, and the precision of the aircraft agent model is improved; based on actual control instruction calculation design of an aircraft model proxy network and nonlinear model interpolation, the adaptive capacity of the aircraft to interference and uncertainty is improved, and online high-precision control of the aircraft is realized.

Description

Technical Field

[0001] The present invention relates to an intelligent adaptive control method for a gliding aircraft driven by dynamic model update, belonging to the technical field of aircraft control. Background Art

[0002] The existing aircraft control system design methods mainly rely on fixed models. However, in practical applications, the dynamic characteristics and models of aircraft will be affected by factors such as environmental disturbances and mission changes, resulting in poor adaptability and decreased control accuracy of traditional control methods. Although the recently emerging intelligent control methods can obtain approximately optimal control strategies through training, they are also designed and trained offline according to specific models, with poor generalization ability and difficulty in applying to complex and variable flight environments.

[0003] Therefore, there is an urgent need to develop an intelligent adaptive control method that can dynamically update the aircraft model online to improve the robustness and adaptability of the aircraft control system. Summary of the Invention

[0004] To solve the problems in the background art, the present invention provides an intelligent adaptive control method for a gliding aircraft driven by dynamic model update.

[0005] To achieve the above object, the present invention adopts the following technical solutions: An intelligent adaptive control method for a gliding aircraft driven by dynamic model update, the method comprising the following steps:

[0006] S1: Offline construction of a gliding aircraft proxy model;

[0007] The S1 includes the following steps:

[0008] S101: Collect historical flight data of the gliding aircraft, and respectively construct identification sample libraries for the aerodynamic moment models of its pitch, yaw, and roll channels;

[0009] S102: Establish a fully connected neural network, and use the sample library to train the proxy network of the gliding aircraft three-channel aerodynamic moment model to form a basic proxy network of the gliding aircraft model:

[0010] The network input is six-dimensional flight state, and the output is one-dimensional aerodynamic moment coefficient;

[0011] Wherein:

[0012] The proxy network model of the pitch channel is as follows:

[0013] C mz = net z (Ma, α, β, δ x , δ y , δ z ) (1)

[0014] The yaw channel proxy network model is as follows:

[0015] C my = net y (Ma, α, β, δ x , δ y , δ z ) (2)

[0016] The roll channel proxy network model is as follows:

[0017] C mx = net x (Ma, α, β, δ x , δ y , δ z ) (3)

[0018] In formulas (1)-(3):

[0019] Ma is the Mach number;

[0020] α is the angle of attack;

[0021] β is the sideslip angle;

[0022] δ x is the aileron deflection angle of the roll channel;

[0023] δ y is the rudder deflection angle of the yaw channel;

[0024] δ z is the rudder deflection angle of the pitch channel;

[0025] C mx is the one-dimensional aerodynamic moment coefficient of the roll channel;

[0026] C my is the one-dimensional aerodynamic moment coefficient of the yaw channel;

[0027] C mz is the one-dimensional aerodynamic moment coefficient of the pitch channel.

[0028] S2: Design of the gliding aircraft controller;

[0029] The said S2 includes the following steps:

[0030] S201: Construct a gliding aircraft model oriented to control. The dynamic model of the gliding aircraft around the center of mass is:

[0031]

[0032] In formula (4):

[0033] Θ = [α β σ] Tis the attitude angle vector, where: σ is the bank angle;

[0034] is the first derivative of the attitude angle vector;

[0035] Q is the attitude angular velocity transformation matrix;

[0036] ω b = [ω x ω y ω z T is the attitude angular velocity vector, where: ω x is the roll angular velocity, ω y is the yaw angular velocity, ω z is the pitch angular velocity;

[0037] is the first derivative of the attitude angular velocity vector;

[0038] Δf F is the model uncertainty term;

[0039] I is the moment of inertia of the aircraft;

[0040] M is the control torque;

[0041] M d is the disturbance torque;

[0042] S202: Define the system state variable as x = [α β σ ω x ω y ω z T , and rewrite Equation (4) into the MIMO system form as follows:

[0043]

[0044] In Equation (5):

[0045] is the first derivative of the system state variable;

[0046] f(x) is the system state nonlinear term;

[0047] g(x) is the system input nonlinear term;

[0048] d is the system uncertainty term;

[0049] S203: Perform feedback linearization on Equation (5), and we can get:

[0050]

[0051] In Equation (6):

[0052] ​​ is the second derivative of the attitude angle vector;

[0053] F is the state matrix after feedback linearization;

[0054] E is the equivalent control matrix;

[0055] is the disturbance quantity;

[0056] S204: Design the state feedback control law:

[0057] M = -E -1 F + Π(7)

[0058] In Equation (7):

[0059] Π = [π1 π2 π3] T is the auxiliary control variable, where: π1 is the auxiliary control variable of the pitch channel, π2 is the auxiliary control variable of the yaw channel, and π3 is the auxiliary control variable of the roll channel;

[0060] Then we can get:

[0061]

[0062] Introduce the virtual control quantity U = EΠ, and we can get:

[0063]

[0064] S205: Decouple the system into three second-order systems, design the non-singular fast terminal sliding mode control law for each of the three channels, and obtain the virtual control quantity U;

[0065] The said S205 includes the following steps:

[0066] S20501: Define the desired control output as where: α d is the desired angle of attack, is the first derivative of the desired angle of attack;

[0067] Then there is the state error vector e α as:

[0068] e α = [e α1 , e α2 T = x d - x(10)

[0069] In Equation (10):

[0070] e α1 is the angle of attack error;

[0071] eα2 is the error of the angle of attack change rate;

[0072] S20502: Select a non-singular fast terminal sliding mode surface:

[0073]

[0074] In formula (11):

[0075] k1 > 0, k2 > 0, p, q, m, and n are all constants and are all taken as odd numbers, satisfying

[0076] S20503: Obtain the first derivative of the sliding mode surface:

[0077]

[0078] In formula (12):

[0079] is the second derivative of the desired angle of attack α d ;

[0080] u1 is the control quantity of the pitch channel;

[0081] S20504: Divide the control quantity into the equivalent control law u eq on the sliding mode surface and the switching control u sw , that is, the control quantity of the pitch channel u1 = u 1eq + u 1sw , let It can be obtained:

[0082]

[0083] S20505: Adopt the idea of a terminal attractor to design the switching control u 1sw :

[0084]

[0085] In formula (14):

[0086] k3 > 0, k4 > 0, k and l are all constants and are all taken as odd numbers, satisfying

[0087] S20506: Substitute formula (14) into formula (12) and combine with formula (13) to obtain:

[0088]

[0089] Then the control quantity of the pitch channel is:

[0090]

[0091] S20507: Similarly, it can be obtained that:

[0092] The control quantity of the yaw channel is:

[0093]

[0094] The control quantity of the roll channel is:

[0095]

[0096] S20508: Obtain the virtual control quantity U = [u1 u2 u3] T .

[0097] S206: Use Equation (7) to obtain the desired control torque:

[0098] M = -E -1 F + E -1 U (19).

[0099] S3: Obtain the actual control command;

[0100] The said S3 includes the following steps:

[0101] S301: According to the current state of the gliding aircraft and the desired control torque M of the pitch channel z , the desired aerodynamic torque coefficient of the pitch channel can be obtained

[0102]

[0103] In Equation (20):

[0104] q is the dynamic pressure;

[0105] S is the reference area;

[0106] L is the reference length;

[0107] S302: Obtain the actual control command of the pitch channel;

[0108] The said S302 includes the following steps:

[0109] S30201: Divide the pitch rudder channel deflection angle into characteristic point intervals according to the capabilities of the gliding aircraft, and input the current Mach number, angle of attack, sideslip angle, and pitch rudder deflection angle sequence into the aerodynamic network proxy model, i.e., Equation (1), to obtain the corresponding desired aerodynamic torque coefficient sequence of the pitch channel

[0110] S30202: Find the maximum and minimum values of the sequence and judge whether the desired aerodynamic torque coefficient is greater than the sequence the maximum value in or less than the sequence the minimum value in

[0111] When the above conditions are met, the rudder deflection angle corresponding to the maximum or minimum value in the sequence is taken as the actual control command

[0112] On the contrary, starting from the initial value of the sequence judge in turn whether the expected aerodynamic moment coefficient is located between two adjacent values in the sequence When the above conditions are met, the expected aerodynamic moment coefficient corresponding rudder deflection angle δ z is obtained by interpolation. When there are multiple rudder deflection angles that meet the conditions, the rudder deflection angle with the smallest absolute deviation from the previous moment's rudder deflection angle is taken as the actual control command

[0113] S303: Similarly, obtain the actual control commands for the yaw channel and the roll channel.

[0114] S4: Dynamic update and adaptive control adjustment of the gliding aircraft agent model.

[0115] The S4 includes the following steps:

[0116] S401: Using the principles of flight mechanics, convert the measured data during the flight of the gliding aircraft into the input and output of the agent network model, accumulate this data, and adopt a sliding mechanism to form a model dynamic update sample set;

[0117] S402: Use the model dynamic update sample set and the MAS incremental learning method to continuously update and adjust the gliding aircraft agent network model online, so that the agent network model is updated dynamically online;

[0118] The S402 includes the following steps:

[0119] S40201: Assume that the parameters of the original agent network model are θ old , and calculate the importance Ω of the network parameters using the gradient of the output with respect to the input;

[0120] S40202: Calculate the network loss function Y:

[0121] Y = Y new + λΩ(θ - θ old ) 2 (21)

[0122] In formula (21):

[0123] Y newIt is the loss function calculated by using the model to dynamically update the sample set;

[0124] λ is a parameter for measuring importance;

[0125] θ is the parameter of the surrogate network model optimized in the current generation;

[0126] S40203: Optimize Equation (21) using the gradient descent method, and the optimization of the surrogate network model parameter θ can be achieved, thereby realizing the dynamic update of the surrogate network model.

[0127] S403: Replace the pneumatic network surrogate model used in S3 with the dynamically updated model in S402 to obtain a more accurate aircraft control command δ * .

[0128] Compared with the prior art, the beneficial effects of the present invention are:

[0129] The present invention constructs a surrogate model of a gliding aircraft using a neural network, avoiding modeling errors and providing a good foundation for the online dynamic update of the aircraft model; adopts a sliding mechanism and an incremental learning method based on memory-aware synapses for the online dynamic update of the aircraft surrogate model, realizes the online update of the aircraft model and the adaptive adjustment of the control system, enhances the adaptability of the aircraft control system to the environment and tasks, and effectively improves the accuracy of the aircraft surrogate model; based on the design of actual control command calculation of the aircraft model surrogate network and non-linear model interpolation, effectively improves the adaptability of the aircraft to interference and uncertainty, and realizes the online high-precision control of the aircraft. Description of the Drawings

[0130] Figure 1 is the flow chart of the present invention. Detailed Embodiments

[0131] Next, the technical solutions in the present invention will be clearly and completely described in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0132] An intelligent adaptive control method for a gliding aircraft driven by dynamic model update, the method comprising the following steps:

[0133] S1: Offline construction of a gliding aircraft surrogate model;

[0134] The above S1 includes collecting the simulation and flight test data of the gliding aircraft to form a sample set, constructing an aircraft surrogate network model, and using the sample set for training to form the basic surrogate network of the aircraft model, which specifically includes the following steps:

[0135] S101: Collect the historical flight data of the gliding aircraft, and respectively construct the identification sample libraries of the aerodynamic moment models of its pitch, yaw, and roll channels;

[0136] S102: Establish a fully connected neural network, and use the Adam algorithm to train the surrogate network of the three-channel aerodynamic moment model of the gliding aircraft using the sample library to form the basic surrogate network of the gliding aircraft model:

[0137] The network input is six-dimensional flight states, including: Mach number, angle of attack, sideslip angle, and the rudder deflections of the pitch, yaw, and roll channels; the output is a one-dimensional aerodynamic moment coefficient;

[0138] Among them:

[0139] The surrogate network model of the pitch channel is as follows:

[0140] C mz = net z (Ma, α, β, δ x , δ y , δ z ) (1)

[0141] The surrogate network model of the yaw channel is as follows:

[0142] C my = net y (Ma, α, β, δ x , δ y , δ z ) (2)

[0143] The surrogate network model of the roll channel is as follows:

[0144] C mx = net x (Ma, α, β, δ x , δ y , δ z ) (3)

[0145] In formulas (1)-(3):

[0146] Ma is the Mach number;

[0147] α is the angle of attack;

[0148] β is the sideslip angle;

[0149] δ x is the rudder deflection of the roll channel;

[0150] δ y is the rudder deflection angle of the yaw channel;

[0151] δ z is the rudder deflection angle of the pitch channel;

[0152] C mx is the one-dimensional aerodynamic moment coefficient of the roll channel;

[0153] C my is the one-dimensional aerodynamic moment coefficient of the yaw channel;

[0154] C mz is the one-dimensional aerodynamic moment coefficient of the pitch channel.

[0155] S2: Design of the gliding vehicle controller;

[0156] The S2 includes constructing a control-oriented model based on the characteristics of the gliding vehicle and designing a non-singular fast terminal sliding mode control law to obtain the required desired control moment, which specifically includes the following steps:

[0157] S201: Construct a control-oriented gliding vehicle model. The dynamic model of the gliding vehicle around the center of mass is:

[0158]

[0159] In Equation (4):

[0160] Θ = [α β σ] T is the attitude angle vector, where: σ is the bank angle;

[0161] is the first derivative of the attitude angle vector;

[0162] Q is the attitude angular velocity transformation matrix;

[0163] ω b = [ω x ω y ω z T is the attitude angular velocity vector, where: ω x is the roll angular velocity, ω y is the yaw angular velocity, ω z is the pitch angular velocity;

[0164] is the first derivative of the attitude angular velocity vector;

[0165] Δf F is the model uncertainty;

[0166] I is the moment of inertia of the aircraft;

[0167] ​M is the control torque;

[0168] M d is the disturbance torque;

[0169] S202: Define the system state variables as x = [α β σω x ω y ω z T , and rewrite Equation (4) into the MIMO system form as follows:

[0170]

[0171] In Equation (5):

[0172] is the first derivative of the system state variables;

[0173] f(x) is the system state nonlinear term;

[0174] g(x) is the system input nonlinear term;

[0175] d is the system uncertainty term;

[0176] S203: Perform feedback linearization on Equation (5), and we can get:

[0177]

[0178] In Equation (6):

[0179] is the second derivative of the attitude angle vector;

[0180] F is the state matrix after feedback linearization;

[0181] E is the equivalent control matrix;

[0182] is the disturbance quantity;

[0183] S204: Design the state feedback control law:

[0184] M = -E -1 F + Π (7)

[0185] In Equation (7):

[0186] Π = [π1 π2 π3] T is the auxiliary control variable, where: π1 is the auxiliary control variable of the pitch channel, π2 is the auxiliary control variable of the yaw channel, and π3 is the auxiliary control variable of the roll channel;

[0187] Then we can get:

[0188] ​

[0189] Introduce the virtual control quantity \(U = E\Pi\), and we can get:

[0190]

[0191] S205: The system is decoupled into three second-order systems, and the nonsingular fast terminal sliding mode control law is designed for each of the three channels to obtain the virtual control quantity \(U\);

[0192] The said S205 includes the following steps:

[0193] S20501: Define the desired control output as where: \(\alpha\) d is the desired angle of attack, and \(\dot{\alpha}\) is the first derivative of the desired angle of attack;

[0194] Then the state error vector \(e\) α is:

[0195] \(e\) α = [\(e\) α1 , \(e\) α2 T = \(x\) d - \(x_{(10)}\)

[0196] In formula (10):

[0197] \(e\) α1 is the angle of attack error;

[0198] \(e\) α2 is the angle of attack rate error;

[0199] S20502: Select the nonsingular fast terminal sliding mode surface:

[0200]

[0201] In formula (11):

[0202] \(k_1>0\), \(k_2>0\), \(p\), \(q\), \(m\) and \(n\) are all constants and all take odd numbers, satisfying

[0203] S20503: Obtain the first derivative of the sliding mode surface:

[0204]

[0205] In formula (12):

[0206] \(\ddot{\alpha}\) is the second derivative of the desired angle of attack \(\alpha\) d ;

[0207] \(u_1\) is the control quantity of the pitch channel;

[0208] S20504: Divide the control quantity into the equivalent control law u on the sliding mode surface eq and the switching control u sw , that is, the control quantity u1 of the pitch channel = u 1eq +u 1sw . Let It can be obtained that:

[0209]

[0210] S20505: Adopt the idea of the terminal attractor to design the switching control u 1sw :

[0211]

[0212] In formula (14):

[0213] k3>0, k4>0, k and l are all constants and are all taken as odd numbers, satisfying

[0214] S20506: Substitute formula (14) into formula (12) and combine with formula (13) to obtain:

[0215]

[0216] Then the control quantity of the pitch channel is:

[0217]

[0218] S20507: Similarly, it can be obtained that:

[0219] The control quantity of the yaw channel is:

[0220]

[0221] The control quantity of the roll channel is:

[0222]

[0223] S20508: Obtain the virtual control quantity U = [u1 u2 u3] T .

[0224] S206: Use formula (7) to obtain the desired control torque:

[0225] M = -E -1 F + E -1 U (19).

[0226] S3: Obtain the actual control instruction;

[0227] S3 includes calculating the corresponding aerodynamic moment coefficient based on the desired control moment obtained from S2, and then obtaining the actual control command (rudder deflection angle) based on the glider vehicle model proxy network and the non-linear model interpolation method; specifically, it includes the following steps:

[0228] S301: According to the current state of the glider vehicle and the desired control moment M of the pitch channel z , the desired aerodynamic moment coefficient of the pitch channel can be obtained

[0229]

[0230] In formula (20):

[0231] q is the dynamic pressure;

[0232] S is the reference area;

[0233] L is the reference length;

[0234] S302: Obtain the actual control command of the pitch channel;

[0235] The said S302 includes the following steps:

[0236] S30201: Divide the pitch rudder channel deflection angle into characteristic point intervals (such as [-20, -10, -5, 0, 5, 10, 20]°) according to the glider vehicle's ability, and input the current Mach number, angle of attack, sideslip angle, and pitch rudder deflection angle sequence into the aerodynamic network proxy model, that is, formula (1), to obtain the corresponding desired aerodynamic moment coefficient sequence of the pitch channel

[0237] S30202: Considering that the non-linearity of the aerodynamic model is relatively strong, and multiple rudder deflection angles can achieve the same aerodynamic moment coefficient. To avoid sudden changes in the calculated rudder deflection angle and affect the control performance, the following non-linear interpolation method is used to solve the rudder deflection angle.

[0238] Find the maximum and minimum values of the sequence , and judge whether the desired aerodynamic moment coefficient is greater than the maximum value in the sequence or less than the minimum value in the sequence ;

[0239] When the above conditions are met, then take the rudder deflection angle corresponding to the maximum or minimum value in the sequence as the actual control command

[0240] Otherwise, start from the starting value of the sequence , and judge in turn whether the desired aerodynamic moment coefficient is located in the sequence Between two adjacent values, when the condition is satisfied, interpolation is used to obtain the desired aerodynamic moment coefficient The corresponding rudder deflection angle δ z When there are multiple rudder deflection angles that satisfy the condition, the rudder deflection angle with the smallest absolute deviation from the previous moment's rudder deflection angle is taken as the actual control command

[0241] S303: Similarly, obtain the actual control commands for the yaw channel and the roll channel

[0242] S4: Dynamic update of the gliding vehicle proxy model and adaptive control adjustment

[0243] The S4 includes collecting online flight data to form a model dynamic update sample set, and using a sliding mechanism and an incremental learning method based on Memory Aware Synapses (MAS) to update the parameters of the gliding vehicle proxy network model, enabling the proxy model to be updated online dynamically, providing an accurate proxy model for S3 to obtain more accurate aircraft control commands. It specifically includes the following steps

[0244] S401: Using the principles of flight mechanics to convert the measurement data during the flight of the gliding vehicle into the input and output of the proxy network model, accumulating this data, and using a sliding mechanism to form a model dynamic update sample set

[0245] S402: Using the model dynamic update sample set and the MAS incremental learning method to continuously update and adjust the gliding vehicle proxy network model online, enabling the proxy network model to be updated online dynamically, so as to more accurately represent the aircraft model. Among them, the MAS incremental learning method has a low computational cost and can meet the online adjustment time requirements

[0246] The S402 includes the following steps

[0247] S40201: Assume that the parameters of the original proxy network model are θ old , and calculate the importance Ω of the network parameters using the gradient of the output with respect to the input (i.e., sensitivity)

[0248] S40202: Calculate the network loss function Y

[0249] Y = Y new + λΩ(θ - θ old ) 2 (21)

[0250] In formula (21):

[0251] Y new is the loss function calculated using the model dynamic update sample set

[0252] λ is a parameter to measure importance

[0253] θ is the parameter of the surrogate network model optimized in the current generation;

[0254] S40203: Optimize Equation (21) using the gradient descent method, and the optimization of the surrogate network model parameter θ can be achieved, thereby realizing the dynamic update of the surrogate network model.

[0255] S403: Replace the pneumatic network surrogate model used in S3 with the dynamically updated model in S402, so as to provide an accurate surrogate model for S3 to obtain more accurate aircraft control commands δ * 。

[0256] For those skilled in the art, it is obvious that the present invention is not limited to the details of the above exemplary embodiments, and the present invention can be implemented in other forms without departing from the spirit or basic characteristics of the present invention. Therefore, from any point of view, the embodiments should be regarded as exemplary and non-limiting. The scope of the present invention is defined by the appended claims rather than the above description. Therefore, all changes falling within the meaning and scope of the equivalent conditions of the claims are intended to be included in the present invention. Any reference signs in the claims should not be regarded as limiting the claims involved.

[0257] In addition, it should be understood that although this specification is described according to embodiments, not every embodiment only contains an independent technical solution. This narrative way of the specification is only for clarity. Those skilled in the art should regard the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. An intelligent adaptive control method for a gliding aircraft based on dynamic model update driving, characterized in that: The method includes the following steps: S1: Construction of an offline gliding aircraft agent model; S2: Design of a gliding aircraft controller; S3: Obtaining actual control commands; S4: Dynamic update of the gliding aircraft agent model and adaptive control adjustment.

2. The intelligent adaptive control method for a gliding aircraft based on dynamic model update drive according to claim 1, characterized in that: The S1 includes the following steps: S101: Collect historical flight data of the gliding aircraft, and respectively construct identification sample libraries for the aerodynamic moment models of its pitch, yaw, and roll channels; S102: Establish a fully connected neural network, and use the sample libraries to train the agent network of the gliding aircraft's three-channel aerodynamic moment model to form a basic agent network of the gliding aircraft model: The network input is a six-dimensional flight state, and the output is a one-dimensional aerodynamic moment coefficient; Where: The pitch channel agent network model is as follows: C mz = net z (Ma, α, β, δ x , δ y , δ z ) (1) The yaw channel agent network model is as follows: C my = net y (Ma, α, β, δ x , δ y , δ z ) (2) The roll channel agent network model is as follows: C mx = net x (Ma, α, β, δ x , δ y , δ z ) (3) In formulas (1)-(3): Ma is the Mach number; α is the angle of attack; β is the sideslip angle; δ x is the rudder deflection angle of the roll channel; δ y is the rudder deflection angle of the yaw channel; δ z is the rudder deflection angle of the pitch channel; C mx is the one-dimensional aerodynamic moment coefficient for the roll channel; C my is the one-dimensional aerodynamic moment coefficient of the yaw channel; C mz is the one-dimensional aerodynamic moment coefficient of the pitch channel.

3. The intelligent adaptive control method for a gliding aircraft based on dynamic model update drive according to claim 2, characterized in that: The S2 includes the following steps: S201: Construct a gliding aircraft model for control. The dynamic model of the gliding aircraft around the center of mass is: In formula (4): Θ = [αβσ] T is the attitude angle vector, where: σ is the bank angle; is the first derivative of the attitude angle vector; Q is the attitude angular velocity transformation matrix; ω b = [ω x ω y ω z T is the attitude angular velocity vector, where: ω x is the roll angular velocity, ω y is the yaw angular velocity, ω z is the pitch angular velocity;​ is the first derivative of the attitude angular velocity vector; Δf F is the model uncertainty; I is the moment of inertia of the aircraft; M is the control moment; M d is the disturbance torque; S202: Define the system state variable as \(x = [\alpha\ \beta\ \sigma\ \omega x \omega y \omega z \ T , and rewrite Equation (4) into the MIMO system form as follows: In formula (5): is the first derivative of the system state variable; f(x) is the system state nonlinear term; g(x) is the system input nonlinear term; d is the system uncertainty term; S203: Perform feedback linearization on formula (5), and the following can be obtained: In formula (6): is the second derivative of the attitude angle vector; F is the state matrix after feedback linearization; E is the equivalent control matrix; is the disturbance quantity; S204: Design a state feedback control law: M = -E -1 F + Π (7) In formula (7): Π = [π1 π2 π3] T is an auxiliary control variable, where: π1 is the auxiliary control variable for the pitch channel, π2 is the auxiliary control variable for the yaw channel, and π3 is the auxiliary control variable for the roll channel; Then the following can be obtained: Introduce the virtual control quantity U = EΠ, and the following can be obtained: S205: Decouple the system into three second-order systems, and design a non-singular fast terminal sliding mode control law for each of the three channels to obtain the virtual control quantity U; S20501: Define the expected control output as where: α d is the expected angle of attack, is the first derivative of the expected angle of attack; Then there is a state error vector e α which is: e α = [e α1 , e α2 T = x d - x(10)​ In formula (10): e α1 is the angle of attack error; e α2 is the angle of attack change rate error; S20502: Select a non-singular fast terminal sliding mode surface: In formula (11): k1 > 0, k2 > 0, p, q, m, and n are all constants and are all odd numbers, satisfying S20503: Obtain the first derivative of the sliding mode surface: In formula (12): is the second derivative of the desired angle of attack α d ; u1 is the control quantity of the pitch channel; S20504: Divide the control quantity into the equivalent control law u on the sliding mode surface eq and the switching control u sw , that is, the control quantity u1 of the pitch channel = u 1eq + u 1sw , let It can be obtained that: S20505: Design the switching control u using the idea of terminal attractor 1sw : In formula (14): k3 > 0, k4 > 0, k and l are both constants and both take odd values, satisfying S20506: Substitute formula (14) into formula (12) and combine with formula (13) to obtain: Then the control quantity of the pitch channel is: S20507: Similarly, it can be obtained that: The control quantity of the yaw channel is: The control quantity of the roll channel is: S20508: Obtain the virtual control quantity U = [u1 u2 u3] T ; S206: Use formula (7) to obtain the desired control moment: M = -E -1 F + E -1 U (19).

4. An intelligent adaptive control method for a gliding aircraft based on dynamic model update driving according to claim 3, characterized in that: The S3 includes the following steps: S301: Based on the current state of the gliding aircraft and the desired control moment M of the pitch channel z , the desired aerodynamic moment coefficient of the pitch channel can be obtained In formula (20): q is the dynamic pressure; S is the reference area; L is the reference length; S302: Obtain the actual control command of the pitch channel; S303: Similarly, obtain the actual control commands of the yaw channel and the roll channel.

5. A method for intelligent adaptive control of a gliding aircraft based on dynamic model update driving, as claimed in claim 4, wherein: The S302 includes the following steps: S30201: Divide the pitch rudder channel deflection angle into characteristic point intervals according to the capabilities of the gliding aircraft, and input the current Mach number, angle of attack, sideslip angle, and pitch rudder deflection angle sequence into the aerodynamic network proxy model, i.e., Equation (1), to obtain the corresponding expected aerodynamic moment coefficient sequence of the pitch channel S30202: Search for sequence to find the maximum and minimum values, and determine whether the expected aerodynamic moment coefficient is greater than the maximum value in the sequence or less than the minimum value in the sequence , When the above conditions are met, the rudder deflection angle corresponding to the maximum or minimum value in the sequence is taken as the actual control command Conversely, starting from the starting value of the sequence judge in sequence whether the expected aerodynamic moment coefficient is located between two adjacent values in the sequence When the above conditions are met, the rudder deflection angle δ corresponding to the expected aerodynamic moment coefficient is obtained by interpolation z When there are multiple rudder deflection angles that meet the conditions, the rudder deflection angle with the smallest absolute value of deviation from the previous moment's rudder deflection angle is taken as the actual control command 6. The intelligent adaptive control method of a gliding aircraft based on dynamic model update drive according to claim 5, wherein: The S4 includes the following steps: S401: Use the principles of flight mechanics to convert the measured data during the flight of the gliding aircraft into the input and output of the agent network model, accumulate this data, and adopt a sliding mechanism to form a model dynamic update sample set; S402: Use the model dynamic update sample set and the MAS incremental learning method to continuously update and adjust the gliding aircraft agent network model online, so that the agent network model is updated dynamically online; S403: Replace the pneumatic network proxy model used in S3 with the dynamically updated model in S402 to obtain a more accurate aircraft control command δ * .

7. The intelligent adaptive control method for a gliding aircraft based on dynamic model update drive according to claim 6, wherein: The S402 includes the following steps: S40201: Assume that the parameters of the original proxy network model are θ old , and calculate the importance Ω of the network parameters using the gradient of the output with respect to the input; S40202: Calculate the network loss function Y: Y = Y new + λΩ(θ - θ old ) 2 (21) In formula (21): Y new is the loss function calculated by using the model to dynamically update the sample set; λ is a parameter for measuring importance; θ is the parameter of the surrogate network model optimized in the current generation; S40203: Optimize formula (21) using the gradient descent method, and the optimization of the parameter θ of the surrogate network model can be achieved, thereby realizing the dynamic update of the surrogate network model.

Images