Aero-engine infield prediction method based on combination of POD dimensionality reduction and machine learning

Abstract

The invention provides an aero-engine infield prediction method based on combination of POD dimensionality reduction and machine learning. According to the method, dimension reduction is carried out on a three-dimensional numerical simulation result of the whole aero-engine through a POD method, main characteristic modes are extracted, and a time mode matrix and a space mode matrix are generated. And then training the time modal matrix by using a machine learning model, predicting the time modal matrix of the test working condition, and reconstructing a predicted field value matrix in combination with the space modal matrix. The method has remarkable effects in the aspects of reducing the data dimension and improving the calculation efficiency and the prediction precision, and is suitable for rapid prediction of the Mach number field, the total temperature field and the total pressure field of the whole aero-engine.

Description

Technical Field

[0001] The present invention relates to the technical field of aircraft engine performance prediction, and in particular to an aircraft engine internal field prediction method based on the combination of POD dimensionality reduction and machine learning. Background Art

[0002] Aircraft engine performance prediction is a crucial component of engine design, operation, and maintenance. Traditional prediction methods primarily rely on physical models and experimental data. While these methods can reflect engine operating conditions to a certain extent, they often exhibit limitations when dealing with complex operating conditions and high-dimensional data. For example, physical models require precise boundary conditions and parameter settings, while experimental data acquisition is costly and time-consuming. Furthermore, engine operating data is often high-dimensional, nonlinear, and nonstationary, further complicating the prediction challenges of traditional methods.

[0003] As a data-driven approach, machine learning can automatically extract features from massive amounts of data and achieve high-precision predictions through model training. In recent years, the application of machine learning in aero-engine performance prediction has gradually increased. However, machine learning methods still face challenges when processing high-dimensional data. High-dimensional data not only increases computational complexity but can also lead to model overfitting and reduced prediction performance. Furthermore, most research focuses on engine fault diagnosis, specifically dealing with one-dimensional results. In recent years, dimensionality reduction techniques have gained widespread attention as an effective means of solving high-dimensional data problems in three-dimensional numerical simulations. The POD method, a classic dimensionality reduction technique, can extract the main characteristic modes of the data by projecting high-dimensional data into low dimensions, thereby significantly reducing computational complexity and improving the prediction accuracy of the model. It has been widely used in fields such as fluid mechanics and structural dynamics. However, the POD method itself is a linear method, and its ability to describe the internal engine environment with strong nonlinear characteristics (such as turbulence and combustion environments) is limited.

[0004] Currently, predictive field technology, combining POD and machine learning techniques, is primarily used to analyze flow fields near wings or temperature fields within compressors and turbines. Research on predicting the multi-physics fields within an entire engine is lacking. Therefore, developing a method for rapidly predicting the multi-physics fields within an aeroengine is crucial for optimizing engine design, improving performance, and ensuring reliability. Summary of the Invention

[0005] The present invention aims to solve one of the technical problems in the related art at least to a certain extent.

[0006] The present invention proposes an aircraft engine internal field prediction method based on the combination of POD dimensionality reduction and machine learning, aiming to address the limitations of traditional methods in processing high-dimensional, nonlinear and non-stationary data when predicting the multi-physical fields inside the entire aircraft engine, while improving computational efficiency and prediction accuracy.

[0007] Another object of the present invention is to propose an aircraft engine interior field prediction system based on the combination of POD dimensionality reduction and machine learning.

[0008] To achieve the above objectives, the present invention proposes, on one hand, a method for predicting the interior field of an aircraft engine based on a combination of POD dimensionality reduction and machine learning, comprising:

[0009] Acquire high-dimensional data of the entire aircraft engine through three-dimensional numerical simulation; wherein the high-dimensional data includes Mach number field, total temperature field and total pressure field data;

[0010] The POD dimensionality reduction method based on the proper orthogonal decomposition is used to reduce the dimensionality of high-dimensional data to extract characteristic modes and generate spatial mode matrix and temporal mode matrix;

[0011] Inputting the time modal matrix and the operating parameters of the aircraft engine into the machine learning model for model training to obtain a trained machine learning model;

[0012] The trained machine learning model is used to predict the time modal matrix of the test condition, and the predicted field value matrix is ​​reconstructed based on the spatial modal matrix.

[0013] The aircraft engine interior field prediction method based on the combination of POD dimensionality reduction and machine learning in the embodiment of the present invention may also have the following additional technical features:

[0014] In one embodiment of the present invention, obtaining high-dimensional data of an entire aircraft engine through three-dimensional numerical simulation includes:

[0015] The KJ66 micro-turbojet engine was simulated in three dimensions using STARCCM+ fluid dynamics simulation software. The fluid domain meshes of varying precision were generated. The generated fluid domain meshes were then imported into ANSYS-FLUENT for solution. Preliminary numerical calculation results were generated through mesh independence verification.

[0016] Determining the engine speed, ambient pressure, and gas-fuel ratio based on the preliminary numerical calculation results, and selecting a plurality of operating points using a uniform stratified sampling method to perform numerical calculations on the operating points to obtain numerical calculation results for each operating point;

[0017] The numerical calculation results are imported into the Tecplot program to extract the required geometric parts and key parameter data to generate a .dat file that can be read by Tecplot.

[0018] In one embodiment of the present invention, a dimensionality reduction method based on proper orthogonal decomposition (POD) is used to reduce the dimensionality of high-dimensional data to extract characteristic modes, and to generate a spatial modal matrix and a time modal matrix after dimensionality reduction, including:

[0019] Perform POD decomposition on the parameter part of the .dat file generated by Tecplot, and perform POD decomposition on a high-dimensional information matrix U with a specific distribution in time and space. mn =[u(x i ,t j )], each column represents the signal at all locations in local time, and each row represents the signal at all times at the local location. Calculate the information mean field data:

[0020]

[0021] Calculate the information fluctuation matrix:

[0022]

[0023] Calculate the covariance matrix R:

[0024]

[0025] Perform eigenvalue decomposition and obtain the characteristic matrix D and eigenvector matrix of the covariance matrix through eigenvalue decomposition

[0026] R·Φ=Φ·D

[0027] The characteristic matrix D is a diagonal matrix, and the eigenvalues ​​λ on the diagonal are i is the corresponding eigenvector The eigenvalues ​​and eigenvector matrices of Eigenvector is the corresponding time mode;

[0028] Finally, the spatial mode Ψ is calculated, where each column is a spatial mode:

[0029]

[0030] In one embodiment of the present invention, the time modal matrix and the operating parameters of the aircraft engine are input into a machine learning model for model training to obtain a trained machine learning model, including:

[0031] Acquiring training data, preprocessing the training data, and dividing the preprocessed data into a training set, a test set, and a validation set; the training data includes operating parameters and a time modal matrix of the aircraft engine;

[0032] Construct a multivariate support vector machine (SVR) model, specifying that each column of the time modal matrix corresponds to a single SVR regressor;

[0033] During training, the input operating parameters of the aircraft engine are used as input features of each SVR regressor, and the column values ​​of the corresponding time modal matrix are used as target values; an SVR regressor is trained for each column of the time modal matrix based on the training set, and a multivariate SVR model including multiple SVR regressors is finally obtained;

[0034] The trained multivariate SVR model is tested and verified based on the test set and validation set to obtain a trained multivariate SVR model.

[0035] In one embodiment of the present invention, a trained machine learning model is used to predict the time modal matrix of the test condition, and a predicted field value matrix is ​​reconstructed based on the spatial modal matrix, including:

[0036] Perform a dot product operation on the spatial modal matrix and the predicted test condition time modal matrix, and add the result to the mean field data stored during POD dimensionality reduction to obtain the predicted field data;

[0037] The relative root mean square error and fitting coefficient are used to evaluate the difference and fitting degree between the predicted field data and the actual field data to obtain the data evaluation results;

[0038] Based on the data evaluation results, the predicted field data replaces the parameter part in the original Tecplot .dat file to generate a visual prediction result.

[0039] To achieve the above objectives, the present invention further proposes an aircraft engine interior field prediction system based on a combination of POD dimensionality reduction and machine learning, comprising:

[0040] A high-dimensional data acquisition module is used to obtain high-dimensional data of the entire aircraft engine through three-dimensional numerical simulation; wherein the high-dimensional data includes Mach number field, total temperature field and total pressure field data;

[0041] The intrinsic orthogonal decomposition module is used to reduce the dimensionality of high-dimensional data using the intrinsic orthogonal decomposition (POD) dimensionality reduction method to extract characteristic modes and generate spatial mode matrix and temporal mode matrix;

[0042] A machine learning model training module is used to input the time modal matrix and the operating parameters of the aircraft engine into the machine learning model for model training to obtain a trained machine learning model;

[0043] The reconstruction prediction module is used to use the trained machine learning model to predict the time modal matrix of the test condition and reconstruct the predicted field value matrix based on the spatial modal matrix.

[0044] The aircraft engine internal field prediction method and system based on the combination of POD dimensionality reduction and machine learning in the embodiment of the present invention adjusts the retention order of the time matrix obtained by POD dimensionality reduction to minimize the data volume while ensuring prediction accuracy, thereby further improving the learning efficiency of machine learning prediction.

[0045] Additional aspects and advantages of the present invention will be set forth in part in the description which follows and, in part, will be obvious from the description which follows, or may be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:

[0047] Figure 1 Flowchart of an aircraft engine interior field prediction method based on the combination of POD dimensionality reduction and machine learning according to an embodiment of the present invention;

[0048] Figure 2 1 is an architecture diagram of an aircraft engine interior field prediction method based on a combination of POD dimensionality reduction and machine learning according to an embodiment of the present invention;

[0049] Figure 3 is a schematic diagram of n-th order modal energy according to an embodiment of the present invention;

[0050] Figure 4 is a schematic diagram of determination coefficients of different machine learning methods according to an embodiment of the present invention;

[0051] Figure 5 2 is a structural diagram of an aircraft engine interior field prediction system based on the combination of POD dimensionality reduction and machine learning according to an embodiment of the present invention. DETAILED DESCRIPTION

[0052] It should be noted that, in the absence of conflict, the embodiments and features of the embodiments of the present invention can be combined with each other. The present invention will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0053] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of the present invention.

[0054] The following describes, with reference to the accompanying drawings, an aircraft engine interior field prediction method and system based on the combination of POD dimensionality reduction and machine learning according to an embodiment of the present invention.

[0055] Figure 1 FIG. 1 is a flow chart of an aircraft engine interior prediction method based on a combination of POD dimensionality reduction and machine learning according to an embodiment of the present invention. Figure 1 Shown and Figure 2 As shown, the method includes:

[0056] S1, obtaining high-dimensional data of the entire aircraft engine through three-dimensional numerical simulation; wherein the high-dimensional data includes Mach number field, total temperature field and total pressure field data.

[0057] In this embodiment, a three-dimensional numerical simulation of a KJ66 micro-turbojet engine was performed using STARCCM+ fluid dynamics simulation software. Meshes of varying degrees of accuracy were generated for the fluid domain, and the simulation was solved using ANSYS-FLUENT. After verifying mesh independence, numerical calculations were performed at specific operating points scientifically sampled within a given parameter space to generate a training dataset.

[0058] The parameters used in this embodiment are engine speed, ambient pressure and gas-fuel ratio. Uniform stratified sampling is used to obtain a total of 1000 operating points for calculation and results are obtained. The calculation results are used to extract the required geometric parts and parameter data of interest using the Tecplot program to generate a .dat file that can be read by Tecplot.

[0059] S2, uses the POD dimensionality reduction method based on the intrinsic orthogonal decomposition to reduce the dimensionality of high-dimensional data to extract characteristic modes and generate spatial modal matrix and temporal modal matrix.

[0060] Specifically, the POD method of the present invention includes: establishing an information matrix and removing the information mean field data to obtain an information fluctuation matrix; calculating the covariance matrix and performing eigenvalue decomposition to obtain a characteristic matrix and an eigenvector matrix; calculating the spatial mode and using the time matrix of the first i-order modes to train the neural network.

[0061] Specifically, the POD method is used to reduce the dimensionality of high-dimensional data, and each working condition is regarded as a "time sequence point". The main characteristic modes are extracted to generate the time modal matrix and the space modal matrix.

[0062] Specifically, the parameter part of the .dat file generated by Tecplot is decomposed into POD, the main characteristic modes are extracted, and the time mode matrix and the space mode matrix are generated. mn =[u(x i ,t j)], each column represents the signal at all locations in local time, and each row represents the signal at all times at the local location. Calculate the information mean data:

[0063]

[0064] Calculate the information fluctuation matrix:

[0065]

[0066] Then calculate the covariance matrix R

[0067]

[0068] Then perform eigenvalue decomposition to obtain the characteristic matrix D and eigenvector matrix of the covariance matrix

[0069] R·Φ=Φ·D

[0070] The characteristic matrix D is a diagonal matrix, and the eigenvalues ​​λ on the diagonal are i is the corresponding eigenvector The eigenvalues ​​and eigenvector matrices of Eigenvector is the corresponding time mode.

[0071] Finally, the spatial modes Ψ are calculated, where each column is a spatial mode.

[0072]

[0073] like Figure 3 As shown, the retained time matrix order is selected according to the decrease in the energy of the n-th order mode. In this embodiment, the retained time matrix order is selected to be 20-100.

[0074] S3, inputting the time modal matrix and the operating parameters of the aircraft engine into the machine learning model for model training to obtain a trained machine learning model.

[0075] Specifically, the engine operating parameters (ambient pressure, gas-fuel ratio, speed, etc.) are used as input, and the time modal matrix after dimensionality reduction is used as output to train the machine learning model. Figure 4 We used various machine learning models to train and test the reduced temporal modal matrix. In this example, we selected a multivariate support vector machine (SVM) model as the optimal model. This model applied a single SVR output regressor to each column of the temporal matrix. In the SVR model, ε was set to the default value of 0.1, the regularization parameter was 1.0, and the RBF kernel function was selected. After training, the model was stored.

[0076] The embodiment of the present invention obtains training data, preprocesses the training data, and divides the preprocessed data into a training set, a test set, and a validation set; the training data includes operating parameters and a time modal matrix of an aircraft engine;

[0077] Construct a multivariate support vector machine (SVR) model, specifying that each column of the time modal matrix corresponds to a single SVR regressor;

[0078] During training, the input operating parameters of the aircraft engine are used as input features of each SVR regressor, and the column values ​​of the corresponding time modal matrix are used as target values; an SVR regressor is trained for each column of the time modal matrix based on the training set, and a multivariate SVR model including multiple SVR regressors is finally obtained;

[0079] The trained multivariate SVR model is tested and verified based on the test set and validation set to obtain a trained multivariate SVR model.

[0080] S4, using the trained machine learning model to predict the time modal matrix of the test condition, and reconstructing the predicted field value matrix based on the spatial modal matrix.

[0081] Specifically, the trained model is used to predict the time modal matrix of the test condition, and the predicted field value matrix is ​​reconstructed in combination with the spatial modal matrix. The present invention utilizes the trained model to multiply the spatial mode and the predicted time matrix, and adds the mean field data stored during POD dimensionality reduction to obtain the predicted field data. When verifying the prediction accuracy, the data evaluation results are obtained by judging using relative root mean square error and fitting coefficient, and when intuitively comparing the error between the predicted field and the original field, the predicted data can be substituted for the parameter part in the original Tecplot .dat file to obtain a visual result. In this example, training, prediction and reconstruction are carried out for the Mach number field, total temperature field and total pressure field, and the predicted field value matrix includes the Mach number field, total temperature field and total pressure field.

[0082] The beneficial effects of the present invention are:

[0083] 1) The present invention uses the POD method to reduce the dimensionality of the three-dimensional numerical simulation results, reducing the spatial order of the number of grid points in the original calculation conditions to the "time" order of the number of conditions. By retaining the spatial modal matrix with the same size as the data of a single condition, the subsequent machine learning model is significantly reduced in its dependence on the amount of high-fidelity simulation or experimental data, thereby reducing the data acquisition cost and computing resource consumption.

[0084] 2) The present invention performs multi-physics field training and prediction on the entire aircraft engine, and is capable of processing high-dimensional, nonlinear, and non-stationary data under complex internal operating conditions of the aircraft engine. It is suitable for performance prediction of the entire engine and its components and has wide applicability.

[0085] 3) The present invention only takes about 24.3 seconds to predict 200 cases in the test set, while the traditional numerical simulation method takes about 2 hours to converge the calculation of a single case. This method significantly improves the computational efficiency and is suitable for rapid prediction and real-time applications.

[0086] The aircraft engine internal field prediction method based on the combination of POD dimensionality reduction and machine learning according to an embodiment of the present invention is intended to address the limitations of traditional methods in processing high-dimensional, nonlinear and non-stationary data when predicting the multi-physical fields inside the entire aircraft engine, while improving computational efficiency and prediction accuracy. The present invention adjusts the order of the time matrix obtained by POD dimensionality reduction to minimize the amount of data while ensuring prediction accuracy, thereby further improving the learning efficiency of machine learning prediction. The present invention is suitable for performance prediction of the KJ66 micro turbojet engine, with computational efficiency significantly higher than that of traditional numerical simulation methods, and prediction accuracy that meets engineering application requirements. It can also be applied to temperature and pressure prediction at the aircraft engine component level.

[0087] In order to implement the above embodiment, Figure 5 As shown, this embodiment also provides an aircraft engine interior field prediction system 10 based on the combination of POD dimensionality reduction and machine learning, including:

[0088] The high-dimensional data acquisition module 100 is used to obtain high-dimensional data of the entire aircraft engine through three-dimensional numerical simulation; wherein the high-dimensional data includes Mach number field, total temperature field and total pressure field data;

[0089] The intrinsic orthogonal decomposition module 200 is used to perform dimensionality reduction operations on high-dimensional data using the intrinsic orthogonal decomposition (POD) dimensionality reduction method to extract characteristic modes and generate a spatial mode matrix and a temporal mode matrix;

[0090] A machine learning model training module 300 is configured to input the time modal matrix and the operating parameters of the aircraft engine into the machine learning model for model training to obtain a trained machine learning model;

[0091] The reconstruction prediction module 400 is used to use the trained machine learning model to predict the time modal matrix of the test condition and reconstruct the predicted field value matrix based on the spatial modal matrix.

[0092] Furthermore, the high-dimensional data acquisition module 100 is further configured to:

[0093] The KJ66 micro-turbojet engine was simulated in three dimensions using STARCCM+ fluid dynamics simulation software. The fluid domain meshes of varying precision were generated. The generated fluid domain meshes were then imported into ANSYS-FLUENT for solution. Preliminary numerical calculation results were generated through mesh independence verification.

[0094] Determining the engine speed, ambient pressure, and gas-fuel ratio based on the preliminary numerical calculation results, and selecting a plurality of operating points using a uniform stratified sampling method to perform numerical calculations on the operating points to obtain numerical calculation results for each operating point;

[0095] The numerical calculation results are imported into the Tecplot program to extract the required geometric parts and key parameter data to generate a .dat file that can be read by Tecplot.

[0096] Furthermore, the intrinsic orthogonal decomposition module 200 is further configured to:

[0097] Perform POD decomposition on the parameter part of the .dat file generated by Tecplot, and perform POD decomposition on a high-dimensional information matrix U with a specific distribution in time and space. mn =[u(x i ,t j )], each column represents the signal at all locations in local time, and each row represents the signal at all times at the local location. Calculate the information mean field data:

[0098]

[0099] Calculate the information fluctuation matrix:

[0100]

[0101] Calculate the covariance matrix R:

[0102]

[0103] Perform eigenvalue decomposition and obtain the characteristic matrix D and eigenvector matrix of the covariance matrix through eigenvalue decomposition

[0104] R·D=Φ·D

[0105] The characteristic matrix D is a diagonal matrix, and the eigenvalues ​​λ on the diagonal are i is the corresponding eigenvector The eigenvalues ​​and eigenvector matrices of Eigenvector is the corresponding time mode;

[0106] Finally, the spatial mode Ψ is calculated, where each column is a spatial mode:

[0107]

[0108] Furthermore, the machine learning model training module 300 is also used to:

[0109] Acquiring training data, preprocessing the training data, and dividing the preprocessed data into a training set, a test set, and a validation set; the training data includes operating parameters and a time modal matrix of the aircraft engine;

[0110] Construct a multivariate support vector machine (SVR) model, specifying that each column of the time modal matrix corresponds to a single SVR regressor;

[0111] During training, the input operating parameters of the aircraft engine are used as input features of each SVR regressor, and the column values ​​of the corresponding time modal matrix are used as target values; an SVR regressor is trained for each column of the time modal matrix based on the training set, and a multivariate SVR model including multiple SVR regressors is finally obtained;

[0112] The trained multivariate SVR model is tested and verified based on the test set and validation set to obtain a trained multivariate SVR model.

[0113] Furthermore, the reconstruction prediction module 400 is further configured to:

[0114] Perform a dot product operation on the spatial modal matrix and the predicted test condition time modal matrix, and add the result to the mean field data stored during POD dimensionality reduction to obtain the predicted field data;

[0115] The relative root mean square error and fitting coefficient are used to evaluate the difference and fitting degree between the predicted field data and the actual field data to obtain the data evaluation results;

[0116] Based on the data evaluation results, the predicted field data replaces the parameter part in the original Tecplot .dat file to generate a visual prediction result.

[0117] The aircraft engine internal field prediction system based on a combination of POD dimensionality reduction and machine learning, according to an embodiment of the present invention, aims to address the limitations of traditional methods in processing high-dimensional, nonlinear, and non-stationary data when predicting the multi-physical fields within an aircraft engine, while simultaneously improving computational efficiency and prediction accuracy. By adjusting the order of the time matrix obtained by POD dimensionality reduction, the present invention minimizes the amount of data while ensuring prediction accuracy, thereby further improving the learning efficiency of machine learning predictions.

[0118] In the description of this specification, the reference terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" mean that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art can combine and combine different embodiments or examples described in this specification and features of different embodiments or examples without contradiction.

[0119] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of the technical features being referred to. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one such feature. In the description of the present invention, "plurality" means at least two, such as two, three, etc., unless otherwise specifically defined.

Claims

1. A method for predicting the internal field of an aircraft engine based on the combination of POD dimensionality reduction and machine learning, characterized in that: include: Acquire high-dimensional data of the entire aircraft engine through three-dimensional numerical simulation; wherein the high-dimensional data includes Mach number field, total temperature field and total pressure field data; The POD dimensionality reduction method based on the proper orthogonal decomposition is used to reduce the dimensionality of high-dimensional data to extract characteristic modes and generate spatial mode matrix and temporal mode matrix; Inputting the time modal matrix and the operating parameters of the aircraft engine into the machine learning model for model training to obtain a trained machine learning model; The trained machine learning model is used to predict the time modal matrix of the test condition, and the predicted field value matrix is ​​reconstructed based on the spatial modal matrix.

2. The method according to claim 1, characterized in that Obtain high-dimensional data of the entire aircraft engine through three-dimensional numerical simulation, including: The KJ66 micro-turbojet engine was simulated in three dimensions using STARCCM+ fluid dynamics simulation software. The fluid domain meshes of varying precision were generated. The generated fluid domain meshes were then imported into ANSYS-FLUENT for solution. Preliminary numerical calculation results were generated through mesh independence verification. Determining the engine speed, ambient pressure, and gas-fuel ratio based on the preliminary numerical calculation results, and selecting a plurality of operating points using a uniform stratified sampling method to perform numerical calculations on the operating points to obtain numerical calculation results for each operating point; The numerical calculation results are imported into the Tecplot program to extract the required geometric parts and key parameter data to generate a .dat file that can be read by Tecplot.

3. The method according to claim 2, characterized in that The POD dimensionality reduction method based on the intrinsic orthogonal decomposition is used to reduce the dimensionality of high-dimensional data to extract characteristic modes, and generate a spatial modal matrix and a time modal matrix after dimensionality reduction, including: Perform POD decomposition on the parameter part of the .dat file generated by Tecplot, and perform POD decomposition on a high-dimensional information matrix U with a specific distribution in time and space. mn =[u(x i ,t j )], each column represents the signal at all locations in local time, and each row represents the signal at all times at the local location. Calculate the information mean field data: Calculate the information fluctuation matrix: Calculate the covariance matrix R: Perform eigenvalue decomposition and obtain the characteristic matrix D and eigenvector matrix of the covariance matrix through eigenvalue decomposition R·Φ=Φ·D The characteristic matrix D is a diagonal matrix, and the eigenvalues ​​λ on the diagonal are i is the corresponding eigenvector The eigenvalues ​​and eigenvector matrices of Eigenvector is the corresponding time mode; Finally, the spatial mode Ψ is calculated, where each column is a spatial mode:

4. The method according to claim 1, wherein The time modal matrix and the operating parameters of the aircraft engine are input into the machine learning model for model training to obtain a trained machine learning model, including: Acquiring training data, preprocessing the training data, and dividing the preprocessed data into a training set, a test set, and a validation set; the training data includes operating parameters and a time modal matrix of the aircraft engine; Construct a multivariate support vector machine (SVR) model, specifying that each column of the time modal matrix corresponds to a single SVR regressor; During training, the input operating parameters of the aircraft engine are used as input features of each SVR regressor, and the column values ​​of the corresponding time modal matrix are used as target values; an SVR regressor is trained for each column of the time modal matrix based on the training set, and a multivariate SVR model including multiple SVR regressors is finally obtained; The trained multivariate SVR model is tested and verified based on the test set and validation set to obtain a trained multivariate SVR model.

5. The method according to claim 3, characterized in that The trained machine learning model is used to predict the time modal matrix of the test condition, and the predicted field value matrix is ​​reconstructed based on the spatial modal matrix, including: Perform a dot product operation on the spatial modal matrix and the predicted test condition time modal matrix, and add the result to the mean field data stored during POD dimensionality reduction to obtain the predicted field data; The relative root mean square error and fitting coefficient are used to evaluate the difference and fitting degree between the predicted field data and the actual field data to obtain the data evaluation results; Based on the data evaluation results, the predicted field data replaces the parameter part in the original Tecplot .dat file to generate a visual prediction result.

6. An aircraft engine internal field prediction system based on the combination of POD dimension reduction and machine learning, characterized in that: include: A high-dimensional data acquisition module is used to obtain high-dimensional data of the entire aircraft engine through three-dimensional numerical simulation; wherein the high-dimensional data includes Mach number field, total temperature field and total pressure field data; The intrinsic orthogonal decomposition module is used to reduce the dimensionality of high-dimensional data using the intrinsic orthogonal decomposition (POD) dimensionality reduction method to extract characteristic modes and generate spatial mode matrix and temporal mode matrix; A machine learning model training module is used to input the time modal matrix and the operating parameters of the aircraft engine into the machine learning model for model training to obtain a trained machine learning model; The reconstruction prediction module is used to use the trained machine learning model to predict the time modal matrix of the test condition and reconstruct the predicted field value matrix based on the spatial modal matrix.

7. The system according to claim 6, characterized in that The high-dimensional data acquisition module is also used to: The KJ66 micro-turbojet engine was simulated in three dimensions using STARCCM+ fluid dynamics simulation software. The fluid domain meshes of varying precision were generated. The generated fluid domain meshes were then imported into ANSYS-FLUENT for solution. Preliminary numerical calculation results were generated through mesh independence verification. Determining the engine speed, ambient pressure, and gas-fuel ratio based on the preliminary numerical calculation results, and selecting a plurality of operating points using a uniform stratified sampling method to perform numerical calculations on the operating points to obtain numerical calculation results for each operating point; The numerical calculation results are imported into the Tecplot program to extract the required geometric parts and key parameter data to generate a .dat file that can be read by Tecplot.

8. The system according to claim 7, characterized in that The intrinsic orthogonal decomposition module is also used for: Perform POD decomposition on the parameter part of the .dat file generated by Tecplot, and perform POD decomposition on a high-dimensional information matrix U with a specific distribution in time and space. mn =[u(x i ,t j )], each column represents the signal at all locations in local time, and each row represents the signal at all times at the local location. Calculate the information mean field data: Calculate the information fluctuation matrix: Calculate the covariance matrix R: Perform eigenvalue decomposition and obtain the characteristic matrix D and eigenvector matrix of the covariance matrix through eigenvalue decomposition R·D=Φ·D The characteristic matrix D is a diagonal matrix, and the eigenvalues ​​λ on the diagonal are i is the corresponding eigenvector The eigenvalues ​​and eigenvector matrices of Eigenvector is the corresponding time mode; Finally, the spatial mode Ψ is calculated, where each column is a spatial mode:

9. The system according to claim 6, wherein: The machine learning model training module is also used to: Acquiring training data, preprocessing the training data, and dividing the preprocessed data into a training set, a test set, and a validation set; the training data includes operating parameters and a time modal matrix of the aircraft engine; Construct a multivariate support vector machine (SVR) model, specifying that each column of the time modal matrix corresponds to a single SVR regressor; During training, the input aircraft engine operating parameters are used as the input features of each SVR regressor, and the column values ​​of the corresponding time modal matrix are used as target values; Based on the training set, an SVR regressor is trained for each column of the time modality matrix, and finally a multivariate SVR model including multiple SVR regressors is obtained; The trained multivariate SVR model is tested and verified based on the test set and validation set to obtain a trained multivariate SVR model.

10. The system according to claim 8, wherein: Reconstructed prediction module, also used for: Perform a dot product operation on the spatial modal matrix and the predicted test condition time modal matrix, and add the result to the mean field data stored during POD dimensionality reduction to obtain the predicted field data; The relative root mean square error and fitting coefficient are used to evaluate the difference and fitting degree between the predicted field data and the actual field data to obtain the data evaluation results; Based on the data evaluation results, the predicted field data replaces the parameter part in the original Tecplot .dat file to generate a visual prediction result.

Images