A model-free attitude control method for a helicopter / air composite aircraft considering model uncertainties

Abstract

This invention proposes a model-free attitude control method for a straight / air hybrid aircraft that considers model uncertainty. During the control law design process, this method utilizes an LSTM neural network to fit the system terms, making it independent of prior model information. Based on the Lyapunov method, the online update law of the neural network weights is derived, and the system stability under the dual effects of the control law and the network weight update law is proven. In the simulation, ASMMFC is applied to an aircraft attitude control scenario with model uncertainty and time-varying parameters. Simulation results show that ASMMFC outperforms other existing adaptive sliding mode control methods.

Description

Technical Field

[0001] This invention relates to the field of aircraft attitude control technology, and in particular to a model-free attitude control method for a vertical / air hybrid aircraft that takes into account model uncertainties. Background Technology

[0002] Nonlinear model uncertainties and control model parameter perturbations are among the reasons for the difficulty in designing aircraft attitude controllers, leading to a decline in controller quality and even divergence. For helicopter / aircraft with combined control, the impact of these problems is even more severe due to jet interference effects and drastic changes in aircraft state, making controller design even more challenging. With the development of neural network technology, algorithms combining neural networks with traditional control methods are increasingly being applied to control system design. Considering that external disturbances and model parameter perturbations can cause difficulties in controller design and a decline in control quality, combining artificial neural networks with sliding mode control can improve controller quality and reduce the difficulty of control design. Summary of the Invention

[0003] The purpose of this invention is to solve the problem of attitude control of existing vertical / air composite control aircraft with model uncertainty, and to propose a model-free attitude control method for vertical / air composite aircraft that considers model uncertainty.

[0004] This invention is achieved through the following technical solution: This invention proposes a model-free attitude control method for a helicopter / air composite aircraft considering model uncertainties, the method comprising the following steps:

[0005] Step 1: Establish a three-channel attitude dynamics model for a straight / aerodynamic composite control aircraft that considers aerodynamic parameter perturbations;

[0006] Step 2: Design a sliding mode controller based on the nominal system without considering external disturbances and parameter perturbations;

[0007] Step 3: Use an LSTM neural network to fit the system terms in the sliding mode control law and design an adaptive sliding mode model-free controller (ASMMFC).

[0008] Step 4: Based on the Lyapunov method, design the weight update law of the LSTM neural network and ensure the system stability under the dual constraints of sliding mode control law and weight update law.

[0009] Step 5: Combining the model and control strategy designed in the previous steps, conduct simulation experiments and analysis of the three-channel attitude control of the straight / air hybrid control aircraft.

[0010] Furthermore, in step one, the aerodynamic parameters of the aircraft are specifically defined as follows:

[0011] a4 = Y α / (mV), a6=-(J x -J y ) / J z , b4 = -Z β / (mV), b6=-(J z -J x ) / J y , k y =1 / (mV),k z =-1 / (mV),l y =l / J y ,l z =-l / J z ,

[0012] In the formula:

[0013]

[0014] Using the above aerodynamic parameters, the three-channel attitude dynamics model of the direct / aerodynamic composite control aircraft can be modeled as follows:

[0015]

[0016] In the formula:

[0017] X=[γ ω x ω y ω z n y n z δ x δ y δ z F z F y ] T For system state variables; u = [δ xc δ yc δ zc F zc F yc ] T For system control input.

[0018] Furthermore, in step two, consider a class of multi-input multi-output nonlinear systems:

[0019]

[0020] In the formula: Defined as a system state vector, Defined as system control input, f(X)∈R N×NDefined as a system state function, B∈R N×u Defined as a control matrix, d∈R N Defined as external interference;

[0021] The system control command tracking error is defined as:

[0022] e = XX d (3)

[0023] Taking the derivative of the instruction tracking error with respect to time t, we can obtain:

[0024]

[0025] Design a sliding surface as follows:

[0026] S = Ce(5).

[0027] Furthermore, in step two, it is assumed that matrix CB is invertible, and for a specific control system, control matrix B is known; this assumption can be satisfied by selecting an appropriate sliding mode parameter matrix C.

[0028] Taking the derivative of the sliding mode function with respect to time t, we can obtain:

[0029]

[0030] set up And assuming d=0, the equivalent control rate of sliding mode control can be obtained as follows:

[0031]

[0032] To achieve the exponential sliding mode reaching law, the switching control law is designed as follows:

[0033]

[0034] in To determine the right pseudoinverse of the control matrix B, ie

[0035] By combining the equivalent control law (7) and the switching control law (8), the control law of the nonlinear system can be obtained:

[0036] u = u eq +u sw (9)

[0037] Substituting the control law (9) into formula (6), we get:

[0038]

[0039] Where: matrix Since both the external disturbance d and the sliding mode parameter matrix K and P are bounded, the sliding mode can be optimized by selecting appropriate sliding mode parameter matrices K and P. Negative definite; that is, it exists. Sliding mode accessibility is satisfied.

[0040] Furthermore, in step two, when S = 0, the Lyapunov equation is designed as follows:

[0041]

[0042] Differentiating formula (11) with respect to time t and introducing the control law (9), we can obtain:

[0043]

[0044] For the system function f(X) at the point X = X d Performing a first-order Taylor expansion at the given point, we obtain:

[0045]

[0046] Define ψ = Δf(X) - B(CB) -1 CΔf(X), where Substituting formula (13) into formula (12), we get:

[0047]

[0048] Ensure the linear matrix inequality ψ=(Δf(X)-B(CB)) is met. -1 If CΔf(X))<0 holds, then we can obtain Negative definiteness, according to Lyapunov stability theory, means the system is asymptotically stable, i.e., it exists... Furthermore, the values ​​of the sliding mode parameter matrix C can be obtained by solving this linear matrix inequality.

[0049] Furthermore, in the process of fitting the nonlinear equation y=f(X) by the LSTM neural network, there exists an optimal weight W. * Standard bias term and optimal gate unit weights The optimal output of a neural network can be represented as y * =W * h * +ε, where h * Let represent the optimal value of h, and ε be the bounded fitting error, i.e., ||ε|| ≤ ε. b , ε b It is a normal number, considering that all optimal weights are constants or slow time-varying parameters.

[0050] Furthermore, define:

[0051] y * =W * h * +ε=f(X) (15)

[0052] The actual output value of the LSTM neural network during the fitting process is defined as:

[0053]

[0054] in: For the actual scaling matrix, This represents the actual hidden state of the LSTM neural network. For a nonlinear function with 13 independent variables, its implicit expression can be written as:

[0055]

[0056] Substituting the fitted values ​​from the LSTM neural network into the control law yields the new control law:

[0057]

[0058] The fitting error of an LSTM neural network for a nonlinear function can be defined as:

[0059]

[0060] In the formula: Defined as approximation error, Expanding the weight matrix of the gated unit of the LSTM neural network into vector form, i.e.:

[0061]

[0062] According to the Taylor expansion formula, we can obtain:

[0063]

[0064] In the formula: O h Indicates the approximation error. This represents the Jacobian matrix.

[0065] Furthermore, the weight update law for the LSTM neural network is designed as follows:

[0066]

[0067]

[0068] In the formula: W ci Let represent the i-th column of matrix W.

[0069] The present invention also proposes an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the model-free attitude control method for a straight / air composite aircraft considering model uncertainty.

[0070] The present invention also proposes a computer-readable storage medium for storing computer instructions, which, when executed by a processor, implement the steps of the model-free attitude control method for a straight / air composite aircraft considering model uncertainties.

[0071] Compared with the prior art, the beneficial effects of the present invention are:

[0072] This invention addresses the attitude control problem of direct force / aerodynamic composite aircraft with model uncertainty, proposing an Adaptive Sliding Mode Model-Free Control (ASMMFC) algorithm based on a Long-Short-Term Memory (LSTM) neural network. Compared to existing neural network sliding mode algorithms, ASMMFC relaxes its basic assumptions, thus broadening its applicability. During the control law design process, an LSTM neural network is used to fit the system terms, making it independent of prior model information. The online update law of the neural network weights is derived based on the Lyapunov method, and the system stability under the dual effects of the control law and the network weight update law is proven. In simulations, ASMMFC is applied to an aircraft attitude control scenario with model uncertainty and time-varying parameters. Simulation results show that ASMMFC outperforms other existing adaptive sliding mode control methods. Attached Figure Description

[0073] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0074] Figure 1 This is a schematic diagram of the body coordinate system and velocity coordinate system of a straight / air-based composite control aircraft.

[0075] Figure 2 This is a schematic diagram of a vertical / air-based composite control aircraft.

[0076] Figure 3 This is a diagram of the LSTM neural network structure;

[0077] Figure 4This is a diagram of the adaptive model-free sliding mode control algorithm.

[0078] Figure 5 This is a schematic diagram of the aircraft's angle-of-attack response curve;

[0079] Figure 6 This is a schematic diagram of the aircraft's sideslip angle response curve;

[0080] Figure 7 This is a schematic diagram of the aircraft's roll angle response curve;

[0081] Figure 8 This is a schematic diagram of the overload response curve of the aircraft's pitch channel;

[0082] Figure 9 This is a schematic diagram of the overload response curve of the aircraft's yaw channel;

[0083] Figure 10 This is a schematic diagram of the aircraft pitch channel control input curve;

[0084] Figure 11 This is a schematic diagram of the yaw channel control input curve for an aircraft;

[0085] Figure 12 This is a schematic diagram of the control input curve for the aircraft's roll channel. Detailed Implementation

[0086] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0087] Combination Figures 1-12 This invention proposes a model-free attitude control method for a helicopter / air-air hybrid aircraft that considers model uncertainties. The method includes the following steps:

[0088] Step 1: Establish a three-channel attitude dynamics model for a straight / aerodynamic composite control aircraft that considers aerodynamic parameter perturbations;

[0089] Step 2: Design a sliding mode controller based on the nominal system without considering external disturbances and parameter perturbations;

[0090] Step 3: Use an LSTM neural network to fit the system terms in the sliding mode control law and design an adaptive sliding mode model-free controller (ASMMFC).

[0091] Step 4: Based on the Lyapunov method, design the weight update law of the LSTM neural network and ensure the system stability under the dual constraints of sliding mode control law and weight update law.

[0092] Step 5: Combining the model and control strategy designed in the previous steps, conduct simulation experiments and analysis of the three-channel attitude control of the straight / air hybrid control aircraft.

[0093] In step one, the specific definitions of the aircraft's aerodynamic parameters are as follows:

[0094] a4 = Y α / (mV), a6=-(J x -J y ) / J z , b4 = -Z β / (mV), b6=-(J z -J x ) / J y , k y =1 / (mV),k z =-1 / (mV),l y =l / J y ,l z =-l / J z ,

[0095] In the formula:

[0096]

[0097] Using the above aerodynamic parameters, the three-channel attitude dynamics model of the direct / aerodynamic composite control aircraft can be modeled as follows:

[0098]

[0099] In the formula:

[0100] X=[γ ω x ω y ω z n y n z δ x δ y δ z F z F y ] T For system state variables; u = [δ xc δ yc δ zc F zc F yc ] T For system control input.

[0101] In step two, consider a class of multi-input multi-output nonlinear systems:

[0102]

[0103] In the formula: Defined as a system state vector, Defined as system control input, f(X)∈R N×N Defined as a system state function, B∈R N×u Defined as a control matrix, d∈R N Defined as external interference;

[0104] The system control command tracking error is defined as:

[0105] e = XX d (3)

[0106] Taking the derivative of the instruction tracking error with respect to time t, we can obtain:

[0107]

[0108] Design a sliding surface as follows:

[0109] S = Ce(5).

[0110] In step two, it is assumed that matrix CB is invertible, and for a specific control system, control matrix B is known; this assumption can be satisfied by selecting an appropriate sliding mode parameter matrix C.

[0111] Taking the derivative of the sliding mode function with respect to time t, we can obtain:

[0112]

[0113] set up And assuming d=0, the equivalent control rate of sliding mode control can be obtained as follows:

[0114]

[0115] To achieve the exponential sliding mode reaching law, the switching control law is designed as follows:

[0116]

[0117] in To determine the right pseudoinverse of the control matrix B, ie

[0118] By combining the equivalent control law (7) and the switching control law (8), the control law of the nonlinear system can be obtained:

[0119] u = u eq +u sw (9)

[0120] Substituting the control law (9) into formula (6), we get:

[0121]

[0122] Where: matrix Since both the external disturbance d and the sliding mode parameter matrix K and P are bounded, the sliding mode can be optimized by selecting appropriate sliding mode parameter matrices K and P. Negative definite; that is, it exists. Sliding mode accessibility is satisfied.

[0123] In step two, when S = 0, the Lyapunov equation is designed as follows:

[0124]

[0125] Differentiating formula (11) with respect to time t and introducing the control law (9), we can obtain:

[0126]

[0127] For the system function f(X) at the point X = X d Performing a first-order Taylor expansion at the given point, we obtain:

[0128]

[0129] Define ψ = Δf(X) - B(CB) -1 CΔf(X), where Substituting formula (13) into formula (12), we get:

[0130]

[0131] Ensure the linear matrix inequality ψ=(Δf(X)-B(CB)) is met. -1 If CΔf(X))<0 holds, then we can obtain Negative definiteness, according to Lyapunov stability theory, means the system is asymptotically stable, i.e., it exists... Furthermore, the values ​​of the sliding mode parameter matrix C can be obtained by solving this linear matrix inequality.

[0132] In practical applications, systems typically contain uncertain parameters and unmodeled components, making it extremely difficult to obtain an accurate system model. When designing control laws, if an accurate system model cannot be obtained, the control effect will be reduced, and the system may even diverge. Therefore, this invention designs an adaptive sliding mode control law that does not rely on prior model information, utilizing an LSTM neural network to fit the system terms in the control law.

[0133] In the process of fitting the nonlinear equation y=f(X) by an LSTM neural network, there exists an optimal weight W. * Standard bias term and optimal gate unit weights The optimal output of a neural network can be represented as y * =W * h * +ε, where h * Let represent the optimal value of h, and ε be the bounded fitting error, i.e., ||ε|| ≤ ε. b , ε b It is a sufficiently small normal number, considering that all optimal weights are constants or slow time-varying parameters.

[0134] definition:

[0135] y * =W * h * +ε=f(X) (15)

[0136] The actual output value of the LSTM neural network during the fitting process is defined as:

[0137]

[0138] in: For the actual scaling matrix, This represents the actual hidden state of the LSTM neural network. For a nonlinear function with 13 independent variables, its implicit expression can be written as:

[0139]

[0140] Substituting the fitted values ​​from the LSTM neural network into the control law yields the new control law:

[0141]

[0142] The fitting error of an LSTM neural network for a nonlinear function can be defined as:

[0143]

[0144] In the formula: Defined as approximation error, Expanding the weight matrix of the gated unit of the LSTM neural network into vector form, i.e.:

[0145]

[0146] According to the Taylor expansion formula, we can obtain:

[0147]

[0148] In the formula: O h Indicates the approximation error. This represents the Jacobian matrix.

[0149] The weight update law for the LSTM neural network is designed as follows:

[0150]

[0151]

[0152] In the formula: W ci Let represent the i-th column of matrix W.

[0153] During neural network training, update laws (22) to (24) are used to update the neural network weights, and control law (18) is used to ensure that the training error of the network weights and the tracking error of the system are consistent and asymptotically stable. Specifically:

[0154] Design a Lyapunov function:

[0155]

[0156] V = V1 + V2 (27)

[0157] In the formula:

[0158] Taking the time derivative of the Lyapunov function, we get:

[0159]

[0160] Substituting the control law and the network weight update law, the above equation can be rewritten as:

[0161]

[0162] Substituting formula (15) into the above equation, we get:

[0163]

[0164] Substituting into formula (19), we get:

[0165]

[0166] set up The update law for the scaling matrix W can then be obtained. Expanding the terms on both sides of the equation, we get:

[0167]

[0168] By combining like terms, we can obtain the scaling matrix update law (22).

[0169] set up The update law for the gating unit weights can then be obtained (23).

[0170] Similarly, let The update law (24) for the bias terms of each gate unit can then be obtained.

[0171] Since ε0, O h d and X d Since both are bounded, we can assume that ||Cε0||≤ε E , ||Cd||≤d E , Substituting the network weights and bias update law into the Lyapunov function (27), we obtain:

[0172]

[0173] From the above formula, it can be seen that it can be achieved by setting... To obtain Negative definiteness, according to Lyapunov stability theory, exists According to formulas (11) to (14), we can obtain That is, the system tracking error eventually converges uniformly.

[0174] Simulation section

[0175] The aircraft attitude dynamics equation (1) was used in the simulation to verify the ASMMFC's ability to track overload commands. All aerodynamic parameters were designed with the aircraft operating at an altitude of 30 km in mind.

[0176] When an aircraft operates at an altitude of 30 km, the decrease in atmospheric density leads to a decrease in dynamic pressure, which in turn reduces aerodynamic forces and the effectiveness of aerodynamic control surfaces. A purely aerodynamic control mode is no longer sufficient to meet control requirements; therefore, a direct force aerodynamic composite control strategy is needed to improve the dynamic response speed of the aircraft's attitude control. The direct force engine designed in this invention is an attitude control engine. This means that the direct force does not directly generate overload, but rather ensures rapid overload tracking across all channels by quickly establishing angles of attack and sideslip angles. To track sufficiently large overload commands, the aircraft needs to operate at high angles of attack or sideslip angles. At this point, assuming aerodynamic parameters to be time-invariant or slowly variable becomes impractical.

[0177] Therefore, taking aerodynamic parameters a2 and b2 as examples, consider that the two are functions of the angle of attack and sideslip angle respectively, that is:

[0178]

[0179] Other aerodynamic parameters are considered as perturbation parameters, namely:

[0180]

[0181] Since the roll angle and roll rate are both very small, we consider c1 and c3 as perturbation parameters. We also consider the external disturbance d as Gaussian white noise.

[0182] Substituting formulas (37) to (39) into the three-channel attitude dynamics model (1) of the aircraft, we can obtain the system model used in this simulation.

[0183] The initial values ​​of the neural network weights are set to be randomly selected in the interval [-1, 1]. Considering practical application scenarios, the aerodynamic rudder deflection angle and the magnitude of the direct force are both bounded, meaning the control input has saturation constraints. Therefore, |δ xc |≤30°,|δ yc |≤30°,|δ zc |≤30°,|F z |≤3000N,|F y |≤3000N. The initial value of the system state is defined as X=[3 / 57.3 0 0 0 00 0 0] T .

[0184] Sine maneuvering is a common aircraft maneuvering mode. Therefore, in the simulation experiment, the overload commands for the pitch and yaw channels are set to sinusoidal functions, and the roll angle command is set to 0°.

[0185] The sliding mode control parameter matrix is ​​defined as follows:

[0186] K = diag(9.5 9.5 9.5 5.5 5.5)

[0187] eta1=0.01, eta2=eta3=...=eta9=0.1

[0188] Table 1 Aircraft Parameters

[0189]

[0190] The aerodynamic parameters of the yaw channel are the same as those of the pitch channel.

[0191] The present invention also proposes an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the model-free attitude control method for a straight / air composite aircraft considering model uncertainty.

[0192] The present invention also proposes a computer-readable storage medium for storing computer instructions, which, when executed by a processor, implement the steps of the model-free attitude control method for a straight / air composite aircraft considering model uncertainties.

[0193] The memory in this application embodiment can be volatile memory or non-volatile memory, or it can include both volatile and non-volatile memory. The non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDRSDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous linked dynamic random access memory (SLDRAM), and direct rambus RAM (DRRAM). It should be noted that the memory used in the methods described in this invention is intended to include, but is not limited to, these and any other suitable types of memory.

[0194] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions. When the computer instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium accessible to a computer or a data storage device such as a server or data center that integrates one or more available media. The available media may be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., high-density digital video discs (DVDs)), or semiconductor media (e.g., solid-state disks (SSDs)).

[0195] In implementation, each step of the above method can be completed by integrated logic circuits in the processor's hardware or by instructions in software. The steps of the method disclosed in the embodiments of this application can be directly implemented by a hardware processor, or by a combination of hardware and software modules in the processor. The software modules can reside in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. This storage medium is located in memory, and the processor reads information from the memory and, in conjunction with its hardware, completes the steps of the above method. To avoid repetition, detailed descriptions are omitted here.

[0196] It should be noted that the processor in the embodiments of this application can be an integrated circuit chip with signal processing capabilities. During implementation, each step of the above method embodiments can be completed by the integrated logic circuitry in the processor's hardware or by instructions in software form. The processor can be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this application. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this application can be directly embodied as being executed by a hardware decoding processor, or executed by a combination of hardware and software modules in the decoding processor. The software modules can be located in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. This storage medium is located in memory, and the processor reads the information in the memory and, in conjunction with its hardware, completes the steps of the above methods.

[0197] The above provides a detailed description of a model-free attitude control method for a straight / air composite aircraft that considers model uncertainty, as proposed in this invention. Specific examples have been used to illustrate the principles and implementation methods of this invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of this invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this invention. Therefore, the content of this specification should not be construed as a limitation of this invention.

Claims

1. A model-free attitude control method for a helicopter / air composite aircraft considering model uncertainties, characterized in that, The method includes the following steps: Step 1: Establish a three-channel attitude dynamics model for a straight / aerodynamic composite control aircraft that considers aerodynamic parameter perturbations; Step 2: Design a sliding mode controller based on the nominal system without considering external disturbances and parameter perturbations; Step 3: Use an LSTM neural network to fit the system terms in the sliding mode control law and design an adaptive sliding mode model-free controller (ASMMFC). Step 4: Based on the Lyapunov method, design the weight update law of the LSTM neural network and ensure the system stability under the dual constraints of sliding mode control law and weight update law. Step 5: Combining the model and control strategy designed in the previous steps, conduct simulation experiments and analysis of the three-channel attitude control of the straight / air hybrid control aircraft; Fitting nonlinear equations in LSTM neural networks In the process, there exists an optimal weight. Standard bias term , , , and optimal gate unit weights The optimal output of a neural network can be expressed as: ,in express The optimal value, The fitting error is bounded, i.e. , It is a normal number, considering that all optimal weights are constants or slow time-varying parameters; definition: (15) The actual output value of the LSTM neural network during the fitting process is defined as: (16) in: For the actual scaling matrix, This represents the actual hidden state of the LSTM neural network. For a nonlinear function with 13 independent variables, its implicit expression can be written as: (17) Substituting the fitted values ​​from the LSTM neural network into the control law yields the new control law: (18) The fitting error of an LSTM neural network for a nonlinear function can be defined as: (19) In the formula: Defined as approximation error, ; Expanding the weight matrix of the gated unit of the LSTM neural network into vector form, i.e.: (20) According to the Taylor expansion formula, we can obtain: (21) In the formula: Indicates the approximation error. This represents the Jacobian matrix.

2. The method according to claim 1, characterized in that, In step one, the specific definitions of the aircraft's aerodynamic parameters are as follows: , , , , , , , , , , , , , , , , , ， In the formula: ， ， Using the above aerodynamic parameters, the three-channel attitude dynamics model of the direct / aerodynamic composite control aircraft can be modeled as follows: （1） In the formula: For system state variables; For system control input.

3. The method according to claim 2, characterized in that, In step two, consider a class of multi-input multi-output nonlinear systems: （2） In the formula: Defined as a system state vector, Defined as system control input, Defined as a system state function, Defined as a control matrix, Defined as external interference; The system control command tracking error is defined as: （3） Instruction tracking error with respect to time Taking the derivative, we get: （4） Design a sliding surface as follows: （5）。 4. The method according to claim 3, characterized in that, In step two, assume the matrix It is reversible; for a specific control system, the control matrix... It is known; this assumption can be made by selecting an appropriate sliding mode parameter matrix. To satisfy; For the sliding mode function with respect to time Taking the derivative, we get: (6) set up And assume The equivalent control rate of sliding mode control can be obtained as follows: (7) To achieve the exponential sliding mode reaching law, the switching control law is designed as follows: (8) in For control matrix The right pseudo-inverse, ie ; By combining the equivalent control law (7) and the switching control law (8), the control law of the nonlinear system can be obtained: (9) Substituting the control law (9) into formula (6), we get: (10) Where: matrix and external interference All parameters are bounded, therefore, by selecting an appropriate sliding mode parameter matrix... and , making Negative definite; that is, it exists. Sliding mode accessibility is satisfied.

5. The method according to claim 4, characterized in that, In step two, when At that time, the Lyapunov equations were designed as follows: (11) For formula (11) with respect to time By taking the derivative and introducing the control law (9), we can obtain: (12) For system functions At point Performing a first-order Taylor expansion at this point, we obtain: (13) definition ,in Substituting formula (13) into formula (12), we get: (14) Guarantee linear matrix inequalities Once established, one can obtain Negative definiteness, according to Lyapunov stability theory, means the system is asymptotically stable, i.e., it exists... Furthermore, the sliding mode parameter matrix can be obtained by solving this linear matrix inequality. The value of .

6. The method according to claim 5, characterized in that, The weight update law for the LSTM neural network is designed as follows: (22) (23) (24) In the formula: Representation matrix The List.

7. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1-6.

8. A computer-readable storage medium for storing computer instructions, characterized in that, When the computer instructions are executed by the processor, they implement the steps of the method according to any one of claims 1-6.

Images