Aero-engine dynamic real-time modeling method based on ODENet

Abstract

The present application relates to the technical field of aero-engine control, and particularly relates to an aero-engine dynamic real-time modeling method based on ODE Net, comprising: obtaining input data and output data of an engine component-level model, constructing an initial ODE Net model, and training the initial ODE Net model to obtain an ODE Net model. The method of the present application avoids the shortcomings of insufficient precision based on mechanism modeling and insufficient generalization ability and lack of mechanism support of a neural network algorithm based on data, and compared with the method of the prior art, the method of the present application has higher precision and stronger real-time performance.

Description

Technical Field

[0001] This invention relates to the field of aero-engine control technology, and more specifically to a real-time dynamic modeling method for aero-engines based on ODENet. Background Technology

[0002] In the Full Authority Digital Electronic Control (FADEC) system for real-time modeling and control of aero-engines, the engine dynamic model is the foundation for performance optimization control, health management, and fault diagnosis. The real-time performance and accuracy of the model directly affect the effectiveness of engine control and fault diagnosis.

[0003] Currently, aero-engine modeling methods are mainly divided into three categories: mechanism modeling, data-driven modeling, and hybrid modeling. Mechanism modeling has good interpretability and high accuracy; however, as engine structures become increasingly complex, determining accurate mathematical expressions becomes more difficult, and mechanism-based models may face accuracy degradation. In contrast, neural network-based modeling methods do not need to consider the complex physical structure inside the engine system, simplifying the model construction process and effectively compensating for the shortcomings of mechanism modeling. However, pure neural network methods may not be reliable when facing scenarios with high requirements for model generalization ability and result interpretability. Therefore, hybrid modeling methods based on mechanism and data have emerged, combining the interpretability of mechanism modeling with the flexibility of data-driven modeling, thus improving the model's generalization ability. State-space models are crucial for multi-data performance analysis and controller design of aero-engines; however, in the field of aero-engine modeling, traditional state-space models rely on the linearization of component-level models, resulting in poor model accuracy and real-time performance.

[0004] Therefore, it is necessary to apply data-driven methods to state-space modeling, and propose a hybrid modeling method with state-space model as prior knowledge. Summary of the Invention

[0005] This invention provides a dynamic real-time modeling method for aero-engines based on ODENet. This method is a hybrid modeling method that uses a state-space model as prior knowledge, in order to solve the problem of insufficient accuracy and real-time performance caused by the reliance on linearization of component-level models in existing state-space modeling.

[0006] The present invention provides a real-time dynamic modeling method for aero-engines based on ODENet, which employs the following technical solution:

[0007] Multiple pre-set APRBS excitation signals are imported into the engine component-level model to obtain the output data of the engine component-level model. The APRBS excitation signals are the input data, which include: main fuel flow rate, nozzle throat area opening, and rear bypass ejector opening. The output data are: high-pressure rotor speed, low-pressure rotor speed, engine thrust, low-pressure rotor surge margin, high-pressure rotor surge margin, and turbine inlet temperature. The high-pressure rotor speed and low-pressure rotor speed in the output data are used as the state data of the engine component-level model.

[0008] An initial ODENet model is constructed based on a nonlinear continuous state-space model. The initial ODENet model includes a state network model and an output network model. The state network model includes a state network and an ordinary differential equation solver; the output network model includes an output network.

[0009] The input data and its corresponding state data are used as the input parameters of the state network, and the derivatives of the state data are used as the output parameters of the state network. The derivatives of the state data predicted by the state network are converted into the predicted values ​​of the state data by the ordinary differential equation solver. The loss function is set according to the state data and the predicted values ​​of the state data, and the state network is trained to obtain the trained state network model.

[0010] The input data and its corresponding state data are used as the input parameters of the output network, and the output data corresponding to the APRBS excitation signal are used as the output parameters of the output network. The output network is trained to obtain the trained output network model.

[0011] The ODENet model is constructed based on the trained state network model and the trained output network model.

[0012] Preferably, the expression for the nonlinear continuous state-space model is:

[0013]

[0014] In the formula, Represents the derivative vector of state data; Represents a state network model; Indicates the output network model; This represents the output vector composed of the output data; This represents the input vector composed of the input data; This represents the state vector composed of state data.

[0015] Preferably, the expression for the ODENet model is:

[0016]

[0017] In the formula, Represents the derivative vector of state data; This represents the output vector composed of the output data; Represents a state network model; Indicates the output network model; The state vector is represented at the th... The hidden state at each moment; Indicates the input vector at the th The hidden state at each moment; Hyperparameters of a state network This represents the hyperparameters of the output network.

[0018] Preferably, the step of pre-setting multiple sets of APRBS excitation signals is as follows: generating multiple sets of APRBS excitation signals using amplitude-modulated pseudo-random binary sequences.

[0019] Preferably, when generating multiple sets of APRBS excitation signals using amplitude-modulated pseudo-random binary sequences: the variation range of the main fuel flow rate is [30%, 100%] of the normalized value of the main fuel flow rate in the multiple sets of APRBS excitation signals; the variation range of the nozzle throat area opening is [100%, 140%] of the normalized value of the nozzle throat area opening in the multiple sets of APRBS excitation signals; and the variation range of the rear duct ejector opening is [60%, 140%] of the normalized value of the rear duct ejector opening in the multiple sets of APRBS excitation signals.

[0020] Preferably, the opening of the rear duct ejector is obtained based on the inlet area of ​​the mixing chamber, the inlet area of ​​the mixing chamber, and the outlet area of ​​the mixing chamber of the engine model.

[0021] Preferably, the formula for calculating the opening of the rear duct ejector is:

[0022]

[0023] In the formula, A 63 The area of ​​the inlet of the mixing chamber; A 163 The area of ​​the mixed-use outdoor culvert entrance; A 64 The area of ​​the mixing chamber outlet; α 63 This is a proportional coefficient representing the inlet area of ​​the mixing chamber. α 163 This indicates the opening degree of the rear duct ejector.

[0024] Preferably, the ratio of the cross-sectional area of ​​the nozzle throat to the minimum cross-sectional area of ​​the nozzle throat is used as the nozzle throat area opening.

[0025] The beneficial effects of this invention are:

[0026] The proposed real-time dynamic modeling method for aero-engines based on ODENet imports the designed APRBS excitation signals into the engine component-level model to obtain the corresponding input and output data of the engine model under each set of APRBS excitation signals. Then, based on the input data, state vector, output data, and state derivative vector, the initial ODENet model constructed based on the nonlinear continuous state space model is trained to obtain the ODENet-based model. The ODENet model not only avoids completely "end-to-end" black-box fitting, but also describes the continuous changes of the hidden state through ordinary differential equations, thereby more accurately capturing the dynamic characteristics of the data. In other words, the modeling method of this invention, as a hybrid modeling method based on mechanism and data, avoids the shortcomings of insufficient accuracy of mechanism-based modeling and insufficient generalization ability of data-based neural network methods due to lack of mechanism support. Compared with the original methods, it has higher accuracy and stronger real-time performance. Attached Figure Description

[0027] 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 some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0028] Figure 1 This is a schematic diagram of the overall structure of an embodiment of the real-time dynamic modeling method for aero-engines based on ODENet according to the present invention.

[0029] Figure 2 This diagram illustrates the connection between the CLM-based dynamic real-time model and the ODENet-based dynamic real-time model in this embodiment.

[0030] Figure 3 This is a schematic diagram of the three-dimensional model structure of the variable cycle engine in this embodiment;

[0031] Figure 4 This is a block diagram of the ODENet model training in this embodiment;

[0032] Figure 5 This is a schematic diagram of the APRBS excitation signal generated using an amplitude-modulated pseudo-random binary sequence in this embodiment.

[0033] Figure 6 This is a schematic diagram showing how the MAE of the state network and output network changes with the number of neurons and hidden layers in this embodiment;

[0034] Figure 7 This is a schematic diagram comparing the ODENet model output with the training data in this embodiment;

[0035] Figure 8 This is a schematic diagram comparing the ODENet model output with the test data in this embodiment. Detailed Implementation

[0036] 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.

[0037] An embodiment of the present invention, based on ODENet, for real-time dynamic modeling of aero-engines, is as follows: Figure 1 As shown, it includes:

[0038] S1. Obtain the input and output data of the engine component-level model;

[0039] Specifically, an engine component-level model is constructed, and multiple pre-set APRBS excitation signals are imported into the engine component-level model to obtain the output data of the engine component-level model. Among them, the APRBS excitation signals are the input data, which include: main fuel flow rate, tail nozzle throat area opening, and rear bypass ejector opening; the output data are: high-pressure rotor speed, low-pressure rotor speed, engine thrust, low-pressure rotor surge margin, high-pressure rotor surge margin, and turbine inlet temperature; the high-pressure rotor speed and low-pressure rotor speed in the output data are used as the state data of the engine component-level model.

[0040] In this embodiment, the steps for generating multiple sets of APRBS excitation signals are as follows: Since the preset excitation signals must be able to fully excite the engine characteristics, and Amplitude-modulated pseudo-random binary sequence (APRBS) is a suitable excitation signal for this situation, capable of exciting the aero-engine system over a wide frequency range, this embodiment utilizes APRBS to generate multiple sets of APRBS excitation signals. It should be noted that, in order to effectively capture the control effect of each set of APRBS excitation signals, during the generation of APRBS excitation signals, it is ensured that at most one input signal undergoes a step transition at any given time.

[0041] To fully cover the engine's operating conditions, in this embodiment, when generating multiple sets of APRBS excitation signals using amplitude-modulated pseudo-random binary sequences: the variation range of the main fuel flow rate is [30%, 100%] of the normalized value of the main fuel flow rate in the multiple sets of APRBS excitation signals; the variation range of the nozzle throat area opening is [100%, 140%] of the normalized value of the nozzle throat area opening in the multiple sets of APRBS excitation signals; and the variation range of the rear bypass ejector opening is [60%, 140%] of the normalized value of the rear bypass ejector opening in the multiple sets of APRBS excitation signals.

[0042] Multiple sets of APRBS excitation signals are imported into the engine component-level model to obtain the input and output data of the engine component-level model corresponding to each set of APRBS excitation signals. The input data includes: main fuel flow rate, nozzle throat area opening, and rear bypass ejector opening. The output data includes: high-pressure rotor speed, low-pressure rotor speed, engine thrust, low-pressure rotor surge margin, high-pressure rotor surge margin, and turbine inlet temperature. The high-pressure rotor speed and low-pressure rotor speed in the output data are used as the state data of the engine component-level model.

[0043] In this embodiment, the state vector ,in, This refers to the high-voltage rotor speed; The low-pressure rotor speed; the derivative of the state vector. , This refers to the acceleration of the high-pressure rotor. This refers to the acceleration of the low-pressure rotor.

[0044] S3. Construct the initial ODENet model and train the initial ODENet model to obtain the ODENet model;

[0045] Step 31: The steps to construct the initial ODENet model based on the nonlinear continuous state-space model:

[0046] In this embodiment, a variable cycle engine is taken as the research object, and its structural schematic diagram is as follows. Figure 3 As shown; the section numbers in the figure. A 63 The interior entrance area of ​​the mixing chamber A 163 For the mixed-use outdoor culvert inlet area, A 64 The area of ​​the mixing chamber outlet; the opening of the rear duct ejector can be obtained from the areas of the three. α 163 The calculation formula is:

[0047]

[0048] The expression for the nozzle throat area opening is:

[0049]

[0050] In the formula, A 63 The area of ​​the inlet of the mixing chamber; A 163 The area of ​​the mixed-use outdoor culvert entrance; A 64 The area of ​​the mixing chamber outlet; α 63 This is a proportional coefficient representing the inlet area of ​​the mixing chamber. α 163 Indicates the opening degree of the rear duct ejector; This refers to the opening area of ​​the tail nozzle throat. This is the cross-section of the tail nozzle throat; This is the minimum cross-sectional area of ​​the tailpipe throat.

[0051] Based on prior knowledge of aero-engines, the nonlinear continuous state-space model of a variable cycle engine can be described as follows:

[0052]

[0053] In the formula, Represents the derivative vector of state data. ,in, For high-voltage rotor acceleration and This refers to the acceleration of the low-pressure rotor. Represents a state network model; Indicates the output network model; The output vector represents the output data. Among them, the high-voltage rotor speed Low-pressure rotor speed Engine thrust Low-voltage rotor surge margin High-voltage rotor surge margin and turbine inlet temperature ; The input vector represents the input data. ,in, Main fuel flow; This refers to the opening area of ​​the tail nozzle throat. For the opening of the rear duct ejector; A state vector is a vector composed of state data. ,in, For high-voltage rotor speed and This is the low-pressure rotor speed.

[0054] Specifically, an initial ODENet model is constructed based on the aforementioned nonlinear continuous state-space model. The initial ODENet model includes a state network model and an output network model. The state network model includes a state network and an ordinary differential equation solver; the output network model includes an output network. Its core concept is to use the state network to represent the derivative vector of the state data. The output vector is represented by an output network. Therefore, the dynamic real-time model based on ODENet is as follows:

[0055]

[0056] In the formula, Represents the derivative vector of state data; This represents the output vector composed of the output data; Represents a state network model; Indicates the output network model; The state vector is represented at the th... The hidden state at each moment; Indicates the input vector at the th The hidden state at each moment; Hyperparameters of a state network This represents the hyperparameters of the output network. It should be noted that the APRBS excitation signal vector... Wherein, the APRBS excitation signals are the main fuel flow rate and the main fuel flow rate. Tail nozzle throat area opening and the opening of the duct ejector Output data (i.e., the output vector of the ODENet model) The data are respectively the high-voltage rotor speed. Low-pressure rotor speed Engine thrust Low-voltage rotor surge margin High-voltage rotor surge margin and turbine inlet temperature .

[0057] Step 32: Training the initial ODENet model: A fully connected neural network (MLP) is used to represent the state network and the output network respectively. For the state network, the input is the state vector. x The input vector formed by the APRBS excitation signal u The output of the state network is the derivative of the state vector. The predicted value of the state vector is obtained by solving the derivative of the state vector using an ordinary differential equation solver; for the output network, the input of the output network is still the state vector. x The input vector formed by the APRBS excitation signalu The output data of the output network is the output vector corresponding to the APRBS excitation signal. y Then, the network is trained. The specific training process is as follows: the input data and its corresponding state data are used as the input parameters of the state network, and the derivatives of the state data are used as the output parameters of the state network. The derivatives of the state data predicted by the state network are converted into predicted values ​​of the state data by the ordinary differential equation solver. A loss function is set according to the state data and the predicted values ​​of the state data, and the state network is trained to obtain a trained state network model. The input data and its corresponding state data are used as the input parameters of the output network, and the output data corresponding to the APRBS excitation signal are used as the output parameters of the output network. The output network is trained to obtain a trained output network model. The trained state network model and the trained output network model constitute the ODENet model. The high-pressure rotor speed of the engine can be determined based on the ODENet model. Low-pressure rotor speed Engine thrust Low-voltage rotor surge margin High-voltage rotor surge margin and turbine inlet temperature Make accurate predictions.

[0058] In this embodiment, both the state network and the output network employ fully connected neural networks (MLPs). Fully connected neural networks are a common neural network structure used in function regression; this method is simple, flexible, and has a strong ability to fit nonlinear functions. Therefore, this embodiment uses fully connected neural networks to represent the state network and output network of the ODENet model. For the state network, the input is the state vector. x The input vector formed by the APRBS excitation signal u The output of the state network is the derivative of the state vector. For the output network, the input to the output network is still the state vector. x The input vector formed by the APRBS excitation signal u The output data of the output network is the output vector corresponding to the APRBS excitation signal. y .

[0059] The expression for the state network is:

[0060]

[0061] In the formula, Represents the state network; x is the state vector and u is the input vector composed of APRBS excitation signals; The activation function of the state network is typically represented by tanh, sigmoid, or ReLU. This embodiment uses the sigmoid function. This represents the total number of layers in the state network; The first state network Layer weight variables; The first state network Layer bias variables.

[0062] The expression for the output network is:

[0063]

[0064] In the formula, This represents the output network, where x is the state vector and u is the input vector composed of APRBS excitation signals; This represents the activation function of the output network. Commonly used functions include tanh, sigmoid, and ReLU. This example uses the sigmoid function. Indicates the total number of layers in the output network; Indicates the output network's first... Layer weight variables; Indicates the output network's first... Layer bias variables.

[0065] The training process of ODENet is divided into forward propagation and backward propagation, such as Figure 4 As shown, during the forward propagation of the ODENet model, the input vector composed of each set of APRBS excitation signals and the corresponding state vector are used as the input parameters of the state network, and the derivative of the state vector is used as the output parameter of the state network. An ordinary differential equation solver then converts the derivative of the predicted state vector into the predicted state vector value. The state model works similarly to... (The sentence is incomplete and requires further context to translate accurately.) t i Integrating over the state network:

[0066]

[0067] In the formula, x ( t i+1 )express t i+1 The predicted value of the state vector at time step [time], therefore, given the initial state vector... x ( t 0) and neural network parameterization functions f The state vector at any given time can be derived using an ordinary differential equation solver (ODESolver). For the time series of predicted state vector values, the hidden states at all intermediate time points need to be obtained, i.e. .

[0068]

[0069] In backpropagation, a loss function is needed to calculate the predicted sequence. and truth sequence x The error between them is analyzed, and the optimizer is used to continuously optimize and update the parameters. .

[0070] The output model is trained without the participation of the ODESLver. In forward propagation, the state vector corresponding to each set of APRBS excitation signal vectors is used as the input data to the output network, and the output data corresponding to that set of APRBS excitation signals is used as the output data of the output network. In backpropagation, the predicted sequence is calculated using a loss function. and truth sequence y The error between them is used to update the output network parameters. The ODENet model is derived based on an ordinary differential equation solver, a trained state network, and a trained output network.

[0071] When using ODESolver for integration calculations, this paper selects the fourth-order Runge-Kutta algorithm to solve for the predicted value of the state vector, given its high accuracy and stability. The loss function is defined using the root mean square error (MAE), and the state vectors are calculated accordingly. x and output vector y The corresponding residuals are used to update the data in the state network and the output network. The loss function is shown below:

[0072]

[0073] In the formula, n The number of training data points is represented by `minibatch`, which represents the data samples used in each iteration of model training. Indicates the loss value; This represents the predicted value of the state vector obtained by the ordinary differential equation solver; the main goal of training the ODENet model is to adjust the hyperparameters of the state network and the output network by minimizing the loss function, thereby ensuring that the network can make accurate predictions.

[0074] The following description, in conjunction with the accompanying drawings, illustrates this embodiment:

[0075] like Figure 5As shown, in order to effectively capture the control effect of each input signal, when generating the amplitude-modulated pseudo-random binary sequence, it is ensured that at most one input data undergoes a step transition at each time step, and the opening of the back duct ejector is also guaranteed. α 163 The variation amplitude does not exceed 30%, thus obtaining the distribution of input spatial data points; the sampling time of the engine component-level model (CLM model) is 0.02s, therefore the minimum hold time (i.e., the shortest time for the signal to remain unchanged) is set to 15s to ensure that the engine component-level model is fully excited, thereby obtaining the APRBS excitation signal sequence as follows. Figure 5 As shown in (a), (b), and (c) in the figure, Figure 5 (d) in the image is a magnified view of the gray areas of (a), (b), and (c) within a single figure, allowing for visualization as follows: Figure 5 As seen in (d), the data transition time is asynchronous.

[0076] The APRBS excitation signal sequence was imported into the engine component-level model (CLM model) to obtain a dataset with three inputs and six outputs. Based on experience, the dataset was split, with the first 80% used as the training set and the remaining 20% ​​as the test set.

[0077] The main goal of training the initial ODENet model is to adjust the hyperparameters of the state network and the output network by minimizing the loss function, thereby ensuring that the neural network can make accurate predictions. To obtain the optimal hyperparameters, the impact of the number of hidden layers and neurons on the performance of the state and output networks is analyzed from the perspective of fitting accuracy. For simplicity, it is assumed that the state and output networks have the same number of hidden layers, and all hidden layers have the same number of neurons. During the pre-training of the initial ODENet model, it was found that the root mean square error (MAE) of the initial ODENet model stabilized at 90 steps. To prevent overtraining, the Epoch was set to 90. The results of the root mean square error (MAE) changing with the number of hidden layers and neurons are as follows: Figure 6 As shown.

[0078] The obtained state network and output network of the ODENet model, along with their corresponding training hyperparameters, are shown in Table 1. For the training data in this embodiment, this set of hyperparameters can balance the generalization ability and training efficiency of the neural network. Thus, the dynamic modeling of the variable cycle engine is completed.

[0079] Table 1

[0080]

[0081] Validating the effectiveness of the ODENet model:

[0082] Open-loop simulation of the ODENet model: In the open-loop simulation, the fit accuracy of the ODENet model is evaluated using the fit degree (fit). The larger the fit degree (fit), the better the fit.

[0083]

[0084] In the formula, This represents the predicted value of the output vector; Indicates the output vector; This represents the mean of the output vector.

[0085] like Figures 7 to 8 As shown, in this embodiment, the simulation environment for verifying the effectiveness of the ODENet model is a Windows 11 Home Chinese version operating system, an AMD Ryzen 7 PRO 7840HS CPU with a main frequency of 3.8GHz, and 32G of memory. In the open-loop simulation, the predicted outputs of the training dataset and the ODENet model are first compared and analyzed. Figure 7 In the diagram, the blue line represents the training data, and the yellow dashed line represents the predicted output data of the ODENet model; from Figure 7 From this, we can know that Figure 7 Images (a) to (f) show the fit of the ODENet model output vectors, corresponding to 96.41%, 98.57%, 98.31%, 98.01%, 94.15%, and 96.79%. The ODENet model's accuracy was then validated using the test dataset, and the results are as follows: Figure 8 As shown, Figure 8 (a) to (f) show that the fit of the ODENet model output vector is 96.77%, 98.69%, 98.52%, 97.66%, 94.05%, and 97.04%, respectively. This shows that the ODENet model can track the output of the CLM in real time, and verifies that the engine dynamic model based on ODENet has high accuracy.

[0086] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An aero-engine dynamic real-time modeling method based on ODENet, characterized in that, The method comprises the following steps: The preset multiple groups of APRBS excitation signals are introduced into the engine component level model to obtain output data of the engine component level model, wherein the APRBS excitation signal is input data, which comprises main fuel flow, tail nozzle throat area opening degree and afterburner opening degree; the output data comprises high-pressure rotor speed, low-pressure rotor speed, engine thrust, low-pressure rotor surge margin, high-pressure rotor surge margin and turbine inlet temperature; the high-pressure rotor speed and the low-pressure rotor speed in the output data are used as state data of the engine component level model; An initial ODENet model is constructed based on a nonlinear continuous state space model, the initial ODENet model comprising a state network model and an output network model, the state network model comprising a state network and a common differential equation solver; the output network model comprising an output network; The input data and corresponding state data are used as input parameters of the state network, and the derivative of the state data is used as output parameters of the state network; the derivative of the state data predicted by the state network is converted into a predicted value of the state data by the common differential equation solver; a loss function is set according to the state data and the predicted value of the state data, and the state network is trained to obtain a trained state network model; The input data and corresponding state data are used as input parameters of the output network, and the output data corresponding to the group of APRBS excitation signals is used as output parameters of the output network; the output network is trained to obtain a trained output network model; The trained state network model and the trained output network model are used to construct an ODENet model.

2. The method of claim 1, wherein, The expression of the nonlinear continuous state space model is: wherein denotes a state data derivative vector; denotes a state network model; denotes an output network model; denotes an output vector composed of output data; denotes an input vector composed of input data; denotes a state vector composed of state data.

3. The method of claim 1, wherein, The expression of the ODENet model is: wherein denotes a state data derivative vector; denotes an output vector composed of output data; denotes a state network model; denotes an output network model; denotes a hidden state of the state vector at the time step; denotes a hidden state of the input vector at the time step; denotes a hyperparameter of the state network; denotes a hyperparameter of the output network.

4. The method of claim 1, wherein, The preset multiple groups of APRBS excitation signals are generated by using amplitude modulation pseudo-random binary sequences.

5. The method of claim 4, wherein, When the multiple groups of APRBS excitation signals are generated by using amplitude modulation pseudo-random binary sequences: the variation range of the main fuel flow is [30%, 100%] of the normalized value of the main fuel flow in the multiple groups of APRBS excitation signals; the variation range of the tail nozzle throat area opening degree is [100%, 140%] of the normalized value of the tail nozzle throat area opening degree in the multiple groups of APRBS excitation signals; and the variation range of the afterburner opening degree is [60%, 140%] of the normalized value of the afterburner opening degree in the multiple groups of APRBS excitation signals.

6. The method of claim 1, wherein, The afterburner opening degree is obtained according to the mixed chamber inner duct inlet area, the mixed chamber outer duct inlet area and the mixed chamber outlet area of the engine model.

7. The method of claim 1, wherein, The calculation formula of the afterburner opening degree is: wherein A 63 is the area of the inner containment inlet of the mixing chamber; A 163 is the area of the outer containment inlet of the mixing chamber; A 64 is the area of the outlet of the mixing chamber; α 63 is a proportionality coefficient of the area of the inner containment inlet of the mixing chamber, α 163 denotes the bypass ratio of the afterburner.

8. The method of claim 1, wherein, The ratio of the tail nozzle throat section to the minimum cross-sectional area of the tail nozzle throat is used as the tail nozzle throat area opening degree.

Images