A method, system, device and medium for identifying an unsteady aerodynamic model
By using attitude dynamics networks and loss function optimization, the problem of insufficient identification accuracy and generalization ability of existing unsteady aerodynamic models is solved, and effective identification and accurate prediction of unsteady aerodynamic models are achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- CALCULATION AERODYNAMICS INST CHINA AERODYNAMICS RES & DEV CENT
- Filing Date
- 2026-04-15
- Publication Date
- 2026-06-23
AI Technical Summary
Existing methods for identifying unsteady aerodynamic models suffer from problems such as decreased model accuracy, lack of physical meaning, and limited generalization ability, especially in high-order or nonlinear cases where effective identification is difficult.
An attitude dynamics network is employed. By acquiring a training dataset, an unsteady aerodynamic model prediction network is constructed. An angle-of-attack sequence is input into the attitude dynamics network, and combined with integrator and loss function optimization, the unsteady aerodynamic model can be identified.
It enables effective identification of unsteady aerodynamic models, and the predicted angular velocity can be directly applied in engineering, improving the accuracy and generalization ability of the model.
Smart Images
Figure CN122065731B_ABST
Abstract
Description
Technical Field
[0001] This application relates to the fields of deep learning and aerodynamics, and in particular to a method, system, device and medium for identifying unsteady aerodynamic models. Background Technology
[0002] Existing methods for identifying unsteady aerodynamic models are mainly divided into two categories: traditional mathematical modeling methods and modern data-driven methods.
[0003] Traditional mathematical modeling methods are based on an understanding of the physical mechanisms of unsteady aerodynamics, establishing a mathematical relationship between aerodynamic forces and flight states. These mainly include frequency domain methods, step response model methods, state-space model methods, and differential equation model methods. For frequency domain methods, aerodynamic coefficients are obtained through forced vibration experiments, and Fourier decomposition is performed at reduced frequencies to identify in-phase or out-of-phase derivatives and characteristic time constants. This method is suitable for identifying unsteady aerodynamic forces under the assumption of small amplitude linearity. Step response model methods establish a dynamic relationship between aerodynamic forces and the input by measuring the transient response of aerodynamic forces to a unit step input. The core of this method is to use a step function or step response function to describe the time-varying process of aerodynamic forces. However, this method suffers from decreased accuracy in higher-order or nonlinear cases if the step input fails to excite all the dynamics of the system. State-space models struggle to capture strongly nonlinear flow characteristics, and the physical meaning of the parameters in differential equation models is unclear, making them unsuitable for aerodynamic modeling of large-amplitude, multi-degree-of-freedom coupled motions.
[0004] Modern data-driven methods mainly include decision tree methods, BP neural networks, RBF neural network modeling methods, fuzzy logic modeling methods, Kriging methods, Gaussian process methods, recurrent neural networks, etc. Data-driven methods mainly have shortcomings such as lack of internal physical meaning of the model, high requirements for data volume and data quality, and limited generalization ability. Summary of the Invention
[0005] In view of this, this application provides a method, system, device and medium for identifying unsteady aerodynamic models.
[0006] This application discloses a method for identifying unsteady aerodynamic models, which includes:
[0007] Step 1: Obtain the training dataset; each training sample in the training dataset includes the aircraft's angular velocity, steady aerodynamic torque coefficient, dynamic pressure matrix, and angular velocity label value;
[0008] Step 2: Input the aircraft's angle of attack sequence into the attitude dynamics network to obtain the predicted value of the aircraft's angular velocity at the next moment; the attitude dynamics network includes an unsteady aerodynamic model prediction network, an attitude dynamics model, and an integrator connected in sequence;
[0009] Step 3: Based on the next moment's angular velocity label value, the predicted value of the aircraft's next moment's angular velocity, and the unsteady aerodynamic model prediction network parameters, construct the loss function of the attitude dynamics network; input the training dataset into the attitude dynamics network for training, with the goal of minimizing the loss function, and output the optimized unsteady aerodynamic model prediction network parameters and the attitude dynamics network model; the attitude dynamics network model is used to identify unsteady aerodynamic models.
[0010] Further, step 2 includes:
[0011] Step 21: Input the aircraft's angle of attack sequence into the unsteady aerodynamic model prediction network to obtain the torques;
[0012] Step 22: Combine the torque and input attitude dynamics model to obtain the predicted angular acceleration value of the aircraft. ;
[0013] Step 23: Predict the angular acceleration value of the aircraft The input is integrated using an integrator to obtain the predicted value of the aircraft's angular velocity at the next moment. .
[0014] Further, step 21 includes:
[0015] Step 211: Construct an unsteady aerodynamic model;
[0016]
[0017] in, This represents the longitudinal pitch unsteady aerodynamic moment. This represents the angle of attack of the aircraft at the current time t. This represents the angle of attack of the aircraft at the previous P time points. Indicates the aircraft's angle of attack sequence. This represents the mapping relationship from the angle of attack sequence to the pitch unsteady aerodynamic moment;
[0018] Step 212: Set the aircraft's angle of attack sequence Input an unsteady aerodynamic model prediction network, and output the longitudinal pitch unsteady aerodynamic moment. Predicted value ;
[0019] Step 213: Calculate the predicted values of the longitudinal pitch unsteady aerodynamic moments. With steady aerodynamic torque coefficient Add them together to get the sum of the torques.
[0020] Further, step 22 includes:
[0021] The predicted value of the aircraft's angular acceleration is obtained using the following formula. :
[0022]
[0023] in, This represents the predicted angular acceleration of the aircraft. Represents the dynamic pressure matrix. Q Indicates dynamic pressure. s Indicates the reference area. b Indicates the lateral reference length. c Indicates the longitudinal reference length. Represents a diagonal matrix. Represents the steady aerodynamic torque coefficient. , This represents the predicted value of the unsteady aerodynamic moment coefficient. This represents the predicted value of the longitudinal pitch unsteady aerodynamic moment coefficient. I This represents the moment of inertia matrix of the aircraft. This indicates the current angular velocity of the aircraft.
[0024] Further, step 23 includes:
[0025] The predicted angular velocity of the aircraft at the next moment is obtained by integrating the predicted angular acceleration value of the aircraft using the following formula:
[0026]
[0027] in, This represents the predicted angular velocity of the aircraft at the next moment. Indicates the integral symbol, This represents the predicted angular acceleration of the aircraft. Indicates time.
[0028] Furthermore, the unsteady aerodynamic model prediction network includes an input layer, a hidden layer, and an output layer connected in sequence;
[0029] The input layer is used to input the angle of attack sequence of the aircraft, the hidden layer is used to perform a nonlinear transformation on the angle of attack sequence of the aircraft to obtain the predicted value of the longitudinal pitch unsteady aerodynamic moment, and the output layer is used to output the predicted value of the longitudinal pitch unsteady aerodynamic moment.
[0030] Further, step 3 includes:
[0031] Step 31: Construct the loss function:
[0032] The expression for the loss function of the attitude dynamics network is:
[0033]
[0034] in, Represents the loss function. Indicates the angular velocity label value at the next moment. And the predicted value of the angular velocity of the aircraft at the next moment The mean square error between them It's a hyperparameter. This is used to reduce the complexity of prediction networks for unsteady aerodynamic models and improve their generalization ability. This represents the parameter vector of the unsteady aerodynamic model prediction network. express The i-th unsteady aerodynamic model prediction network parameter in the diagram, where n represents the parameter vector. The number of parameters in the middle, Indicates the cumulative symbol, angular velocity label value. Used to indicate the aircraft's true angular velocity at the next moment;
[0035] Step 32: Obtain the network gradient of the loss function based on the loss function:
[0036]
[0037] in, The network gradient represents the loss function. Represents the parameter vector Find the gradient;
[0038] Step 33: During each iteration of training the attitude dynamics network, based on the calculated network gradient, the parameters of the unsteady aerodynamic model prediction network are adjusted using the gradient descent algorithm to minimize the loss function. The updated unsteady aerodynamic model prediction network parameters are then used to replace the parameters of the unsteady aerodynamic model prediction network from the previous iteration. When the preset number of iterations is reached, the optimized unsteady aerodynamic model prediction network parameters and the attitude dynamics network model are output.
[0039] This application also discloses an identification system for unsteady aerodynamic models, implementing the aforementioned method for identifying unsteady aerodynamic models, comprising:
[0040] The data acquisition module is used to acquire the training dataset; each training sample in the training dataset includes the aircraft's angular velocity, steady aerodynamic torque coefficient, dynamic pressure matrix, and angular velocity label value.
[0041] The network construction module is used to input the aircraft's angle of attack sequence into the attitude dynamics network to obtain the predicted value of the aircraft's angular velocity at the next moment; the attitude dynamics network includes an unsteady aerodynamic model prediction network, an attitude dynamics model, and an integrator connected in sequence.
[0042] The function construction module is used to construct the loss function of the attitude dynamics network based on the next moment's angular velocity label value, the predicted value of the aircraft's next moment's angular velocity, and the unsteady aerodynamic model prediction network parameters.
[0043] The network training module is used to input the training dataset into the attitude dynamics network for training, with the goal of minimizing the loss function, and output the optimized unsteady aerodynamic model prediction network parameters and attitude dynamics network model; the attitude dynamics network model is used to identify unsteady aerodynamic models.
[0044] This application also discloses an electronic device, including a memory, a processor, and a computer program stored in the memory and executed by the processor, wherein the processor executes the computer program to implement the above-described method for identifying unsteady aerodynamic models.
[0045] This application also discloses a non-transitory computer-readable storage medium that stores computer instructions for causing a computer to execute the above-described method for identifying unsteady aerodynamic models.
[0046] Due to the adoption of the above technical solution, this application has the following advantages: This application obtains the angular velocity prediction by integrating the loss function, and the label value of the angular velocity prediction can be directly measured in engineering applications; The network input in this application is the time series data of the angle of attack, which can effectively identify unsteady aerodynamic models, while the existing deep networks with embedded attitude dynamics can only identify steady aerodynamic models. Attached Figure Description
[0047] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments recorded in the embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings.
[0048] Figure 1 This is a schematic diagram of the attitude dynamics network structure according to an embodiment of this application. Detailed Implementation
[0049] The present application will be further described in conjunction with the accompanying drawings and embodiments. The described embodiments are only some, not all, of the embodiments of the present application. All other embodiments obtained by those skilled in the art should fall within the protection scope of the embodiments of the present application.
[0050] See Figure 1This application provides an embodiment of a method for identifying unsteady aerodynamic models, which includes:
[0051] Step 1: Obtain the training dataset; each training sample in the training dataset includes the aircraft's angular velocity, steady aerodynamic torque coefficient, dynamic pressure matrix, and angular velocity label value;
[0052] Step 2: Input the aircraft's angle of attack sequence into the attitude dynamics network to obtain the predicted value of the aircraft's angular velocity at the next moment; the attitude dynamics network includes an unsteady aerodynamic model prediction network, an attitude dynamics model, and an integrator connected in sequence;
[0053] Step 3: Based on the next moment's angular velocity label value, the predicted value of the aircraft's next moment's angular velocity, and the unsteady aerodynamic model prediction network parameters, construct the loss function of the attitude dynamics network; input the training dataset into the attitude dynamics network for training, with the goal of minimizing the loss function, and output the optimized unsteady aerodynamic model prediction network parameters and the attitude dynamics network model; the attitude dynamics network model is used to identify unsteady aerodynamic models.
[0054] In one embodiment of this application, step 2 includes:
[0055] Step 21: Input the aircraft's angle of attack sequence into the unsteady aerodynamic model prediction network to obtain the torques;
[0056] Step 22: Combine the torque and input attitude dynamics model to obtain the predicted angular acceleration value of the aircraft. ;
[0057] Step 23: Predict the angular acceleration value of the aircraft The input is integrated using an integrator to obtain the predicted value of the aircraft's angular velocity at the next moment. .
[0058] In one embodiment of this application, step 21 includes:
[0059] Step 211: Construct an unsteady aerodynamic model;
[0060]
[0061] in, This represents the longitudinal pitch unsteady aerodynamic moment. This represents the angle of attack of the aircraft at the current time t. This represents the angle of attack of the aircraft at the previous P time points. Indicates the aircraft's angle of attack sequence. This represents the mapping relationship from the angle of attack sequence to the pitch unsteady aerodynamic moment;
[0062] Step 212: Set the aircraft's angle of attack sequence Input an unsteady aerodynamic model prediction network, and output the longitudinal pitch unsteady aerodynamic moment. Predicted value ;
[0063] Step 213: Calculate the predicted values of the longitudinal pitch unsteady aerodynamic moments. With steady aerodynamic torque coefficient Add them together to get the sum of the torques.
[0064] In one embodiment of this application, step 22 includes:
[0065] The predicted value of the aircraft's angular acceleration is obtained using the following formula. :
[0066]
[0067] in, This represents the predicted angular acceleration of the aircraft. Represents the dynamic pressure matrix. Q Indicates dynamic pressure. s Indicates the reference area. b Indicates the lateral reference length. c Indicates the longitudinal reference length. Represents a diagonal matrix. Represents the steady aerodynamic torque coefficient. , This represents the predicted value of the unsteady aerodynamic moment coefficient. This represents the predicted value of the longitudinal pitch unsteady aerodynamic moment coefficient. I This represents the moment of inertia matrix of the aircraft. This indicates the current angular velocity of the aircraft.
[0068] Figure 1 middle, , It represents the aerodynamic torque.
[0069] In one embodiment of this application, step 23 includes:
[0070] The predicted angular velocity of the aircraft at the next moment is obtained by integrating the predicted angular acceleration value of the aircraft using the following formula:
[0071]
[0072] in, This represents the predicted angular velocity of the aircraft at the next moment. Indicates the integral symbol, This represents the predicted angular acceleration of the aircraft. Indicates time.
[0073] In one embodiment of this application, the unsteady aerodynamic model prediction network includes an input layer, a hidden layer, and an output layer connected in sequence.
[0074] The input layer is used to input the angle of attack sequence of the aircraft, the hidden layer is used to perform a nonlinear transformation on the angle of attack sequence of the aircraft to obtain the predicted value of the longitudinal pitch unsteady aerodynamic moment, and the output layer is used to output the predicted value of the longitudinal pitch unsteady aerodynamic moment.
[0075] In one embodiment of this application, step 3 includes:
[0076] Step 31: Construct the loss function:
[0077] The expression for the loss function of the attitude dynamics network is:
[0078]
[0079] in, Represents the loss function. Indicates the angular velocity label value at the next moment. And the predicted value of the angular velocity of the aircraft at the next moment The mean square error between them It's a hyperparameter. This is used to reduce the complexity of prediction networks for unsteady aerodynamic models and improve their generalization ability. This represents the parameter vector of the unsteady aerodynamic model prediction network. express The i-th unsteady aerodynamic model prediction network parameter in the diagram, where n represents the parameter vector. The number of parameters in the middle, Indicates the cumulative symbol, angular velocity label value. Used to indicate the aircraft's true angular velocity at the next moment;
[0080] Step 32: Obtain the network gradient of the loss function based on the loss function:
[0081]
[0082] in, The network gradient represents the loss function. Represents the parameter vector Find the gradient;
[0083] Step 33: During each iteration of training the attitude dynamics network, based on the calculated network gradient, the parameters of the unsteady aerodynamic model prediction network are adjusted using the gradient descent algorithm to minimize the loss function. The updated unsteady aerodynamic model prediction network parameters are then used to replace the parameters of the unsteady aerodynamic model prediction network from the previous iteration. When the preset number of iterations is reached, the optimized unsteady aerodynamic model prediction network parameters and the attitude dynamics network model are output.
[0084] This application also provides an identification system for unsteady aerodynamic models, implementing the identification method for unsteady aerodynamic models described in the above embodiments, which includes:
[0085] The data acquisition module is used to acquire the training dataset; each training sample in the training dataset includes the aircraft's angular velocity, steady aerodynamic torque coefficient, dynamic pressure matrix, and angular velocity label value.
[0086] The network construction module is used to input the aircraft's angle of attack sequence into the attitude dynamics network to obtain the predicted value of the aircraft's angular velocity at the next moment; the attitude dynamics network includes an unsteady aerodynamic model prediction network, an attitude dynamics model, and an integrator connected in sequence.
[0087] The function construction module is used to construct the loss function of the attitude dynamics network based on the next moment's angular velocity label value, the predicted value of the aircraft's next moment's angular velocity, and the unsteady aerodynamic model prediction network parameters.
[0088] The network training module is used to input the training dataset into the attitude dynamics network for training, with the goal of minimizing the loss function, and output the optimized unsteady aerodynamic model prediction network parameters and attitude dynamics network model; the attitude dynamics network model is used to identify unsteady aerodynamic models.
[0089] This application also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executed by the processor. When the processor executes the computer program, it implements the method for identifying unsteady aerodynamic models described in the above embodiments.
[0090] This application also provides a non-transitory computer-readable storage medium that stores computer instructions for causing a computer to execute the unsteady aerodynamic model identification method described in the above embodiments.
[0091] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application and not to limit them. Although this application has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of this application. Any modifications or equivalent substitutions that do not depart from the spirit and scope of this application should be covered within the protection scope of the claims of this application.
Claims
1. A method of identifying an unsteady aerodynamic model, characterized in that, include: Step 1: Obtain the training dataset; each training sample in the training dataset includes the aircraft's angular velocity, steady aerodynamic torque coefficient, dynamic pressure matrix, and angular velocity label value; Step 2: Input the aircraft's angle of attack sequence into the attitude dynamics network to obtain the predicted value of the aircraft's angular velocity at the next moment; the attitude dynamics network includes an unsteady aerodynamic model prediction network, an attitude dynamics model, and an integrator connected in sequence; Step 3: Based on the angular velocity label value at the next moment, the predicted angular velocity value of the aircraft at the next moment, and the parameters of the unsteady aerodynamic model prediction network, construct the loss function of the attitude dynamics network; input the training dataset into the attitude dynamics network for training, with the goal of minimizing the loss function, and output the optimized parameters of the unsteady aerodynamic model prediction network and the attitude dynamics network model; The attitude dynamics network model is used to identify unsteady aerodynamic models; Step 2 includes: Step 21: Input the aircraft's angle of attack sequence into the unsteady aerodynamic model prediction network to obtain the torque; Step 22: Use the moment and input attitude dynamics model to obtain an aircraft angular acceleration prediction ; Step 23: Predict the vehicle angular acceleration Integrate the input integrator to obtain the predicted value of the vehicle angular velocity at the next time ; Step 3 includes: Step 31: Construct the loss function: The expression for the loss function of the attitude dynamics network is: in, Represents the loss function. Indicates the angular velocity label value at the next moment. The predicted value of the angular velocity of the aircraft at the next moment. The mean square error between them It's a hyperparameter. This is used to reduce the complexity of prediction networks for unsteady aerodynamic models and improve their generalization ability. This represents the parameter vector of the unsteady aerodynamic model prediction network. express The i-th unsteady aerodynamic model predictor network parameter in the diagram, where n represents the parameter vector. The number of parameters in the middle, Indicates the cumulative symbol, angular velocity label value. Used to indicate the aircraft's true angular velocity at the next moment; Step 32: Obtain the network gradient of the loss function based on the loss function: in, The network gradient represents the loss function. Represents the parameter vector Find the gradient; Step 33: During each iteration of training the attitude dynamics network, based on the calculated network gradient, the parameters of the unsteady aerodynamic model prediction network are adjusted using the gradient descent algorithm to minimize the loss function. The updated unsteady aerodynamic model prediction network parameters are then used to replace the parameters of the unsteady aerodynamic model prediction network from the previous iteration. When the preset number of iterations is reached, the optimized unsteady aerodynamic model prediction network parameters and the attitude dynamics network model are output.
2. The method for identifying unsteady aerodynamic models according to claim 1, characterized in that, Step 21 includes: Step 211: Construct an unsteady aerodynamic model; in, This represents the longitudinal pitch unsteady aerodynamic moment. This represents the angle of attack of the aircraft at the current time t. This represents the angle of attack of the aircraft at the previous P time points. Indicates the aircraft's angle of attack sequence. This represents the mapping relationship from the angle of attack sequence to the pitch unsteady aerodynamic moment; Step 212: Set the aircraft's angle of attack sequence Input an unsteady aerodynamic model prediction network, and output the longitudinal pitch unsteady aerodynamic moment. Predicted value ; Step 213: Calculate the predicted values of the longitudinal pitch unsteady aerodynamic moments. With steady aerodynamic torque coefficient Add them together to get the sum of the torques.
3. The method for identifying unsteady aerodynamic models according to claim 2, characterized in that, Step 22 includes: The predicted value of the aircraft's angular acceleration is obtained using the following formula. : in, This represents the predicted angular acceleration of the aircraft. Represents the dynamic pressure matrix. Q Indicates dynamic pressure. s Indicates the reference area. b Indicates the lateral reference length. c Indicates the longitudinal reference length. Represents a diagonal matrix. Represents the steady aerodynamic torque coefficient. , This represents the predicted value of the unsteady aerodynamic moment coefficient. This represents the predicted value of the longitudinal pitch unsteady aerodynamic moment coefficient. I This represents the moment of inertia matrix of the aircraft. This indicates the current angular velocity of the aircraft.
4. The method for identifying unsteady aerodynamic models according to claim 3, characterized in that, Step 23 includes: The predicted angular velocity of the aircraft at the next moment is obtained by integrating the predicted angular acceleration value of the aircraft using the following formula: in, This represents the predicted angular velocity of the aircraft at the next moment. Indicates the integral symbol, This represents the predicted angular acceleration of the aircraft. Indicates time.
5. The method for identifying unsteady aerodynamic models according to any one of claims 1-4, characterized in that, The unsteady aerodynamic model prediction network includes an input layer, a hidden layer, and an output layer connected in sequence. The input layer is used to input the angle of attack sequence of the aircraft, the hidden layer is used to perform a nonlinear transformation on the angle of attack sequence of the aircraft to obtain the predicted value of the longitudinal pitch unsteady aerodynamic moment, and the output layer is used to output the predicted value of the longitudinal pitch unsteady aerodynamic moment.
6. A system for identifying unsteady aerodynamic models, implementing the method for identifying unsteady aerodynamic models according to any one of claims 1-5, characterized in that, include: The data acquisition module is used to acquire the training dataset; Each training sample in the training dataset includes the aircraft's angular velocity, steady aerodynamic torque coefficient, dynamic pressure matrix, and angular velocity label value; The network construction module is used to input the aircraft's angle of attack sequence into the attitude dynamics network to obtain the predicted value of the aircraft's angular velocity at the next moment; the attitude dynamics network includes an unsteady aerodynamic model prediction network, an attitude dynamics model, and an integrator connected in sequence. The function construction module is used to construct the loss function of the attitude dynamics network based on the next moment's angular velocity label value, the predicted value of the aircraft's next moment's angular velocity, and the unsteady aerodynamic model prediction network parameters. The network training module is used to input the training dataset into the attitude dynamics network for training, with the goal of minimizing the loss function, and output the optimized unsteady aerodynamic model prediction network parameters and attitude dynamics network model; the attitude dynamics network model is used to identify unsteady aerodynamic models.
7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executed by the processor, characterized in that, When the processor executes the computer program, it implements the method for identifying unsteady aerodynamic models as described in any one of claims 1 to 5.
8. A non-transitory computer-readable storage medium, characterized in that, The non-transitory computer-readable storage medium stores computer instructions for causing the computer to execute the identification method for an unsteady aerodynamic model as described in any one of claims 1 to 5.