Method and system for optimizing fractional order rbf network controller of hypersonic vehicle

Abstract

The application provides a hypersonic vehicle fractional order RBF network controller optimization method and system, which comprises the following steps: constructing a longitudinal channel dynamics model according to modeling information of a hypersonic vehicle, constructing a fractional order RBF network controller according to the longitudinal channel dynamics model, and interacting a pre-constructed intelligent agent with a hypersonic vehicle flight simulation environment constructed according to the fractional order RBF network controller to obtain a controller parameter setting intelligent agent; inputting real-time flight states of the hypersonic vehicle into the controller parameter setting intelligent agent to perform parameter setting optimization, obtaining optimal controller parameters, inputting the optimal controller parameters into the fractional order RBF network controller to generate flight control instructions, and performing flight control on the hypersonic vehicle according to the flight control instructions, so that the control precision of the controller is improved, the difficulty of controller parameter setting optimization is effectively reduced, and rapid and stable control of the hypersonic vehicle is realized.

Description

Technical Field

[0001] This invention relates to the field of hypersonic vehicle control technology, and in particular to an optimization method and system for a fractional-order RBF network controller for hypersonic vehicles based on deep reinforcement learning. Background Technology

[0002] Hypersonic vehicles exhibit large variations in Mach number and angle of attack during flight and are susceptible to external interference, making it difficult for traditional control methods to achieve satisfactory performance. Furthermore, hypersonic vehicles typically employ controllers with fixed parameters, which often underperform in the face of significant external disturbances and rapidly changing flight environments, potentially leading to system instability and even disaster. Moreover, hypersonic vehicle controllers are often highly complex, with numerous parameters, making parameter tuning extremely challenging, particularly determining the optimal parameter values.

[0003] Therefore, there is an urgent need to provide a controller optimization method for hypersonic vehicles that can adapt to complex environmental changes, so as to improve the control accuracy of the controller and thus improve the steady state and reliability of the hypersonic vehicle during flight. Summary of the Invention

[0004] The purpose of this invention is to provide an optimization method for a fractional-order RBF network controller for hypersonic vehicles. By employing deep reinforcement learning optimization techniques to optimize the control parameters of a fractional-order RBF neural network controller designed based on fractional-order calculus theory, optimal control of hypersonic vehicles can be achieved. This effectively solves the application defects of existing hypersonic vehicle controllers, effectively improves the control accuracy of the controller, and also effectively reduces the difficulty of controller parameter tuning and optimization, thus realizing rapid and stable control of hypersonic vehicles.

[0005] To achieve the above objectives, it is necessary to provide an optimization method and system for a fractional-order RBF network controller for hypersonic vehicles, addressing the aforementioned technical problems.

[0006] In a first aspect, embodiments of the present invention provide an optimization method for a fractional-order RBF network controller of a hypersonic vehicle, the method comprising the following steps:

[0007] Based on the modeling information of the hypersonic vehicle, a longitudinal channel dynamics model of the hypersonic vehicle is constructed; the modeling information includes the vehicle configuration; the longitudinal channel dynamics model includes a velocity subsystem and an altitude subsystem.

[0008] Based on the longitudinal channel dynamics model, a fractional-order RBF network controller is constructed, and based on the fractional-order RBF network controller and the longitudinal channel dynamics model, a hypersonic vehicle flight simulation environment is constructed.

[0009] The pre-built intelligent agent is interactively trained with the hypersonic vehicle flight simulation environment to obtain the controller parameter tuning intelligent agent;

[0010] The real-time flight status of the hypersonic vehicle is obtained, and the real-time flight status is input into the controller parameter tuning agent for parameter tuning and optimization to obtain the optimal controller parameters.

[0011] The optimal controller parameters are input into the fractional-order RBF network controller to generate corresponding flight control commands, and the hypersonic vehicle is controlled to fly according to the flight control commands.

[0012] Furthermore, the aircraft configuration is symmetrical; the longitudinal channel dynamics model is expressed as:

[0013]

[0014] Among them, V and Let h and represent the flight velocity and corresponding first derivative of the hypersonic vehicle, respectively; Represent the flight altitude of the hypersonic vehicle and the corresponding first derivative, respectively; γ and Let α and β represent the trajectory inclination angle and the corresponding first derivative of the hypersonic vehicle, respectively; Let Q and Q represent the angle of attack and the corresponding first derivative of the hypersonic vehicle, respectively; Represent the trajectory tilt angle, pitch rate, and corresponding first derivative of the hypersonic vehicle, respectively; η i , and Let m and I represent the elastic state of the hypersonic vehicle, and their corresponding first and second derivatives, respectively; yy and M yy ζ represents the mass, moment of inertia, and pitching moment of a hypersonic vehicle, respectively; T, D, and L represent thrust, drag, and lift, respectively; i ω i and N i Let represent the i-th damping ratio, natural frequency, and generalized elastic force, respectively; g represents gravitational acceleration.

[0015] Furthermore, the fractional-order RBF network controller includes a velocity subsystem controller and an altitude subsystem controller;

[0016] The step of constructing a fractional-order RBF network controller based on the longitudinal channel dynamics model includes:

[0017] Based on fractional calculus theory, a fractional RBF neural network is pre-constructed; the weights of the fractional RBF neural network are updated using the fractional gradient descent method.

[0018] The fractional-order RBF neural network is used to approximate the nonlinear functions of speed tracking, tilt tracking, angle of attack tracking, and pitch rate tracking, respectively, to obtain the corresponding approximate nonlinear functions of speed tracking, tilt tracking, angle of attack tracking, and pitch rate tracking.

[0019] Based on the approximate nonlinear function for speed tracking, the control model of the speed subsystem controller is obtained; the control model of the speed subsystem controller is expressed as:

[0020]

[0021] In the formula,

[0022] z V =e V -χ V

[0023] x V =V

[0024] Where, φ c Indicates throttle opening or air intake volume; And Δ represents a constant; e V and z V These represent the velocity error and the corresponding tracking error, respectively; V represents the aircraft's flight speed; sig(·) represents the sign function; V d and These represent the reference command velocity and its corresponding first-order reciprocal, respectively. This represents an approximate nonlinear function for speed tracking;

[0025] Based on the approximate nonlinear functions for tilt tracking, angle-of-attack tracking, and pitch angular velocity tracking, the control model of the altitude subsystem controller is obtained; the control model of the altitude subsystem controller is expressed as follows:

[0026]

[0027] In the formula,

[0028] z Q =e Q -χ Q

[0029] χ Q =[V,h,γ,α,Q]

[0030] Where, γ c Indicates the track inclination control value; and h represents a constant. d and These represent the reference flight altitude and its corresponding first-order reciprocal, respectively; V represents the flight speed; e h Indicates the tracking error; α c Indicates the angle of attack control amount; and γ represents a constant; d and These represent the reference track inclination angle and its corresponding first-order reciprocal, respectively; e γ Indicates the tracking error of the trajectory tilt angle; Q represents an approximate nonlinear function for tilt tracking. c Indicates the pitch angular velocity control value; and e represents a constant; α Indicates the angle-of-attack tracking error; α d and These represent the reference angle of attack and its corresponding first-order reciprocal, respectively. This represents an approximate nonlinear function for angle-of-attack tracking; δ e Indicates the elevator deflection angle; And Δ represents a constant; e Q and z Q These represent the pitch angular velocity error and the corresponding tracking error, respectively; Q d and These represent the reference pitch angular velocity and its corresponding first-order reciprocal, respectively. represents the pitch angular velocity tracking approximate nonlinear function; sig(·) represents the sign function.

[0031] Furthermore, the weight update of the fractional-order RBF neural network is expressed as:

[0032]

[0033] Where E represents the network approximation error index; and These represent the update values ​​of the i-th weight in the (t-1), t, and t+1 iterations, respectively; τ represents the network approximation error index E for the i-th weight in the t-th iteration. The fractional order of the derivative; β represents the learning rate; α represents the momentum factor; n represents the number of weights.

[0034] Furthermore, the step of constructing a hypersonic vehicle flight simulation environment based on the fractional-order RBF network controller and the longitudinal channel dynamics model includes:

[0035] Acquire flight data of the hypersonic vehicle; the flight data includes initial state information and initial control parameter information;

[0036] Based on the initial state information, the state information set of the longitudinal channel dynamics model is initialized, and based on the initial control parameter information, the control parameter set of the fractional-order RBF network controller is initialized.

[0037] Based on the longitudinal channel dynamics model and the fractional-order RBF network controller, a flight simulation environment for the hypersonic vehicle is constructed.

[0038] Furthermore, the intelligent agent includes value function-based neural networks and policy-based neural networks;

[0039] The step of interactively training a pre-built intelligent agent with the hypersonic vehicle flight simulation environment to obtain a controller parameter tuning intelligent agent includes:

[0040] Based on the state information set and the control parameter set, select the initial state to be simulated and the control parameters to be simulated;

[0041] Based on the initial state to be simulated and the control parameters to be simulated, the interaction simulation between the agent and the hypersonic vehicle flight simulation environment is executed. When the preset simulation termination condition is reached, the simulation flight of the current round is terminated. The cumulative reward value of each state point in the current simulation flight trajectory is evaluated according to the preset reward function, and the parameters of the value function-based neural network are updated according to the cumulative reward value.

[0042] Based on the updated value function-based neural network, the expected cumulative reward value of each state point in the current simulated flight trajectory is predicted, and the parameters of the policy-based neural network are updated based on the cumulative reward value and the expected cumulative reward value.

[0043] Determine whether the policy-based neural network has reached the preset convergence condition. If it has, stop the simulation training and obtain the controller parameter tuning agent. Otherwise, reselect the initial state to be simulated and the control parameters to be simulated based on the state information set and the control parameter set, and start the next round of interactive simulation training.

[0044] Furthermore, the preset reward function is expressed as:

[0045] r t (s j ,a j ) = r t1 (s j ,a j )+rt2 (s j ,a j )+r t3 (s j ,a j )

[0046] In the formula,

[0047]

[0048]

[0049]

[0050] Where, r t (s j ,a j ), r t1 (s j ,a j ), r t2 (s j ,a j ) and r t3 (s j ,a j ) represent the flight state s at time t, respectively. j Lower control parameter a j The corresponding total reward value, flight status reward function, track tilt angle reward function, and pitch rate reward function; z Q (s j ,a j )and These represent flight states s and s respectively. j Lower control parameter a j The pitch angular velocity tracking error and its corresponding first derivative; z V (s j ,a j )and These represent the flight speed tracking error and its corresponding first derivative, respectively; e h (s j ,a j )and These represent the flight altitude tracking error and its corresponding first derivative, respectively; e γ (s j ,a j )and These represent the tracking error of the trajectory inclination angle and its corresponding first derivative, respectively; e α (s j ,a j )and r1, r2, and r3 represent the angle-of-attack tracking error and its corresponding first derivative, respectively; r1, r2, and r3 represent constants; δ max Indicates the elevator deflection angle δe (s j ,a j The upper limit of ); e γmax This represents the maximum absolute value of the track inclination error; This represents the maximum absolute value of the pitch angular velocity.

[0051] In a second aspect, embodiments of the present invention provide a fractional-order RBF network controller optimization system for hypersonic vehicles, the system comprising:

[0052] The model building module is used to construct a longitudinal channel dynamics model of the hypersonic vehicle based on the modeling information of the hypersonic vehicle; the modeling information includes the vehicle configuration; the longitudinal channel dynamics model includes a velocity subsystem and an altitude subsystem.

[0053] The simulation building module is used to construct a fractional-order RBF network controller based on the longitudinal channel dynamics model, and to construct a hypersonic vehicle flight simulation environment based on the fractional-order RBF network controller and the longitudinal channel dynamics model.

[0054] The strategy training module is used to interactively train a pre-built agent with the hypersonic vehicle flight simulation environment to obtain a controller parameter tuning agent.

[0055] The parameter tuning module is used to acquire the real-time flight status of the hypersonic vehicle and input the real-time flight status into the controller parameter tuning agent for parameter tuning optimization to obtain the optimal controller parameters.

[0056] The control optimization module is used to input the optimal controller parameters into the fractional-order RBF network controller, generate corresponding flight control commands, and perform flight control on the hypersonic vehicle according to the flight control commands.

[0057] Thirdly, embodiments of the present invention also provide a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described method.

[0058] Fourthly, embodiments of the present invention also provide a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the above-described method.

[0059] This application provides a method and system for optimizing a fractional-order RBF network controller for a hypersonic vehicle. The method involves constructing a longitudinal channel dynamics model based on the modeling information of the hypersonic vehicle, building a fractional-order RBF network controller based on the longitudinal channel dynamics model, interactively training the constructed hypersonic vehicle flight simulation environment with a pre-built intelligent agent to obtain a controller parameter tuning agent, inputting the acquired real-time flight state of the hypersonic vehicle into the controller parameter tuning agent for parameter tuning optimization to obtain optimal controller parameters, inputting the optimal controller parameters into the fractional-order RBF network controller to generate flight control commands, and performing flight control on the hypersonic vehicle based on the flight control commands. Compared with existing technologies, this method for optimizing a fractional-order RBF network controller for hypersonic vehicles achieves optimal control of the hypersonic vehicle by combining deep reinforcement learning optimization technology and a fractional-order RBF neural network controller. This effectively improves the control accuracy of the controller while reducing the difficulty of controller parameter tuning optimization, achieving rapid and stable control of the hypersonic vehicle, and has high application value. Attached Figure Description

[0060] Figure 1 This is a flowchart illustrating the optimization method for a fractional-order RBF network controller of a hypersonic vehicle in an embodiment of the present invention.

[0061] Figure 2 This is a schematic diagram of a fractional-order RBF neural network controller in an embodiment of the present invention;

[0062] Figure 3 This is a schematic diagram illustrating the interaction between the intelligent agent and the hypersonic vehicle flight simulation environment in an embodiment of the present invention;

[0063] Figure 4 This is a schematic diagram of the structure of the intelligent agent in an embodiment of the present invention;

[0064] Figure 5 This is a schematic diagram of the process of interactive training between the intelligent agent and the hypersonic vehicle flight simulation environment in an embodiment of the present invention;

[0065] Figure 6 This is a schematic diagram of the structure of the fractional-order RBF network controller optimization system for hypersonic vehicles in an embodiment of the present invention;

[0066] Figure 7 This is an internal structural diagram of the computer device in an embodiment of the present invention. Detailed Implementation

[0067] To make the objectives, technical solutions, and beneficial effects of this application clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Obviously, the embodiments described below are only part of the embodiments of the present invention and are used to illustrate the present invention, but are not intended to limit the scope of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0068] The fractional-order RBF network controller optimization method for hypersonic vehicles provided by this invention can be applied to the optimization control of symmetrically configured hypersonic vehicles. This enables the hypersonic vehicle to adapt to complex environmental changes while effectively improving the controller's control accuracy, thereby enhancing the steady-state performance and reliability of the hypersonic vehicle during flight. It has high application value. The following embodiments will provide a detailed description of the fractional-order RBF network controller optimization method for hypersonic vehicles of this invention.

[0069] In one embodiment, such as Figure 1 As shown, a method for optimizing a fractional-order RBF network controller for a hypersonic vehicle is provided, including the following steps:

[0070] S11. Based on the modeling information of the hypersonic vehicle, construct the longitudinal channel dynamics model of the hypersonic vehicle; wherein, the modeling information can be understood as the relevant assumptions for model construction, including the vehicle configuration, etc. The specific construction process of the longitudinal channel dynamics model of the hypersonic vehicle is as follows:

[0071] Assuming the aircraft configuration is symmetrical, then I xy =I yz =I xz =0, I xy I yz and I xz Let represent the product of inertia when the aircraft is longitudinally symmetrical; we assume that the position of the center of mass does not change, and that both the center of mass and the reference moment are on the axisymmetric plane of the aircraft; correspondingly, the structural parameters of the hypersonic aircraft are shown in Table 1:

[0072] <![CDATA[I yy ]]> <![CDATA[6.779×10 5 kg·m 2 ]]> S 1.5m2

[0073] Based on the above assumptions, the longitudinal channel dynamics model of a hypersonic vehicle can be obtained as follows:

[0074]

[0075] Among them, V and Let h and represent the flight velocity and corresponding first derivative of the hypersonic vehicle, respectively; Represent the flight altitude of the hypersonic vehicle and the corresponding first derivative, respectively; γ and Let α and β represent the trajectory inclination angle and the corresponding first derivative of the hypersonic vehicle, respectively; Let Q and Q represent the angle of attack and the corresponding first derivative of the hypersonic vehicle, respectively; Represent the trajectory tilt angle, pitch rate, and corresponding first derivative of the hypersonic vehicle, respectively; η i , and Let m and I represent the elastic state of the hypersonic vehicle, and their corresponding first and second derivatives, respectively; yy and M yy ζ represents the mass, moment of inertia, and pitching moment of a hypersonic vehicle, respectively; T, D, and L represent thrust, drag, and lift, respectively; i ω i and N i Let represent the i-th damping ratio, natural frequency, and generalized elastic force, respectively; g represents gravitational acceleration.

[0076] In the above formula (1), T, D, L and M yy The corresponding calculation formula is shown in equation (2):

[0077]

[0078] in, There are 6 elastic states; q is the dynamic pressure, and the calculation formula is q = 0.5ρV. 2 ρ is air density; S is the aircraft reference area (maximum cross-section); φ is fuel equivalence ratio; δ e Elevator deflection angle; These are aerodynamic parameters, expressed as:

[0079]

[0080] Define the state variables as x = [V, γ, α, β, h] T The control input is u = [φ, δ] e ] T The system output is y = [V, h]. T Considering that hypersonic vehicles typically need to maintain a fixed altitude and speed during the cruise phase, the longitudinal model of a hypersonic vehicle is divided into a velocity subsystem and an altitude subsystem:

[0081] (1) Velocity subsystem:

[0082]

[0083] (2) Height Subsystem:

[0084]

[0085] In the formula,

[0086]

[0087] Among them, F V F γ F α and F Q These represent the nonlinear functions for velocity tracking, tilt tracking, angle of attack tracking, and pitch velocity tracking, respectively. They are affected by changes in aerodynamic parameters and external disturbances, and include system uncertainties and unknown external disturbances.

[0088] S12. Based on the longitudinal channel dynamics model, construct a fractional-order RBF network controller, and based on the fractional-order RBF network controller and the longitudinal channel dynamics model, construct a hypersonic vehicle flight simulation environment; wherein, the fractional-order RBF network controller can be understood as a controller designed based on the longitudinal channel dynamics model shown in equation (1), considering the physical constraints of the actuator, and using fractional-order calculus theory to track and control the speed and altitude of the hypersonic vehicle, which corresponds to the longitudinal channel dynamics model and includes a speed subsystem controller and an altitude subsystem controller;

[0089] Specifically, the step of constructing a fractional-order RBF network controller based on the longitudinal channel dynamics model includes:

[0090] Based on fractional calculus theory, a fractional RBF neural network is pre-constructed; where the fractional RBF neural network is as follows: Figure 2 The diagram shows three neural network layers: an input layer, a hidden layer, and an output layer. The activation function of the neurons in the hidden layer is shown in equation (5).

[0091]

[0092] Among them, b k Let represent the Gaussian function width of the k-th hidden layer, and b = [b1, b2, ..., bk]. n ]; m is the number of nodes in the hidden layer; c k It is the center vector of the kth hidden layer.

[0093] The output of the output layer of the fractional-order RBF neural network is shown in equation (6):

[0094] y = W T h(x) (6)

[0095] Where W = [W1, W2, ..., W n[] represents the weights; h is the Gaussian function, h = [h1, h2, ..., h2] m ]∈R m x is the input vector, and x = [x1, x2, ..., xn]. n ]∈R n .

[0096] The fractional-order RBF neural network is used to approximate the nonlinear functions of speed tracking, tilt tracking, angle of attack tracking, and pitch rate tracking, respectively, to obtain the corresponding approximate nonlinear functions of speed tracking, tilt tracking, angle of attack tracking, and pitch rate tracking.

[0097] It should be noted that, in practical applications, the above fractional-order RBF neural network approximates the nonlinear function f using equation (6), resulting in the nonlinear approximation function shown in equation (7):

[0098] f = W *T h(x)+ε,|ε|<ε upper (7)

[0099] in, This represents the optimal weights that minimize the approximation error ε; ε upper To approximate the upper bound of the error;

[0100] When there are enough hidden layer nodes, equation (7) will necessarily hold. However, increasing the number of hidden nodes will increase the complexity of the neural network structure, which is not conducive to real-time control. For hypersonic vehicle control systems with high real-time requirements, the existing integer-order RBF neural network cannot be directly applied and needs to be improved to meet the real-time requirements. At the same time, in order to further solve the problem of increased neural network complexity and reduced real-time performance due to the increase of training samples and hidden nodes, the fractional-order controller designed based on fractional-order calculus theory adopted in this embodiment can not only meet the requirements of nonlinear function approximation, but also has greater freedom and flexibility, faster response speed, stronger robustness and stability, which can improve the system's response speed, control accuracy and robustness.

[0101] In this embodiment, the controller approximates the nonlinear functions of speed tracking, tilt tracking, angle of attack tracking, and pitch angular velocity tracking to obtain the approximation result shown in equation (8):

[0102]

[0103] In the formula,

[0104]

[0105] in, h V ε V and ε V These represent the optimal weights, Gaussian function, approximation error, and upper bound of the approximation error for the speed tracking approximate nonlinear function, respectively. h γ ε γ and ε γ These represent the optimal weights, Gaussian function, approximation error, and upper bound of the approximation error for the tilt tracking approximation nonlinear function, respectively. h α ε α and ε α These represent the optimal weights, Gaussian function, approximation error, and upper bound of the approximation error for the angle-of-attack tracking approximate nonlinear function, respectively. h Q ε Q and ε Q These represent the optimal weights, Gaussian function, approximation error, and upper bound of the approximation error for the pitch angular velocity tracking approximate nonlinear function, respectively.

[0106] The optimal weights obtained from each of the above approximations are all updated using the fractional gradient descent method. The corresponding fractional RBF neural network uses fractional calculus theory to update W. * Online updates can be represented as:

[0107]

[0108] Where E represents the network approximation error index; and These represent the update values ​​of the i-th weight in the (t-1), t, and t+1 iterations, respectively; τ represents the network approximation error index E for the i-th weight in the t-th iteration. The fractional order of the derivative, and τ∈(0,1); β represents the learning rate, and β∈(0,1); α represents the momentum factor, and α∈(0,1); n represents the number of weights.

[0109] In this embodiment, the fractional gradient descent method of Equation (9) is used to update the weights of the RBF network, which can effectively improve the response speed and system robustness, thereby improving the response speed and control accuracy of the controller.

[0110] Based on the approximate nonlinear function for speed tracking, the control model of the speed subsystem controller is obtained; wherein, the control model of the speed subsystem controller can be understood as the control model of the speed subsystem controller, and can be expressed as:

[0111]

[0112] In the formula,

[0113] z V =e V -χ V

[0114] x V =V

[0115] Where, φ c Indicates throttle opening or air intake volume; And Δ represents a constant, and k V1 k V2 >0; e V and z V These represent the velocity error and the corresponding tracking error, respectively; V represents the aircraft's flight speed; sig(·) represents the sign function; V d and These represent the reference command velocity and its corresponding first-order reciprocal, respectively. This represents an approximate nonlinear function for speed tracking;

[0116] The specific construction process of the control model for the speed subsystem controller is as follows:

[0117] (1) Define the reference command speed as V d And consider the actual constraints shown in equation (11):

[0118]

[0119] in, φ and These represent the upper and lower bounds of the fuel equivalence ratio, respectively; e V For speed error;

[0120] (2) Introducing the compensation system shown in equation (12):

[0121]

[0122] (3) Define the tracking error z V for:

[0123] z V =e V -χ V

[0124] (4) The control law shown in formula (10) is used to control the throttle opening or intake volume:

[0125]

[0126] in, yes |ε V |≤εV Approximate;

[0127] (5) Approximate function weight The update law is:

[0128]

[0129] Among them, E V F represents V The approximation error index; τ V For E V right Find the fractional order of the derivative and τ V ∈(0,1); and Let β represent the updated values ​​of the weights in the (t-1)th, tth, and (t+1)th iterations, respectively; V This represents the learning rate, and β V ∈(0,1); α V Denotes the momentum factor, and α V ∈(0,1);

[0130] Based on the approximate nonlinear function for tilt tracking, the approximate nonlinear function for angle of attack tracking, and the approximate nonlinear function for pitch angular velocity tracking, the control model of the altitude subsystem controller is obtained; wherein, the control model of the altitude subsystem controller can be understood as the control model of the altitude subsystem controller, as shown in equation (14):

[0131]

[0132] In the formula,

[0133] z Q =e Q -χ Q

[0134] χ Q =[V,h,γ,α,Q]

[0135] Where, γ c Indicates the track inclination control value; and Denotes a constant, and k h1 >0, k h2 >0; h d and These represent the reference flight altitude and its corresponding first-order reciprocal, respectively; V represents the flight speed; e h Indicates the tracking error; α c Indicates the angle of attack control amount; and γ represents a constant; d and These represent the reference track inclination angle and its corresponding first-order reciprocal, respectively; e γ Indicates the tracking error of the trajectory tilt angle; Q represents an approximate nonlinear function for tilt tracking. c Indicates the pitch angular velocity control value; and e represents a constant; α Indicates the angle-of-attack tracking error; α d and These represent the reference angle of attack and its corresponding first-order reciprocal, respectively. This represents an approximate nonlinear function for angle-of-attack tracking; δ e Indicates the elevator deflection angle; And Δ represents a constant; e Q and z Q These represent the pitch angular velocity error and the corresponding tracking error, respectively; Q d and These represent the reference pitch angular velocity and its corresponding first-order reciprocal, respectively. represents an approximate nonlinear function for pitch angular velocity tracking; sig(·) represents the sign function;

[0136] The specific construction process of the control model for the altitude subsystem controller is as follows:

[0137] (1) Define the reference flight altitude h d Design the following control law to make h quickly track h. d :

[0138]

[0139] (2) Define a second-order tracking differentiator and solve for γ. c First derivative:

[0140]

[0141] Among them, l γ , λ γ ξ is a positive constant to be designed; γ1 and ξ γ2 They are γ d and Estimated value.

[0142] (3) Define the reference track inclination angle γ d Design the following control law to achieve control over γ d Rapid tracking:

[0143]

[0144] in, yes |ε γ |≤ε γ Approximate;

[0145] (4) Approximate function weight The update law is shown in equation (15):

[0146]

[0147] Among them, E γ For F γ The approximation error index; τ γ For E γ right Find the fractional order of the derivative and τ γ ∈(0,1); β γ This represents the learning rate, and β γ ∈(0,1); α γ Denotes the momentum factor, and α γ ∈(0,1).

[0148] (5) Define the reference angle of attack α d Design the following control law to achieve α d Quick Tracking:

[0149]

[0150] in, yes |ε α |≤ε α Approximate;

[0151] (6) The following differentiator is used to achieve the differentiability of α. d First-order derivative:

[0152]

[0153] Among them, l α , λ α ξ is a positive constant to be designed. α1 and ξ α2 They are α d and Estimated value;

[0154] (7) Approximate function weight The update law is shown in equation (16):

[0155]

[0156] Among them, E α F represents αThe approximation error index; τ α For E α right Find the fractional order of the derivative and τ α ∈(0,1); β α This represents the learning rate, and β α ∈(0,1); α α Denotes the momentum factor, and α α ∈(0,1);

[0157] (8) Define the pitch angular velocity command Q d Considering the actual constraints of the servo as shown in equation (17):

[0158]

[0159] in, and δ e These are the elevator deflection angles δ e The upper and lower bounds;

[0160] (9) Introducing the compensation system shown in equation (18):

[0161]

[0162] Where, δ c This indicates control inputs that do not take into account actual constraints;

[0163] (10) Define the tracking error z Q :

[0164] z Q =e Q -χ Q

[0165] (11) Obtain the following second-order tracking differentiator.

[0166]

[0167] Among them, l Q , λ Q ξ is a positive constant to be designed; Q1 and ξ Q2 They are Q d and The estimated value;

[0168] (12) The actual control law is designed as follows:

[0169]

[0170] in, yes Approximate;

[0171] (13) Approximate function weight The update law is shown in equation (19):

[0172]

[0173] Among them, E Q For F Q Approximation error index; τ Q For E Q right Find the fractional order of the derivative and τ Q ∈(0,1); β Q This represents the learning rate, and β Q ∈(0,1); α Q Denotes the momentum factor, and α Q ∈(0,1);

[0174] The hypersonic vehicle flight simulation environment can be understood as a flight control simulation environment for hypersonic vehicles used in deep reinforcement learning to train DDPG (Deterministic Policy Gradient) agents; specifically, the steps of constructing the hypersonic vehicle flight simulation environment based on the fractional-order RBF network controller and the longitudinal channel dynamics model include:

[0175] Acquire flight data of the hypersonic vehicle; wherein, flight data can be understood as pre-collected flight control data of a real hypersonic vehicle, used for data initialization of the simulation environment, including initial state information and initial control parameter information;

[0176] Based on the initial state information, the state information set of the longitudinal channel dynamics model is initialized, and based on the initial control parameter information, the control parameter set of the fractional-order RBF network controller is initialized; wherein, the state information set can be defined as S = {s0, s1, s2, ..., s...} t ,…,s n}, s0 represents the initial state of the hypersonic vehicle, s t This indicates that the hypersonic vehicle is in state s t-1 At time t-1, execute control action a. t-1 The state reached afterward; the control parameter set can be defined as A = {a0, a1, a2, ..., a...} t ,…,a n}, where a0 represents the initial control parameter information of the fractional-order RBF neural network controller, a t Indicates that in state s t That is, the control parameters determined by the controller at time t;

[0177] Based on the longitudinal channel dynamics model and the fractional-order RBF network controller, a flight simulation environment for the hypersonic vehicle is constructed.

[0178] S13. The pre-built agent is interactively trained with the hypersonic vehicle flight simulation environment to obtain a controller parameter tuning agent; wherein, the agent is understood as a DDPG agent used to optimize and tune the parameters of the fractional-order RBF network controller, such as... Figure 3 As shown, its output action is the controller parameter vector, and the input is the hypersonic vehicle's flight state S output from the hypersonic vehicle flight simulation environment. t = (V, h, γ, α, Q), the reward function R used to generate the reward value corresponding to the flight state. t And the observations determined based on the control system and the situation. i =(V,h,γ,α,Q,z) V ,e h ,e γ ,e α ,z Q ), where z V e h e γ e α and z Q These are the tracking errors for flight speed V, flight altitude h, track inclination angle γ, angle of attack α, and pitch rate Q, respectively.

[0179] The aforementioned intelligent agent is constructed using a deep neural network, such as... Figure 4 As shown, it includes a value function-based neural network (Critic network) and a policy-based neural network (Actor network). The corresponding network structure includes an input layer, a hidden layer, and an output layer. The corresponding network structure parameters include a learning rate, a reward decay factor, and an exploration probability. The input layer contains m neurons, which are used to input the state of the hypersonic vehicle at time t. The hidden layer contains m neurons and uses an activation function to calculate the state information from the input layer and transmit it to the output layer. The output layer contains n neurons, which are used to output the action value function.

[0180] Specifically, such as Figure 5 As shown, the step of interactively training the pre-built agent with the hypersonic vehicle flight simulation environment to obtain the controller parameter tuning agent includes:

[0181] Based on the state information set and the control parameter set, select the initial state to be simulated and the control parameters to be simulated;

[0182] Based on the initial state to be simulated and the control parameters to be simulated, the interactive simulation between the agent and the hypersonic vehicle flight simulation environment is executed. When a preset simulation termination condition is reached, the simulation flight of the current round is terminated. The cumulative reward value of each state point in the current simulated flight trajectory is evaluated according to a preset reward function, and the parameters of the value function-based neural network are updated based on the cumulative reward value. The preset reward function can be understood as a reward value generation function used to evaluate the effectiveness of each flight state of the hypersonic vehicle. The design of the preset reward function in this embodiment follows the following principles:

[0183] 1) Ensure that the reward function is convergent and has a fast convergence speed;

[0184] 2) Ensure that agent training is conducted within certain constraints and conforms to physical constraints;

[0185] 3) Ensure the system's tracking performance and security.

[0186] Based on the above principles, this embodiment preferably designs penalties for each flight state from three aspects: flight state error, track tilt angle, and pitch angular velocity, to ensure that the agent training is carried out within a certain constraint range, which is more in line with the physical world. The design of the reward function takes into account tracking performance and safety, and the corresponding reward function is shown in Equation (20):

[0187] r t (s j ,a j ) = r t1 (s j ,a j )+r t2 (s j ,a j )+r t3 (s j ,a j (20)

[0188] In the formula,

[0189]

[0190]

[0191]

[0192] Where, r t (s j ,a j ), r t1 (s j ,a j ), r t2 (sj ,a j ) and r t3 (s j ,a j ) represent the flight state s at time t, respectively. j Lower control parameter a j The corresponding total reward value, flight status reward function, track tilt angle reward function, and pitch rate reward function; z Q (s j ,a j )and These represent flight states s and s respectively. j Lower control parameter a j The pitch angular velocity tracking error and its corresponding first derivative; z V (s j ,a j )and These represent the flight speed tracking error and its corresponding first derivative, respectively; e h (s j ,a j )and These represent the flight altitude tracking error and its corresponding first derivative, respectively; e γ (s j ,a j )and These represent the tracking error of the trajectory inclination angle and its corresponding first derivative, respectively; e α (s j ,a j )and Let r1, r2, and r3 represent the angle-of-attack tracking error and the corresponding first derivative, respectively; r1, r2, and r3 represent constants, where r1>0, r2>0, and r3>0; δ max δ e (s j ,a j The upper limit of ); e γmax This represents the maximum absolute value of the track inclination error; This represents the maximum absolute value of the pitch angular velocity;

[0193] Based on equation (20), the cumulative reward value R of the current flight trajectory of the hypersonic vehicle can be obtained. t As shown in equation (21):

[0194]

[0195] Where λ∈[0,1] represents the discount factor, used to represent the discount of the reward at each moment in the flight trajectory. For a reinforcement learning algorithm, the objective is to find a set of policies that maximizes the expected value of the cumulative discount reward of the trajectory. t (s j ,aj ) represents the flight state at time j. j Lower control parameter a j The corresponding total reward value;

[0196] Based on the updated value function-based neural network, the expected cumulative reward value for each state point in the current simulated flight trajectory is predicted, and the parameters of the policy-based neural network are updated based on the cumulative reward value and the expected cumulative reward value; wherein, the expected cumulative reward value can be understood as the value of a certain state s. t Take a certain action a t And it satisfies the policy, i.e., a t =π(s) t When ), the corresponding action value Q π (s t ,a t As shown in equation (22):

[0197]

[0198] Specifically, the process of updating the parameters of the policy-based neural network is as follows: Consider the policy π, and define the performance objective function as shown in equation (23):

[0199]

[0200] in, This represents the gradient of the objective function J with respect to θ. Represents state s t Consistent with distribution ρ π The expected value of Q under the given conditions; Q π (s,a) represents the state s t Next, execute action a according to the deterministic strategy π. t At that time, the action state value Q that it can generate; π θ (s) indicates that in state s t The strategy probability distribution function;

[0201] Based on equation (23), the target Actor network and Critic network are updated. According to the Markov property, Q is updated in two consecutive time intervals as follows:

[0202]

[0203] Among them, Q π (s t ,a t ) indicates that in state s t Next, execute action a according to the deterministic strategy π. t At that time, the action state value Q that it can generate; Represents state s t+1It conforms to the distribution p(·|s t ,a t Expectations under the following circumstances;

[0204] Therefore, the optimal strategy is updated as follows:

[0205]

[0206] Among them, Q * (s t ,a t Optimal action state value; π * (a t |s t ) represents the optimal strategy;

[0207] And, in the optimal strategy π * Below, we obtain Q. * for:

[0208]

[0209] Among them, Q * (s t ,a t ) indicates that in state s t Next, according to the optimal strategy π * Perform action a t At that time, the optimal action state value that it can produce; λ∈[0,1] represents the discount factor; r t (s t ,a t () represents the flight state at time t. t Lower control parameter a t The corresponding total reward value; Represents state s t+1 It conforms to the distribution p(·|s t ,a t Expectations under the following circumstances;

[0210] Determine whether the policy-based neural network has reached a preset convergence condition. If it has, stop the simulation training and obtain the controller parameter tuning agent. Otherwise, reselect the initial state to be simulated and the control parameters to be simulated based on the state information set and the control parameter set, and start the next round of interactive simulation training. The preset convergence condition can be determined based on the performance objective function shown in the figure. The controller parameter tuning agent can be understood as the agent that has reached the training termination condition and whose stored value function Q... * Optimal strategy π * The reward function R and the deep neural network structure parameters are used to tune the parameters of the fractional-order RBF neural network controller for a hypersonic vehicle. The output controller parameters are a = {k1, k2, ..., kn A stable intelligent agent that can meet the performance requirements of the controller;

[0211] S14. Obtain the real-time flight status of the hypersonic vehicle and input the real-time flight status into the controller parameter tuning agent for parameter tuning optimization to obtain the optimal controller parameters; wherein, the process of the controller parameter tuning agent performing parameter tuning optimization based on the real-time flight status of the hypersonic vehicle can refer to the above-mentioned process of training and constructing the controller parameter tuning agent, and will not be repeated here.

[0212] S15. Input the optimal controller parameters into the fractional-order RBF network controller to generate corresponding flight control commands, and perform flight control on the hypersonic vehicle according to the flight control commands.

[0213] This application embodiment constructs a longitudinal channel dynamics model of a hypersonic vehicle, including a velocity subsystem and an altitude subsystem, based on modeling information of the hypersonic vehicle's configuration. A fractional-order RBF network controller is then constructed based on this model. A hypersonic vehicle flight simulation environment is built using the fractional-order RBF network controller and the longitudinal channel dynamics model. A pre-constructed intelligent agent is interactively trained with the hypersonic vehicle flight simulation environment to obtain a controller parameter tuning agent. The real-time flight state of the hypersonic vehicle is then input into the controller parameter tuning agent for parameter tuning optimization to obtain optimal controller parameters. These optimal controller parameters are then input into the fractional-order RBF network controller to generate corresponding flight control commands. The method combines the fractional-order RBF neural network control method with deep reinforcement learning-based optimization techniques, solving the problem of stable, reliable, and precise control of hypersonic vehicles in complex and variable environments. This effectively improves the controller's control accuracy and utilizes deep reinforcement learning techniques to address the difficulty of controller parameter tuning, achieving rapid and stable control of hypersonic vehicles, thus possessing high application value.

[0214] It should be noted that although the steps in the flowchart above are shown sequentially as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order requirement for the execution of these steps, and they can be executed in other orders.

[0215] In one embodiment, such as Figure 6 As shown, a fractional-order RBF network controller optimization system for hypersonic vehicles is provided, the system comprising:

[0216] Model building module 1 is used to construct a longitudinal channel dynamics model of the hypersonic vehicle based on the modeling information of the hypersonic vehicle; the modeling information includes the vehicle configuration; the longitudinal channel dynamics model includes a velocity subsystem and an altitude subsystem;

[0217] Simulation building module 2 is used to construct a fractional-order RBF network controller based on the longitudinal channel dynamics model, and to construct a hypersonic vehicle flight simulation environment based on the fractional-order RBF network controller and the longitudinal channel dynamics model;

[0218] Strategy training module 3 is used to interactively train a pre-built agent with the hypersonic vehicle flight simulation environment to obtain a controller parameter tuning agent.

[0219] The parameter tuning module 4 is used to acquire the real-time flight status of the hypersonic vehicle and input the real-time flight status into the controller parameter tuning agent for parameter tuning optimization to obtain the optimal controller parameters.

[0220] The control optimization module 5 is used to input the optimal controller parameters into the fractional-order RBF network controller, generate corresponding flight control commands, and perform flight control on the hypersonic vehicle according to the flight control commands.

[0221] Specific limitations regarding the optimization system for a fractional-order RBF network controller for hypersonic vehicles can be found in the above description of the optimization method for a fractional-order RBF network controller for hypersonic vehicles; the corresponding technical effects are equivalent and will not be repeated here. Each module in the aforementioned optimization system for a fractional-order RBF network controller for hypersonic vehicles can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the operations corresponding to each module.

[0222] Figure 7 An internal structural diagram of a computer device is shown in one embodiment. This computer device may specifically be a terminal or a server. Figure 7As shown, the computer device includes a processor, memory, network interface, display, camera, and input device connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The network interface is used to communicate with external terminals via a network connection. When the computer program is executed by the processor, it implements a fractional-order RBF network controller optimization method for hypersonic vehicles. The display screen can be an LCD screen or an e-ink display screen. The input device can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0223] Those skilled in the art will understand that Figure 7 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computing devices may include more or fewer components than those shown in the figure, or combine certain components, or have the same component arrangement.

[0224] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the method described above.

[0225] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described method.

[0226] In summary, the hypersonic vehicle fractional-order RBF network controller optimization method and system provided in this invention optimizes the hypersonic vehicle fractional-order RBF network controller by constructing a longitudinal channel dynamics model of the hypersonic vehicle, including a velocity subsystem and an altitude subsystem, based on modeling information including the vehicle configuration. Then, a fractional-order RBF network controller is constructed based on the longitudinal channel dynamics model. Finally, a hypersonic vehicle flight simulation environment is constructed based on the fractional-order RBF network controller and the longitudinal channel dynamics model. The pre-constructed agent is then interactively trained with the hypersonic vehicle flight simulation environment to obtain the controller parameters. The system employs a tuning agent, inputting the real-time flight status of the hypersonic vehicle into the controller parameter tuning agent for parameter tuning and optimization to obtain the optimal controller parameters. These optimal controller parameters are then input into a fractional-order RBF network controller to generate corresponding flight control commands. This system, which combines deep reinforcement learning optimization techniques with a fractional-order RBF neural network controller, achieves optimal control of the hypersonic vehicle. This approach effectively improves the controller's control accuracy while reducing the difficulty of controller parameter tuning and optimization, enabling rapid and stable control of the hypersonic vehicle and demonstrating significant application value.

[0227] The various embodiments in this specification are described in a progressive manner. For directly identical or similar parts of the embodiments, refer to each other. Each embodiment focuses on its differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments. It should be noted that the technical features of the above embodiments can be combined arbitrarily. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as the combination of these technical features does not contradict each other, it should be considered within the scope of this specification.

[0228] The embodiments described above are merely preferred embodiments of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various improvements and substitutions without departing from the technical principles of this invention, and these improvements and substitutions should also be considered within the scope of protection of this application. Therefore, the scope of protection of this patent application should be determined by the scope of the claims.

Claims

1. A method for optimizing a fractional-order RBF network controller for a hypersonic vehicle, characterized in that, The method includes the following steps: Based on the modeling information of the hypersonic vehicle, a longitudinal channel dynamics model of the hypersonic vehicle is constructed considering the elastic state of the vehicle; the modeling information includes the vehicle configuration; the longitudinal channel dynamics model includes a velocity subsystem and an altitude subsystem. Based on the longitudinal channel dynamics model, a fractional-order RBF network controller is constructed, and based on the fractional-order RBF network controller and the longitudinal channel dynamics model, a hypersonic vehicle flight simulation environment is constructed; the fractional-order RBF network controller includes a velocity subsystem controller and an altitude subsystem controller; The pre-built intelligent agent is interactively trained with the hypersonic vehicle flight simulation environment to obtain the controller parameter tuning intelligent agent; The real-time flight status of the hypersonic vehicle is obtained, and the real-time flight status is input into the controller parameter tuning agent for parameter tuning and optimization to obtain the optimal controller parameters. The optimal controller parameters are input into the fractional-order RBF network controller to generate corresponding flight control commands, and the hypersonic vehicle is controlled to fly according to the flight control commands. The step of constructing a fractional-order RBF network controller based on the longitudinal channel dynamics model includes: Based on fractional calculus theory, a fractional RBF neural network is pre-constructed; the weights of the fractional RBF neural network are updated using the fractional gradient descent method. The fractional-order RBF neural network is used to approximate the nonlinear functions of speed tracking, tilt tracking, angle of attack tracking, and pitch rate tracking, respectively, to obtain the corresponding approximate nonlinear functions of speed tracking, tilt tracking, angle of attack tracking, and pitch rate tracking. Based on the aforementioned speed tracking approximate nonlinear function, the control model of the speed subsystem controller is obtained; Based on the approximate nonlinear function for tilt tracking, the approximate nonlinear function for angle of attack tracking, and the approximate nonlinear function for pitch angular velocity tracking, the control model of the altitude subsystem controller is obtained; The reward function of the controller parameter tuning agent is expressed as follows: In the formula, in, , , and They represent t Constant flight status Lower control parameters The corresponding total reward value, flight status reward function, track tilt angle reward function, and pitch rate reward function; and These represent the flight status. Lower control parameters The pitch angular velocity tracking error and its corresponding first derivative; and These represent the flight speed tracking error and the corresponding first derivative, respectively; and These represent the flight altitude tracking error and its corresponding first derivative, respectively. and These represent the tracking error of the trajectory tilt angle and its corresponding first derivative, respectively; and These represent the angle-of-attack tracking error and its corresponding first derivative, respectively. and These represent the trajectory tilt angle, pitch angular velocity, and corresponding first derivative of the hypersonic vehicle, respectively. , and Represents a constant; Indicates elevator deflection angle The upper limit; This represents the maximum absolute value of the track inclination error; This represents the maximum absolute value of the pitch angular velocity.

2. The optimization method for a fractional-order RBF network controller of a hypersonic vehicle as described in claim 1, characterized in that, The aircraft has a symmetrical configuration; the longitudinal channel dynamics model is expressed as follows: in, and Let represent the flight speed of the hypersonic vehicle and its corresponding first derivative, respectively; and These represent the flight altitude of the hypersonic vehicle and its corresponding first derivative, respectively. and represent the trajectory inclination angle and the corresponding first derivative of the hypersonic vehicle, respectively; and represent the angle of attack and the corresponding first derivative of the hypersonic vehicle, respectively; and These represent the trajectory tilt angle, pitch angular velocity, and corresponding first derivative of the hypersonic vehicle, respectively. , and These represent the elastic state of the hypersonic vehicle, and the corresponding first and second derivatives, respectively. , and These represent the mass, moment of inertia, and pitching moment of a hypersonic vehicle, respectively. , and These represent thrust, aerodynamic drag, and lift, respectively. , and They represent the first i Damping ratio, natural frequency, and generalized elastic force; It represents the acceleration due to gravity.

3. The optimization method for a fractional-order RBF network controller of a hypersonic vehicle as described in claim 1, characterized in that, The control model of the speed subsystem controller is represented as follows: In the formula, in, Indicates throttle opening; , , and Represents a constant; and These represent the speed error and the corresponding tracking error, respectively. V Indicates the speed of the aircraft; Represents a symbolic function; and These represent the reference command velocity and its corresponding first-order reciprocal, respectively. This represents an approximate nonlinear function for speed tracking; The control model of the altitude subsystem controller is represented as follows: In the formula, in, Indicates the track inclination control value; and Represents a constant; and These represent the reference flight altitude and its corresponding first-order reciprocal, respectively. V Indicates flight speed; Indicates tracking error; Indicates the angle of attack control amount; and Represents a constant; and These represent the reference track inclination angle and its corresponding first-order reciprocal, respectively; This indicates the tracking error in track tilt angle; This represents an approximate nonlinear function for tilt tracking; Indicates the pitch angular velocity control value; and Represents a constant; Indicates the angle-of-attack tracking error; and These represent the reference angle of attack and its corresponding first-order reciprocal, respectively. This represents an approximate nonlinear function for angle-of-attack tracking. Indicates the elevator deflection angle; , , and Represents a constant; and These represent the pitch angular velocity error and the corresponding tracking error, respectively. and These represent the reference pitch angular velocity and its corresponding first-order reciprocal, respectively. This represents an approximate nonlinear function for pitch angular velocity tracking; Represents a symbolic function.

4. The optimization method for a fractional-order RBF network controller of a hypersonic vehicle as described in claim 3, characterized in that, The weight update of the fractional-order RBF neural network is expressed as follows: in, This represents the network approximation error metric. , and They represent the first i The update values ​​of each weight in the (t-1), t-th, and t+1th iterations; Indicates the network approximation error index For the i-th weight in the t-th iteration (t) Find the fractional order of the derivative; Indicates the learning rate; Indicates the momentum factor; n Indicates the number of weights.

5. The optimization method for a fractional-order RBF network controller of a hypersonic vehicle as described in claim 1, characterized in that, The steps for constructing a hypersonic vehicle flight simulation environment based on the fractional-order RBF network controller and the longitudinal channel dynamics model include: Acquire flight data of the hypersonic vehicle; the flight data includes initial state information and initial control parameter information; Based on the initial state information, the state information set of the longitudinal channel dynamics model is initialized, and based on the initial control parameter information, the control parameter set of the fractional-order RBF network controller is initialized. Based on the longitudinal channel dynamics model and the fractional-order RBF network controller, a flight simulation environment for the hypersonic vehicle is constructed.

6. The optimization method for a fractional-order RBF network controller of a hypersonic vehicle as described in claim 5, characterized in that, The intelligent agent includes value function-based neural networks and policy-based neural networks; The step of interactively training a pre-built intelligent agent with the hypersonic vehicle flight simulation environment to obtain a controller parameter tuning intelligent agent includes: Based on the state information set and the control parameter set, select the initial state to be simulated and the control parameters to be simulated; Based on the initial state to be simulated and the control parameters to be simulated, the interaction simulation between the agent and the hypersonic vehicle flight simulation environment is executed. When the preset simulation termination condition is reached, the simulation flight of the current round is terminated. The cumulative reward value of each state point in the current simulation flight trajectory is evaluated according to the preset reward function, and the parameters of the value function-based neural network are updated according to the cumulative reward value. Based on the updated value function-based neural network, the expected cumulative reward value of each state point in the current simulated flight trajectory is predicted, and the parameters of the policy-based neural network are updated based on the cumulative reward value and the expected cumulative reward value. Determine whether the policy-based neural network has reached the preset convergence condition. If it has, stop the simulation training and obtain the controller parameter tuning agent. Otherwise, reselect the initial state to be simulated and the control parameters to be simulated based on the state information set and the control parameter set, and start the next round of interactive simulation training.

7. A fractional-order RBF network controller optimization system for hypersonic vehicles, characterized in that, The system employs the fractional-order RBF network controller optimization method for hypersonic vehicles as described in claim 1, wherein the system comprises: The model building module is used to construct a longitudinal channel dynamics model of the hypersonic vehicle based on the modeling information of the hypersonic vehicle and considering the elastic state of the vehicle; the modeling information includes the vehicle configuration; the longitudinal channel dynamics model includes a velocity subsystem and an altitude subsystem. The simulation building module is used to construct a fractional-order RBF network controller based on the longitudinal channel dynamics model, and to construct a hypersonic vehicle flight simulation environment based on the fractional-order RBF network controller and the longitudinal channel dynamics model. The strategy training module is used to interactively train a pre-built agent with the hypersonic vehicle flight simulation environment to obtain a controller parameter tuning agent. The parameter tuning module is used to acquire the real-time flight status of the hypersonic vehicle and input the real-time flight status into the controller parameter tuning agent for parameter tuning optimization to obtain the optimal controller parameters. The control optimization module is used to input the optimal controller parameters into the fractional-order RBF network controller, generate corresponding flight control commands, and perform flight control on the hypersonic vehicle according to the flight control commands.

8. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

Images