Uncertainty identification method for closed-loop attitude dynamics of reentry vehicle

Through the physical information neural network enhanced by Euler iteration, a closed-loop attitude dynamics model is built and a deep learning network is trained, which solves the problem of time-varying aerodynamic parameters perturbation identification in the reentry vehicle, and achieves high accuracy and reliability parameter identification , improving the performance stability of the aircraft.

CN118466570BActive Publication Date: 2024-11-08NINGBO INST OF NORTHWESTERN POLYTECHNICAL UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410927850.X
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-07-11
Publication Date
2024-11-08
Estimated Expiration
2044-07-11

AI Technical Summary

Technical Problem

When the existing aerodynamic parameter identification method handles the closed-loop attitude dynamics of the reentered aircraft, it is difficult to effectively identify multiple time-varying aerodynamic parameters perturbations, resulting in model uncertainty and performance instability.

Method used

Using the physical information neural network based on Euler iterative enhancement, a closed-loop attitude dynamic model containing aerodynamic parameters perturbation is constructed, training data is obtained through equal interval sampling, deep learning network is built, residual terms of control equations are added, and physical drive modules are formed. Loss function, train the network until the loss function is minimized to obtain the identification value of the perturbation of the aerodynamic parameter.

Benefits of technology

It realizes the perturbation of multiple time-varying aerodynamic parameters of reentry vehicles in a closed-loop system, improves the accuracy and reliability of parameter identification, and provides theoretical basis and application reference for the research on parameter identification of reentry vehicles.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118466570B_ABST
    Figure CN118466570B_ABST
Patent Text Reader

Abstract

The present invention discloses a method for identifying uncertainty in closed-loop attitude dynamics of a reentry vehicle, comprising the steps of: constructing a closed-loop attitude dynamics model of the reentry process of the reentry vehicle including aerodynamic parameter perturbations, sampling the attitude control task at equal intervals to obtain time series data of state quantities and control quantities as training data; building a deep learning network to fit unknown perturbations, constructing a data-driven model, and adding residual terms of the control equation on the basis of the loss function of the data-driven model to form a loss function of the physical drive module; training the deep learning network based on the training data until the loss function of the physical drive module is minimized, and obtaining an identification value of the aerodynamic parameter perturbation. The present invention provides a method for realizing unknown time-varying parameter targets in a closed-loop system by constructing an aircraft aerodynamic parameter perturbation model and a physical information neural identification network, and performing simulation verification, and can effectively identify multiple time-varying aerodynamic parameter perturbations of the reentry vehicle at the same time.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of re-entry vehicle system identification, and in particular to a re-entry vehicle closed-loop attitude dynamics uncertainty identification method based on Euler iterative enhanced physical information neural network. Background Art

[0002] At present, in the past few decades, reentry vehicles (RVs) have attracted much attention in both military and civilian fields due to their significant advantages in high-speed transportation and low-cost access to space. Reliable mathematical models are crucial to the success of controller design when ensuring that the vehicle completes its mission. However, due to the time-varying and strong coupling characteristics of its model during flight, there are serious uncertainties in the attitude dynamics of the reentry vehicle, among which the aerodynamic parameter perturbation is particularly noteworthy. This inconsistency may even affect the performance and stability of the RV. Therefore, it is very meaningful to carry out research and simulation verification on the above issues.

[0003] At present, the commonly used aerodynamic parameter identification methods can be divided into two categories: model-driven methods and data-driven methods. On the one hand, early aerodynamic parameter identification mainly adopted model-driven methods, including least squares method, maximum likelihood method, Kalman filter algorithm, etc., which required the establishment of an aerodynamic model of the aircraft and reasonable estimation of the initial values ​​of the parameters to be identified. If the model is not established accurately, the aerodynamic data fitting results will be unsatisfactory. If the initial value of the parameter is estimated unreasonably, it is easy to fall into the local minimum. On the other hand, data-based intelligent algorithms, as a data-driven identification modeling method, simulate the input-output relationship of the flight dynamics system through neural networks, and use incremental methods and derivative methods to indirectly extract aerodynamic parameters from neural networks. In addition, purely data-driven models may be very suitable for observed results, but may lead to poor generalization performance due to extrapolation or observation bias, and the predictions may be physically inconsistent or unreliable.

[0004] As one of the ways to solve these problems, the physical information-based neural network (PINN) algorithm framework is a data- and physics-driven modeling framework that can add physical constraints as regular terms into the traditional neural network modeling process, thereby reducing the demand for modeling data. At the same time, PINN also has higher identification accuracy than traditional neural networks. However, the existing PINN method mainly focuses on the identification of fixed-value time-varying unknown parameters in open-loop systems. In actual flight missions, the aircraft models are all closed-loop systems under the action of the controller, and the aerodynamic parameter perturbations are all time-varying values. Therefore, new parameter identification methods are necessary. Summary of the invention

[0005] The purpose of the present invention is to overcome the above-mentioned technical deficiencies and propose a method for identifying the uncertainty of the closed-loop attitude dynamics of a re-entry vehicle, which can effectively identify multiple time-varying aerodynamic parameter perturbations of the re-entry vehicle at the same time, and provide a certain theoretical basis and application reference for the research on re-entry vehicle parameter identification.

[0006] In order to achieve the above technical objectives, in a first aspect, the technical solution of the present invention provides a method for identifying uncertainty in closed-loop attitude dynamics of a reentry vehicle, comprising the following steps:

[0007] A closed-loop attitude dynamics model of the reentry vehicle reentry process including aerodynamic parameter perturbation is constructed, and the attitude control task is sampled at equal intervals to obtain the time series data of the state quantity and control quantity as training data;

[0008] Build a deep learning network to fit unknown perturbations, construct a data-driven model, and add a residual term of the control equation to the loss function of the data-driven model to form a loss function of the physical driving module;

[0009] The deep learning network is trained based on the training data until the loss function of the physical drive module is minimized, thereby obtaining an identification value of the aerodynamic parameter perturbation.

[0010] Compared with the prior art, the method for identifying uncertainty in closed-loop attitude dynamics of a reentry vehicle provided by the present invention has the following beneficial effects:

[0011] In actual flight missions, aircraft models are all closed-loop systems under the action of controllers, and aerodynamic parameter perturbations are all time-varying values. The closed-loop attitude dynamics uncertainty identification method for reentry vehicle provided by the present invention can achieve unknown time-varying parameter targets in the closed-loop system by constructing an aircraft aerodynamic parameter perturbation model and a physical information neural identification network for simulation verification. It can handle unknown parameter estimation tasks in the closed-loop attitude control system of the reentry vehicle, and can effectively identify multiple time-varying aerodynamic parameter perturbations of the reentry vehicle at the same time, providing a certain theoretical basis and application reference for the research on reentry vehicle parameter identification, and has good accuracy and reliability.

[0012] According to some embodiments of the present invention, a closed-loop attitude dynamics model of a reentry vehicle reentry process including aerodynamic parameter perturbations is constructed, comprising the steps of:

[0013] Assuming the vehicle is an ideal rigid body, the reentry thrust is zero, and the attitude angle ignores the influence of the earth's curvature and the earth's rotation, the simplified RV attitude model is as follows:

[0014] ;

[0015] In the formula,

[0016] ;

[0017] ;

[0018] in, They represent the angle of attack, sideslip angle and roll angle respectively; They represent the roll, yaw and pitch angular velocities of the missile system respectively; denote the roll, yaw and pitch moments of inertia respectively; denote the flight path angle, velocity, gravitational acceleration, and vehicle mass, respectively; L and Z represent lift and lateral force respectively, and denote the roll, yaw and pitch control moments respectively; and represent dynamic pressure, reference area and mean aerodynamic chord length respectively; and They represent the aileron angle, rudder angle and elevator angle respectively; are aerodynamic coefficients;

[0019] Since wind tunnel data is different from the actual flight environment, the uncertainties considered in the model are as follows:

[0020] ;

[0021] Indicates the nominal value, is the perturbation variable, and the perturbation range should be limited between -1 and 1; the perturbation values ​​of lift coefficient, pitch moment coefficient and density are set to , and ;

[0022] The state quantity is selected as , , the control input is selected as , the following nonlinear RV attitude control model is obtained

[0023] ;

[0024] in,

[0025] ;

[0026] ;

[0027] ;

[0028] ;

[0029] Using the classic PID method, the attitude angle error is defined as , then the control input is obtained by the following formula

[0030] ;

[0031] in, Indicates attitude angle command, proportional coefficient , integral coefficient , differential coefficient , Represents the controller parameters of the pitch channel.

[0032] According to some embodiments of the present invention, performing equal-interval sampling on the posture control task to obtain time series data of state quantities and control quantities as training data includes the following steps:

[0033] Set the initial attitude angle to , , , the initial angular velocity is , , ; The attitude angle command is , , ; The flight path angle is 0; The control law is limited to ;

[0034] A functional engineering simulator was developed using Matlab / Simulink to build the model. After generating training data using the Euler iteration method with a sampling step of 0.01 seconds, the training data was input into the neural network for parameter identification. The training data included 9 types of data, including three attitude angles, three angular velocities, and three control quantities.

[0035] According to some embodiments of the present invention, building a deep learning network to fit unknown perturbations includes the steps of:

[0036] Define a fully connected feedforward neural network with the same number of fault types; each network consists of: an input layer, multiple hidden layers, and an output layer. The input layer only contains time t, The output layer represents an unknown fault;

[0037] Considering the Layer N Neuron, The input and output forms of the layer are

[0038] ;

[0039] In the formula, is the nonlinear neuron activation function, Indicates The input of the layer, Indicates The output of the layer, is the weight matrix, For the The bias vector corresponding to the layer.

[0040] According to some embodiments of the present invention, constructing a data-driven model includes the steps of:

[0041] The data-driven model includes: 3 deep neural networks, each of which consists of 6 hidden layers. The number of neurons in the hidden layers is 400, 300, 200, 100, 100 and 50 respectively, and the activation function uses the tanh function;

[0042] The input and output of the DNNs module are

[0043] ;

[0044] ;

[0045] Final output of the three deep neural networks Used to approximate .

[0046] According to some embodiments of the present invention, based on the loss function of the data-driven model, a residual term of the control equation is added to form a loss function of the physical driving module, including the steps of:

[0047] Define time interval and represent the initial moment and the terminal moment respectively; and Indicated in Observation data at 1 second, represents the sampling interval, N is the number of sampling points; for the state quantity The difference in time is:

[0048] ;

[0049] time The physical function is defined as:

[0050] ;

[0051] Each subscript f The function represents the differential equation at time The value of each differential equation contains the corresponding fault impact term, and the neural network identifies these parameters in time The value on Respectively expressed in time The state quantity and control quantity;

[0052] The loss function is expressed as:

[0053] .

[0054] According to some embodiments of the present invention, training the deep learning network based on the training data until the loss function of the physical drive module is minimized to obtain the identification value of the aerodynamic parameter perturbation comprises the steps of:

[0055] The training data is substituted into the deep learning network model, the deep learning network model is initialized to generate relevant weights and corresponding output values, and the error function between the output value and the true value is calculated; the result of the error function is optimized by gradient descent, and the optimized correction value is propagated forward to the hidden layer according to the chain rule to correct the weight; after multiple rounds of iterations and weight updates, the optimal aerodynamic parameter perturbation identification value is output.

[0056] In a second aspect, the technical solution of the present invention provides a reentry vehicle closed-loop attitude dynamics uncertainty identification system, comprising:

[0057] The training data acquisition module is used to construct a closed-loop attitude dynamics model of the reentry process of the reentry vehicle including aerodynamic parameter perturbations, and to obtain time series data of state quantities and control quantities by sampling the attitude control tasks at equal intervals as training data;

[0058] A deep learning network building module is used to build a deep learning network to fit unknown perturbations, construct a data-driven model, and add a residual term of the control equation on the basis of the loss function of the data-driven model to form a loss function of the physical driving module;

[0059] The identification value acquisition module trains the deep learning network based on the training data until the loss function of the physical drive module is minimized, thereby obtaining the identification value of the aerodynamic parameter perturbation.

[0060] In a third aspect, the technical solution of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer-executable instructions, and the computer-executable instructions are used to enable a computer to execute a method for identifying uncertainty in closed-loop attitude dynamics of a reentry vehicle as described in any one of the first aspects.

[0061] Additional aspects and advantages of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] The above and / or additional aspects and advantages of the present invention will become apparent and easily understood from the description of the embodiments in conjunction with the following drawings:

[0063] Figure 1 A flow chart of a method for identifying uncertainty in closed-loop attitude dynamics of a reentry vehicle provided by one embodiment of the present invention;

[0064] Figure 2 A flow chart of a parameter identification algorithm design for a closed-loop attitude dynamics uncertainty identification method for a reentry vehicle provided by an embodiment of the present invention;

[0065] Figure 3 A schematic diagram of attitude angle training data for a method for identifying uncertainty in closed-loop attitude dynamics of a reentry vehicle provided by an embodiment of the present invention;

[0066] Figure 4 A schematic diagram of angular velocity training data for a closed-loop attitude dynamics uncertainty identification method for a reentry vehicle provided by an embodiment of the present invention;

[0067] Figure 5 A schematic diagram of control quantity training data for a method for identifying uncertainty in closed-loop attitude dynamics of a reentry vehicle provided by an embodiment of the present invention;

[0068] Figure 6 A schematic diagram of a loss function of an identification result of a reentry vehicle closed-loop attitude dynamics uncertainty identification method provided by an embodiment of the present invention;

[0069] Figure 7 A schematic diagram of lift coefficient perturbation identification results of a closed-loop attitude dynamics uncertainty identification method for a reentry vehicle provided by an embodiment of the present invention;

[0070] Figure 8 A schematic diagram of the pitch moment coefficient perturbation identification results of a reentry vehicle closed-loop attitude dynamics uncertainty identification method provided by an embodiment of the present invention;

[0071] Fig. 9 A schematic diagram of density perturbation identification results of a closed-loop attitude dynamics uncertainty identification method for a reentry vehicle provided in one embodiment of the present invention. DETAILED DESCRIPTION

[0072] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0073] It should be noted that although Figure 3-Figure 9 The functional modules are divided in the system diagram. Figure 1 and Figure 2The flowcharts show the logical sequence, but in some cases, the system may be divided into modules different from the system, or Figure 1 and Figure 2 The steps shown or described are performed in the order in which they are shown in the flowchart. The terms "first", "second", etc. in the specification, claims and the above drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence.

[0074] The embodiments of the present invention are further described below in conjunction with the accompanying drawings.

[0075] Reference Figure 1 , Figure 1 A flow chart of a method for identifying uncertainty in closed-loop attitude dynamics of a re-entry vehicle provided in accordance with an embodiment of the present invention; the method for identifying uncertainty in closed-loop attitude dynamics of a re-entry vehicle includes but is not limited to steps S110 to S130.

[0076] Step S110, constructing a closed-loop attitude dynamics model of the reentry process of the reentry vehicle including aerodynamic parameter perturbations, sampling the attitude control task at equal intervals to obtain time series data of state quantities and control quantities as training data;

[0077] Step S120, building a deep learning network to fit unknown perturbations, constructing a data-driven model, and adding a residual term of the control equation on the basis of the loss function of the data-driven model to form a loss function of the physical driving module;

[0078] Step S130: training the deep learning network based on the training data until the loss function of the physical drive module is minimized, thereby obtaining the identification value of the aerodynamic parameter perturbation.

[0079] In one embodiment, a method for identifying uncertainty in closed-loop attitude dynamics of a reentry vehicle includes the steps of: constructing a closed-loop attitude dynamics model of a reentry vehicle reentry process including aerodynamic parameter perturbations, sampling attitude control tasks at equal intervals to obtain time series data of state quantities and control quantities as training data; building a deep learning network to fit unknown perturbations, constructing a data-driven model, and adding residual terms to the control equation based on the loss function of the data-driven model to form a loss function of a physical drive module; training the deep learning network based on the training data until the loss function of the physical drive module is minimized to obtain an identification value of the aerodynamic parameter perturbations.

[0080] In actual flight missions, aircraft models are all closed-loop systems under the action of controllers, and aerodynamic parameter perturbations are all time-varying values. The closed-loop attitude dynamics uncertainty identification method for reentry vehicle provided by the present invention can achieve unknown time-varying parameter targets in the closed-loop system by constructing an aircraft aerodynamic parameter perturbation model and a physical information neural identification network for simulation verification. It can handle unknown parameter estimation tasks in the closed-loop attitude control system of the reentry vehicle, and can effectively identify multiple time-varying aerodynamic parameter perturbations of the reentry vehicle at the same time, providing a certain theoretical basis and application reference for the research on reentry vehicle parameter identification, and has good accuracy and reliability.

[0081] Reference Figures 2 to 9 , Figure 2 A flow chart of a parameter identification algorithm design for a closed-loop attitude dynamics uncertainty identification method for a reentry vehicle provided by an embodiment of the present invention; Figure 3 A schematic diagram of attitude angle training data for a method for identifying uncertainty in closed-loop attitude dynamics of a reentry vehicle provided by an embodiment of the present invention; Figure 4 A schematic diagram of angular velocity training data for a closed-loop attitude dynamics uncertainty identification method for a reentry vehicle provided by an embodiment of the present invention; Figure 5 A schematic diagram of control quantity training data for a method for identifying uncertainty in closed-loop attitude dynamics of a reentry vehicle provided by an embodiment of the present invention; Figure 6 A schematic diagram of a loss function of an identification result of a reentry vehicle closed-loop attitude dynamics uncertainty identification method provided by an embodiment of the present invention; Figure 7 A schematic diagram of lift coefficient perturbation identification results of a closed-loop attitude dynamics uncertainty identification method for a reentry vehicle provided by an embodiment of the present invention; Figure 8 A schematic diagram of the pitch moment coefficient perturbation identification results of a reentry vehicle closed-loop attitude dynamics uncertainty identification method provided by an embodiment of the present invention; Fig. 9 A schematic diagram of density perturbation identification results of a closed-loop attitude dynamics uncertainty identification method for a reentry vehicle provided in one embodiment of the present invention.

[0082] The uncertainty identification method of closed-loop attitude dynamics of re-entry vehicle based on Euler iterative enhanced physical information neural network is used as a reference for the design process of uncertainty identification of closed-loop attitude dynamics of re-entry vehicle. Figure 2 As shown in FIG. 1 , the uncertainty identification method of closed-loop attitude dynamics of a reentry vehicle based on Euler iterative enhanced physical information neural network includes:

[0083] S1: Establish a closed-loop attitude dynamics model of the reentry vehicle during the reentry process and sample the time series data required for network training;

[0084] Optionally, S1 includes:

[0085] S11: Construct a closed-loop attitude dynamics model of the reentry vehicle during the reentry process including aerodynamic parameter perturbations, and determine the number of unknown parameters;

[0086] The method diagram is as follows Figure 2 As shown in Figure 1, the state and control variables of the closed-loop attitude control system are sampled at equal intervals. In addition, a deep neural network is designed to estimate the unknown parameters. Subsequently, the loss value is calculated based on the sampled variables and the fitting results of the three neural networks, and the optimal identification of the unknown quantity is achieved by minimizing the loss function.

[0087] Assuming that the aircraft is an ideal rigid body, the reentry thrust is zero, the attitude angle is small, and the influence of the earth's curvature and rotation is ignored, the simplified RV attitude model is as follows:

[0088] ;

[0089] In the formula,

[0090] ;

[0091] in, They represent the angle of attack, sideslip angle and roll angle respectively; They represent the roll, yaw and pitch angular velocities of the missile system respectively; denote the roll, yaw and pitch moments of inertia respectively; They represent the flight path angle, velocity, gravitational acceleration, and vehicle mass, respectively. L and Z represents lift and lateral force, and represents the roll, yaw and pitch control moments. At the same time, , and represents dynamic pressure, reference area and mean aerodynamic chord length. and Indicates aileron, rudder and elevator. represents the aerodynamic coefficient.

[0092] Since wind tunnel data is different from the actual flight environment, aerodynamic parameters inevitably have uncertainties. Therefore, in this embodiment, the uncertainties considered in the model are as follows:

[0093] ;

[0094] Indicates the nominal value, represents the perturbation amount, the perturbation range should be limited to between -1 and 1, and the perturbation value is unknown. In this embodiment, the perturbation values ​​of the lift coefficient, pitch moment coefficient and density are set to and .

[0095] The state quantity is selected as , , the control input is selected as , we can get the following nonlinear RV attitude control model

[0096] ;

[0097] in,

[0098] ;

[0099] ;

[0100] ;

[0101] .

[0102] This embodiment adopts the classic PID method, and the attitude angle error is defined as , then the control input is obtained by the following formula

[0103] ;

[0104] in, Indicates attitude angle command, proportional coefficient , integral coefficient , differential coefficient Ensure system stability. represents the controller parameters of the pitch channel, and the same is true for the yaw and roll channels. In this embodiment, the error differential term is replaced by angular velocity.

[0105] S12: Sampling at equal intervals to obtain time series data of state quantities and control quantities.

[0106] The training data comes from the complete posture control task. In this embodiment, the initial posture angle is set to , , , the initial angular velocity is , , ; The attitude angle command is , , ; The flight path angle is 0; The control law, i.e. the rudder angle change limit is .

[0107] The model was built by developing a functional engineering simulator using Matlab / Simulink. After generating the training data using the Euler iteration method with a sampling step of 0.01 seconds, the data was input into the neural network for parameter identification. The entire simulated flight lasted for 10 seconds. After the above processing and calculation, the relevant training set is as follows Figures 3-5 As shown, there are 9 types of data in total, including three attitude angles, three angular velocities and three control quantities. The data set is saved in csv format, with a total of 1000 rows and 9 columns.

[0108] S2: Establish Euler iterative enhanced physical information neural network according to task requirements;

[0109] Step S2 includes:

[0110] S21: Build a deep learning network to fit unknown perturbations and construct a data-driven module. The data-driven module adopts a neural network architecture and sets parameters, including the number of hidden layers, the number of neurons in each layer, and the mapping function between layers.

[0111] The structure of PINN mainly relies on a fully connected feedforward neural network. Theoretically, a multi-layer feedforward network with enough hidden layer neurons can approximate any predetermined continuous function with arbitrarily high accuracy. Therefore, a fully connected feedforward neural network with the same number of fault types is defined. Each network consists of an input layer (containing only time t ), multiple hidden layers, and an output layer (representing an unknown fault).

[0112] Considering the Layer N Neuron, The input and output forms of the layer are

[0113] ;

[0114] In the formula, represents the nonlinear neuron activation function, Indicates The input of the layer, Indicates The output of the layer, represents the weight matrix, Indicates The bias vector corresponding to the layer.

[0115] Generally speaking, there are optimal values ​​for the number of hidden layers and neurons for the corresponding tasks. The network parameter initialization and the type of neural network activation function will affect the accuracy of the results. The empirical test method is used to find the neural test parameters.

[0116] Corresponding to this embodiment, the data driven model is composed of Figure 2 The three DNNs shown in are composed of 6 hidden layers. The number of neurons in each layer is 400, 300, 200, 100, 100 and 50 respectively, and the activation function uses the tanh function.

[0117] Figure 2 The input and output of the DNNs module are

[0118] ;

[0119] ;

[0120] Therefore, the final output of the three DNNs is Used to approximate .

[0121] S22: construct a loss function. On the basis of the loss function of the data-driven module, add the residual term of the control equation to form the loss function of the physical-driven module.

[0122] In the classic PINN, the physical mechanism model is embedded in the loss function, thereby constraining the fitting of the neural network to a specific model. Due to the lack of reliable analytical solutions to the equations, an automatic differentiator is usually used to obtain estimated derivative values ​​when calculating the loss function. However, as far as the identification process is concerned, the data of the entire flight process is known. Therefore, the discretization-based Euler iteration is used to obtain the difference of all sampling moments and incorporate it into the loss function calculation as the derivative value at the corresponding moment.

[0123] Define time interval , and represent the initial time and the terminal time respectively. and Indicated in Observation data at 1 second, represents the sampling interval, N is the number of sampling points. Therefore, in the embodiment, for the state quantity The difference in time is:

[0124] ;

[0125] The existing PINN incorporates the difference between the physical equation before and after iteration into the loss function of the neural network. Incorporating the physical equation into the training process can ensure that the final training result conforms to the laws of physics. The physical function is defined as:

[0126] ;

[0127] Each subscript f The function represents the differential equation at time Each differential equation contains the corresponding fault impact term, and the neural network will identify these parameters in time The value on . Respectively expressed in time state quantity and control quantity.

[0128] The loss function is expressed as:

[0129] ;

[0130] Introducing physical terms into the loss function F , construct the PINN framework. The ordinary differential equation structural loss in the classic PINN is given by the difference between the result obtained by automatic differentiation and the calculated derivative value obtained by substituting the observed data into the model. The estimate of the derivative in the loss function of the physical information neural network based on Euler iteration is calculated based on the forward Euler iteration using the observations of the current and subsequent times, rather than relying on automatic differentiation. Finally, in order to minimize the loss function, the Adam algorithm is used to update the parameters of the DNNs during the training process.

[0131] S3: Train the neural network based on the sampled data until the loss function is minimized, obtain the identification value of the aerodynamic parameter perturbation, and achieve the mission objectives.

[0132] The set of network parameters with the highest model prediction accuracy. All neural networks were trained using the Adam optimizer with an initial learning rate of 0.001 and a decay rate of 95% every 2000 times. In addition, the regularization coefficient ω was set to 0.0001 and the number of training times was set to 200,000.

[0133] The training process of the neural network model is as follows: substitute the data of the training set into the neural network model, initialize the neural network model to generate relevant weights and corresponding output values, calculate the error function between the output value and the true value, where the error function is given by S22; perform gradient descent optimization on the result of the error function, propagate the optimized correction value forward to the hidden layer according to the chain rule, and correct the weight; after multiple rounds of iterations and weight updates, output the optimal aerodynamic parameter perturbation identification value to complete the training of the model. Figures 6 to 9 The control response curve of the dynamic closed loop of the reentry vehicle under the PID controller, the change of the loss function during the identification process, and the identification results of three uncertain disturbances are shown. It can be seen that all three disturbances are successfully identified with very small errors. Therefore, it can be considered that this method can handle the unknown parameter estimation task in the closed-loop attitude control system of the reentry vehicle.

[0134] Based on the above technical scheme, the uncertainty identification method of closed-loop attitude dynamics of reentry vehicle based on Euler iteration enhanced physical information neural network provided by the present invention is verified by constructing an aircraft aerodynamic parameter perturbation model and a physical information neural identification network, and finally it is proved that the proposed method can achieve the unknown time-varying parameter target in the closed-loop system.

[0135] In one embodiment, a method for identifying uncertainty in closed-loop attitude dynamics of a reentry vehicle includes the following steps: constructing a closed-loop attitude dynamics model of a reentry vehicle reentry process including aerodynamic parameter perturbations, sampling attitude control tasks at equal intervals to obtain time series data of state quantities and control quantities as training data; building a deep learning network to fit unknown perturbations, constructing a data-driven model, and adding residual terms to the control equation based on the loss function of the data-driven model to form a loss function of a physical drive module; training the deep learning network based on the training data until the loss function of the physical drive module is minimized to obtain an identification value of the aerodynamic parameter perturbations.

[0136] The deep learning network is trained based on the training data until the loss function of the physical drive module is minimized to obtain the identification value of the aerodynamic parameter perturbation, including the following steps: substituting the training data into the deep learning network model, initializing the deep learning network model to generate relevant weights and corresponding output values, and calculating the error function between the output value and the true value; performing gradient descent optimization on the result of the error function, forward propagating the optimized correction value to the hidden layer according to the chain rule, and correcting the weight; after multiple rounds of iterations and weight updates, outputting the optimal aerodynamic parameter perturbation identification value.

[0137] In one embodiment, a closed-loop attitude dynamics uncertainty identification system for a reentry vehicle includes: a training data acquisition module, which is used to construct a closed-loop attitude dynamics model of a reentry vehicle reentry process including aerodynamic parameter perturbations, and to perform equal-interval sampling on attitude control tasks to obtain time series data of state quantities and control quantities as training data; a deep learning network construction module, which is used to build a deep learning network to fit unknown perturbations, construct a data-driven model, and add residual terms of the control equation based on the loss function of the data-driven model to form a loss function of the physical drive module; an identification value acquisition module, which trains the deep learning network based on the training data until the loss function of the physical drive module is minimized to obtain the identification value of the aerodynamic parameter perturbations.

[0138] As a non-transient computer-readable storage medium, the memory can be used to store non-transient software programs and non-transient computer executable programs. In addition, the memory may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some embodiments, the memory may optionally include a memory remotely disposed relative to the processor, and these remote memories may be connected to the processor via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0139] The device embodiments described above are merely illustrative, and the units described as separate components may or may not be physically separated, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0140] In addition, an embodiment of the present invention also provides a computer-readable storage medium, which stores computer-executable instructions, and the computer-executable instructions are executed by a processor or controller, for example, by a processor in the above-mentioned terminal embodiment, so that the above-mentioned processor can execute the re-entry vehicle closed-loop attitude dynamics uncertainty identification method in the above-mentioned embodiment.

[0141] It will be appreciated by those skilled in the art that all or some of the steps and systems in the methods disclosed above may be implemented as software, firmware, hardware, and appropriate combinations thereof. Some or all physical components may be implemented as software executed by a processor, such as a central processing unit, a digital signal processor, or a microprocessor, or as hardware, or as an integrated circuit, such as an application-specific integrated circuit. Such software may be distributed on a computer-readable medium, which may include a computer storage medium (or non-transitory medium) and a communication medium (or transient medium). As known to those skilled in the art, the term computer storage medium includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules, or other data). Computer storage media include, but are not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital versatile disk (DVD) or other optical disk storage, magnetic cassettes, magnetic tapes, disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and can be accessed by a computer. Furthermore, it is well known to those skilled in the art that communication media typically embodies computer readable instructions, data structures, program modules, or other data in a modulated data signal such as a carrier wave or other transport mechanism, and may include any information delivery media.

[0142] The above is a specific description of the preferred implementation of the present invention, but the present invention is not limited to the above-mentioned implementation mode. Technical personnel familiar with the field can also make various equivalent deformations or substitutions without violating the spirit of the present invention. These equivalent deformations or substitutions are all included in the scope defined by the claims of the present invention.

[0143] The specific implementation of the present invention described above does not constitute a limitation on the protection scope of the present invention. Any other corresponding changes and modifications made based on the technical concept of the present invention should be included in the protection scope of the claims of the present invention.

Claims

1. A method for identifying uncertainty in closed-loop attitude dynamics of a reentry vehicle, characterized in that: The following steps are involved: A closed-loop attitude dynamics model of the reentry vehicle reentry process including aerodynamic parameter perturbation is constructed, and the attitude control task is sampled at equal intervals to obtain the time series data of the state quantity and control quantity as training data; Since wind tunnel data is different from the actual flight environment, the uncertainties considered in the model are as follows: ; Indicates the nominal value, is the perturbation variable, and the perturbation range should be limited between -1 and 1; the perturbation values ​​of lift coefficient, pitch moment coefficient and density are set to , and ; Build a deep learning network to fit unknown perturbations, construct a data-driven model, and add a residual term of the control equation to the loss function of the data-driven model to form a loss function of the physical driving module; Based on the training data, the deep learning network is trained until the loss function of the physical driving module is minimized, thereby obtaining an identification value of the aerodynamic parameter perturbation; Among them, building a deep learning network to fit unknown perturbations includes the following steps: Define a fully connected feedforward neural network with the same number of fault types; each network consists of: an input layer, multiple hidden layers, and an output layer. The input layer only contains time t, The output layer represents an unknown fault; Considering the Layer N Neuron, The input and output forms of the layer are: ; In the formula, is the nonlinear neuron activation function, Indicates The input of the layer, Indicates The output of the layer, is the weight matrix, For the The bias vector corresponding to the layer; Building a data-driven model includes the following steps: The data-driven model includes: 3 deep neural networks, each of which consists of 6 hidden layers. The number of neurons in the hidden layers is 400, 300, 200, 100, 100 and 50 respectively, and the activation function uses the tanh function; The input and output of the DNNs module are ; ; Final output of the three deep neural networks Used to approximate ; On the basis of the loss function of the data-driven model, a residual term of the control equation is added to form a loss function of the physical driving module, including the steps of: Define time interval and represent the initial moment and the terminal moment respectively; and Indicated in Observation data at 1 second, represents the sampling interval, N is the number of sampling points; for the state quantity The difference in time is: ; time The physical function is defined as: ; Each subscript f The function represents the differential equation at time The value of each differential equation contains the corresponding fault impact term, and the neural network identifies these parameters in time The value on Respectively, in time The state quantity and control quantity; The loss function is expressed as: 。 2. The method for identifying uncertainty in closed-loop attitude dynamics of a reentry vehicle according to claim 1, characterized in that: Constructing a closed-loop attitude dynamics model of the reentry vehicle reentry process including aerodynamic parameter perturbations, including the following steps: Assuming the vehicle is an ideal rigid body, the reentry thrust is zero, and the attitude angle ignores the influence of the earth's curvature and the earth's rotation, the simplified RV attitude model is as follows: ; In the formula, ; ; in, They represent the angle of attack, sideslip angle and roll angle respectively; They represent the roll, yaw and pitch angular velocities of the missile system respectively; denote the roll, yaw and pitch moments of inertia respectively; denote the flight path angle, velocity, gravitational acceleration, and vehicle mass, respectively; L and Z represent lift and lateral force respectively, and denote the roll, yaw and pitch control moments respectively; and represent dynamic pressure, reference area and mean aerodynamic chord length respectively; and They represent the aileron angle, rudder angle and elevator angle respectively; are aerodynamic coefficients; The state quantity is selected as , , the control input is selected as , the following nonlinear RV attitude control model is obtained ; in, ; ; ; ; Using the classic PID method, the attitude angle error is defined as , then the control input is obtained by the following formula ; in, Indicates attitude angle command, proportional coefficient , integral coefficient , differential coefficient , Represents the controller parameters of the pitch channel.

3. The method for identifying uncertainty in closed-loop attitude dynamics of a reentry vehicle according to claim 2, characterized in that: The attitude control task is sampled at equal intervals to obtain the time series data of the state quantity and the control quantity as training data, including the following steps: Set the initial attitude angle to , , , the initial angular velocity is , , ; The attitude angle command is , , ; The flight path angle is 0; The control law is limited to ; A functional engineering simulator was developed using Matlab / Simulink to build the model. After generating training data using the Euler iteration method with a sampling step of 0.01 seconds, the training data was input into the neural network for parameter identification. The training data included 9 types of data, including three attitude angles, three angular velocities, and three control quantities.

4. The method for identifying uncertainty in closed-loop attitude dynamics of a reentry vehicle according to claim 1, characterized in that: Training the deep learning network based on the training data until the loss function of the physical drive module is minimized, and obtaining the identification value of the aerodynamic parameter perturbation, comprising the steps of: The training data is substituted into the deep learning network model, the deep learning network model is initialized to generate relevant weights and corresponding output values, and the error function between the output value and the true value is calculated; the result of the error function is optimized by gradient descent, and the optimized correction value is propagated forward to the hidden layer according to the chain rule to correct the weight; after multiple rounds of iterations and weight updates, the optimal aerodynamic parameter perturbation identification value is output.

5. A closed-loop attitude dynamics uncertainty identification system for a reentry vehicle, characterized in that: include: The training data acquisition module is used to construct a closed-loop attitude dynamics model of the reentry process of the reentry vehicle including aerodynamic parameter perturbations, and to obtain time series data of state quantities and control quantities by sampling the attitude control tasks at equal intervals as training data; Since wind tunnel data is different from the actual flight environment, the uncertainties considered in the model are as follows: ; Indicates the nominal value, is the perturbation variable, and the perturbation range should be limited between -1 and 1; the perturbation values ​​of lift coefficient, pitch moment coefficient and density are set to , and ; A deep learning network building module is used to build a deep learning network to fit unknown perturbations, construct a data-driven model, and add a residual term of the control equation on the basis of the loss function of the data-driven model to form a loss function of the physical driving module; An identification value acquisition module, which trains the deep learning network based on the training data until the loss function of the physical driving module is minimized, thereby obtaining an identification value of the aerodynamic parameter perturbation; Among them, when the deep learning network building module builds a deep learning network to fit unknown perturbations, it includes: Define a fully connected feedforward neural network with the same number of fault types; each network consists of: an input layer, multiple hidden layers, and an output layer. The input layer only contains time t, The output layer represents an unknown fault; Considering the Layer N Neuron, The input and output forms of the layer are: ; In the formula, is the nonlinear neuron activation function, Indicates The input of the layer, Indicates The output of the layer, is the weight matrix, For the The bias vector corresponding to the layer; When building a data-driven model, the deep learning network building module includes: The data-driven model includes: 3 deep neural networks, each of which consists of 6 hidden layers. The number of neurons in the hidden layers is 400, 300, 200, 100, 100 and 50 respectively, and the activation function uses the tanh function; The input and output of the DNNs module are ; ; Final output of the three deep neural networks Used to approximate ; The deep learning network building module adds the residual term of the control equation to the loss function of the data-driven model to form the loss function of the physical driving module, including: Define time interval and represent the initial moment and the terminal moment respectively; and Indicated in Observation data at 1 second, represents the sampling interval, N is the number of sampling points; for the state quantity The difference in time is: ; time The physical function is defined as: ; Each subscript f The function represents the differential equation at time The value of each differential equation contains the corresponding fault impact term, and the neural network identifies these parameters in time The value on Respectively, in time The state quantity and control quantity; The loss function is expressed as: 。 6. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, and the computer-executable instructions are used to enable a computer to execute the reentry vehicle closed-loop attitude dynamics uncertainty identification method as described in any one of claims 1 to 4.