Digital twin engine-oriented POD-SVD random forest order reduction method and digital twin engine-oriented POD-SVD random forest order reduction system

Abstract

The invention belongs to the technical field of reduced-order model development, and particularly relates to a POD-SVD random forest reduced-order method and system for digital twin engines. Adjusting a plurality of simulation data generated by the digital twin engine under different working condition boundary conditions; extracting a physical quantity of the region of interest; combining and forming an original matrix required by PODSVD decomposition; decomposing an original matrix into three matrixes by using a PODSVD algorithm, and converting the three matrixes into a modal coefficient matrix, a modal matrix and an energy matrix through a calculation formula; training a random forest algorithm; inputting a boundary condition parameter corresponding to a working condition needing to be predicted into the trained random forest model, and generating a prediction modal coefficient; calculating and restoring a physical quantity needing to be predicted; and adding grid information for the obtained reduced physical quantity, and generating a. Dat file for post-processing. The method is used for breaking through the limitation of tradeoff between cost and model precision in traditional researches such as test / CFD, and meanwhile, the model precision and the evaluation speed are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of reduced-order model development, and specifically relates to a POD-SVD random forest order reduction method and system for digital twin engines. Background Art

[0002] Turbomachinery plays a vital role in fields such as aerospace and energy and chemical engineering. However, simulating its complex internal flows faces a trade-off between high computational cost and model accuracy. In recent years, research methods based on reduced-order models and machine learning have gained attention, aiming to achieve efficient and high-precision simulation predictions.

[0003] As the prime mover for thermal conversion, turbomachinery plays a crucial role in national defense and economic development, including aerospace, energy, and chemical engineering. However, the complex internal flows within turbomachinery make its simulation a typically expensive, high-dimensional problem. Fatigue vibration in jet engine impellers primarily stems from fluid-structure interaction phenomena such as blade flutter. The blade stress error compared to experimental measurements when only centrifugal force is considered is 50%, while the calculated error when considering fluid-structure interaction is within 10%. The impact of engine fluid-structure interaction cannot be ignored, making it essential for engine simulation. To address these characteristics of turbomachinery, three main approaches have been used: experimental, theoretical, and numerical. Traditional research based on experimental and numerical methods often faces a trade-off between cost and model accuracy. To achieve higher-fidelity simulation accuracy, mesh refinement and unsteady computational methods are required, both of which are complex. In recent years, with the rapid development of technologies such as reduced-order models and machine learning, a fourth, data-driven approach has emerged. One goal is to overcome the cost-versus-model accuracy trade-off inherent in traditional research methods like testing and CFD, while simultaneously improving model accuracy and evaluation speed—the so-called "fast and good." This paper combines the Proper Orthogonal Decomposition (POD) method with the random forest algorithm to develop a reduced-order model approach that can rapidly predict compressor blade loads. This approach improves the speed and accuracy of compressor and turbine flow field simulation and fluid-structure interaction performance evaluation. Summary of the Invention

[0004] The present invention provides a POD-SVD random forest order reduction method for digital twin engines, which is used to break through the limitation of "trade-off between cost and model accuracy" in traditional research such as experiments / CFD, while improving model accuracy and evaluation speed.

[0005] The present invention provides a POD-SVD random forest order reduction system for digital twin engines, which is used to implement a POD-SVD random forest order reduction method for digital twin engines.

[0006] The present invention is achieved through the following technical solutions: A POD-SVD random forest order reduction method for a digital twin engine, the method comprising the following steps: Step 1, adjusting a plurality of simulation data generated under different working conditions of the digital twin engine; Step 2, extracting physical quantities of a region of interest from the plurality of simulation data of step 1 through a data acquisition code, the physical quantities of the region of interest including extracting compressor blade surface static pressure, total pressure and total temperature data; Step 3, combining the physical quantities extracted in step 2 to form an original matrix required for POD_SVD decomposition; Step 4, using a POD_SVD algorithm to decompose the original matrix of step 3 into three matrices, and converting the three matrices into a modal coefficient matrix, a modal matrix and an energy matrix through a formula; Step 5, inputting the boundary conditions of the engine and the modal coefficient matrix as training data into a random forest algorithm for training; Step 6, inputting boundary condition parameters corresponding to a working condition to be predicted into the trained random forest model of step 5, and then the random forest model generates predicted modal coefficients; Step 7, calculating and restoring the physical quantities to be predicted by the modal and modal coefficient matrix with an energy ratio of > 99%; Step 8, adding grid information to the restored physical quantities obtained in step 7 to generate a.dat file for post-processing.

[0007] Further, the step 1 is specifically, assuming that the rotational speed and back pressure of the blade are changed, the blade surface load under m working conditions is simulated; The step 2 is specifically, there are n static pressure data on the blade surface of each working condition, then the original data set matrix is constructed , corresponding to the boundary condition value ; Each working condition has two parameters of rotational speed and back pressure ; The working condition to be predicted is .

[0008] Further, the step 3 is specifically, the original data set first calculates the average value of m working conditions:

[0009] Wherein, t represents the row of the matrix, x represents the column of the matrix, means adding each row of data to obtain the average value of the space point static pressure, and the obtained ; Each row of the original data set is subtracted from the average matrix to obtain a matrix for SVD decomposition : .

[0010] Furthermore, the step 4 is specifically as follows: Perform SVD decomposition to obtain three matrices U, S, and V:

[0011] in, , the elements on the diagonal of S are The singular values ​​of , V is the modal matrix of POD.

[0012] Furthermore, the step 5 is specifically as follows: the modal coefficient matrix , the modal energy matrix , the modal coefficient An obtained by POD_SVD decomposition is used as the y value and the boundary condition B is used as the x value to construct the data set , as shown below:

[0013] Use the random forest algorithm to train the model; assuming that the model integrates K decision trees, K different training data are generated based on the training data D , use the data subset to train K decision trees separately.

[0014] Furthermore, the step 6 is specifically as follows: after the training is completed, the operating parameters to be predicted are input. , calculate the mean of the results predicted by K decision trees. The calculation formula is as follows:

[0015] Among them, the generated is the modal coefficient corresponding to the predicted working condition.

[0016] Furthermore, the step 7 is specifically to restore the static pressure on the corresponding blade surface using the following formula:

[0017] Here, k is the number of modes required to achieve 99% energy.

[0018] Furthermore, the step 8 is specifically to restore the predicted static pressure Using the following formula, the average static pressure matrix plus the modal multiplied by the modal coefficient is superimposed, where the restored : .

[0019] A POD-SVD random forest order reduction system for digital twin engines, the system using the above-mentioned POD-SVD random forest order reduction method for digital twin engines, the system comprising: Physical quantity extraction module: adjusts multiple simulation data generated under different operating boundary conditions of the digital twin engine; Extracting physical quantities of a region of interest from a plurality of simulation data using a data acquisition code, wherein the physical quantities of the region of interest include extracting static pressure on a compressor blade surface and blade shape data, and extracting vertex coordinates x, y, and z of the blade shape; Matrix decomposition module: combines the extracted physical quantities into the original matrix required for POD_SVD decomposition; Use the POD_SVD algorithm to decompose the original matrix into three matrices, and convert the three matrices into modal coefficient matrix, modal matrix and energy matrix through the calculation formula; Training module: The engine boundary conditions and modal coefficient matrix are used as training data to input into the random forest algorithm for training; Restoration prediction module: Input the boundary condition parameters corresponding to the working condition to be predicted into the trained random forest model, and the random forest model will generate the predicted modal coefficients; Ensure that the modes and modal coefficient matrices with energy proportions greater than 99% are calculated to restore the physical quantities that need to be predicted; .dat file generation module: adds grid information to the restored physical quantities and generates .dat files for post-processing.

[0020] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the method described above is implemented.

[0021] The beneficial effects of the present invention are: The present invention uses simulation-generated data for training, and can quickly predict the distribution cloud diagrams of physical quantities such as blade surface static pressure, total pressure, and total temperature.

[0022] This method is also applicable to generating data such as Mach number, total temperature, and total pressure for blade height sections, and has a certain degree of applicability to similar problems. It improves the simulation efficiency of components such as compressors and turbines, facilitates rapid analysis of component performance, and accelerates product development. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 It is a flow chart of the fluid-solid weak coupling steps.

[0024] Figure 2 It is a flowchart of the basic steps of POD.

[0025] Figure 3This is a flowchart of the basic steps of the POD_RBF / POD_RBFN reduced-order model.

[0026] Figure 4 It is a flow chart of the random forest algorithm of the present invention.

[0027] Figure 5 This is the overall flow chart of the operation of the POD_SVD_Random Forest algorithm of the present invention.

[0028] Figure 6 This is the flow chart of the POD_SVD_Random Forest core algorithm of the present invention.

[0029] Figure 7 This is the flow chart of the POD_SVD_random forest prediction of blade fluid-solid coupling of the present invention.

[0030] Figure 8 is a graph showing how the first two modal coefficients of the vertex and static pressure predicted by the POD_SVD_Random Forest of the present invention change over time, wherein Figure 8 (a) shows how the first two modal coefficients of the vertex X-axis change over time, Figure 8 (b) shows how the first two modal coefficients of the vertex Y-axis change over time, Figure 8 (c) shows how the first two modal coefficients of the vertex Z-axis change over time, and Figure 8 (d) shows how the first two modes of static pressure change over time.

[0031] Figure 9 shows the static pressure distribution and error distribution on the blade surface, where Figure 9 (a) shows the simulation-prediction-error distribution of the blade suction surface at t=0.0146s, and Figure 9 (b) shows the simulation-prediction-error distribution of the blade pressure surface at t=0.0146s. DETAILED DESCRIPTION

[0032] In the following description, specific details such as specific system structures and technologies are provided for illustration rather than limitation to facilitate a thorough understanding of the embodiments of the present application. However, it should be clear to those skilled in the art that the present application may be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to avoid obstructing the description of the present application with unnecessary details.

[0033] It will be understood that when used in this specification and the appended claims, the term "comprising" indicates the presence of described features, integers, steps, operations, elements and / or components, but does not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof.

[0034] It should also be understood that the terms used in this specification are only for the purpose of describing specific embodiments and are not intended to limit the present application. As used in this specification and the appended claims, the singular forms "a," "an," and "the" are intended to include the plural forms unless the context clearly indicates otherwise.

[0035] The following is a clear and complete description of the technical solutions in the embodiments of this application in conjunction with the drawings in the specification of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0036] In the following description, many specific details are set forth to facilitate a full understanding of the present application. However, the present application may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present application. Therefore, the present application is not limited to the specific embodiments disclosed below.

[0037] Implementation Method 1 Compressor Turbine Flow Field Simulation Compressor and turbine simulations primarily utilize software such as Fluent, CFX, NumecaFINE / Turbo, and STAR-CCM+. Accurately capturing the complex flow phenomena between compressor and turbine rows requires extremely high mesh density, poor convergence in transonic conditions, and a small time step size. These issues result in a massive mesh size for full 3D unsteady computations of multi-stage compressors and turbines, and a single calculation can take up to several weeks.

[0038] Compressor Turbine Fluid-Structure Coupling Simulation (Weak Coupling) The fatigue vibration of compressors and turbines is mainly caused by fluid-structure coupling phenomena such as blade flutter. When only considering the centrifugal force, the blade stress has an error of 50% compared with the experimental measurement value, while the calculation error considering the fluid-structure coupling effect is within 10%. Therefore, when designing compressors and turbines, it is necessary to consider both aerodynamic loads and centrifugal loads, which requires the use of fluid-structure coupling simulation. The commonly used fluid-structure coupling simulation method is the weak coupling method, such as Figure 1 As shown, the fluid-structure interaction problem is split into the fluid field and the solid field. The two fields are calculated separately using the fluid solver and the solid solver, respectively. Physical quantities are exchanged through the fluid-solid interface. The fluid solver calculates blade loads and transmits them to the solid solver, while the solid solver calculates displacements and transmits them to the fluid solver. This coupling is called weak fluid-structure coupling. This coupling requires iterative back-and-forth calculations between the fluid and solid solvers, resulting in greater computational resource overhead and longer time consumption compared to separate fluid simulations.

[0039] POD Technology Proper Orthogonal Decomposition (POD), also known as proper orthogonal decomposition, is a data-based model reduction method widely used in fields such as fluid dynamics and structural mechanics. Its core idea is to reduce model complexity and computational cost by capturing the key dynamic characteristics of the system.

[0040] The core idea of ​​the POD reduction model is to project the high-dimensional data of a complex system into a low-dimensional space to extract the main dynamic characteristics of the system. Through POD reduction, the amount of calculation can be significantly reduced while maintaining high accuracy. The basic steps are as follows: Figure 2 shown.

[0041] Data acquisition: Obtaining data matrices of high-dimensional systems from numerical simulations or experiments.

[0042] Covariance matrix calculation: Use the data matrix to construct the covariance matrix to describe the correlation of the data. Eigenvalue decomposition: Perform eigenvalue decomposition on the covariance matrix to obtain eigenvalues ​​and eigenvectors. Eigenvalues ​​correspond to the energy of the mode, while eigenvectors represent the dominant mode of the system.

[0043] Select dominant modes: Based on the size of the eigenvalues, select the main modes to approximate the original system. Usually, a smaller number of modes can capture most of the energy of the system.

[0044] Reduced-order model construction: A low-dimensional reduced-order model is constructed using the selected dominant modes, which can be used to replace the high-dimensional model for fast simulation.

[0045] Currently, the commonly used modal decomposition methods are divided into three categories: original POD, snapshot POD and POD-SVD.

[0046] Original POD: decomposes the data into modes and modal coefficients. However, due to matrix operations, the decomposition takes a long time when the data volume is large.

[0047] Snapshot POD, also known as POD-SnapShop, is a method that can only decompose small-scale grid data compared to the original POD method. Snapshot POD reduces the matrix to be solved by transposing the matrix, thereby achieving decomposition of large-scale grids.

[0048] POD-SVD: POD based on singular value decomposition. Its principle is similar to snapshot POD. The difference is that snapshot POD uses matrix transposition to reduce the matrix. POD-SVD is implemented based on singular value decomposition. Compared with snapshot POD, the calculation accuracy of singular value decomposition is higher, and the speed is not significantly reduced compared with snapshot POD. For a given matrix , SVD decomposes it into the product of three matrices:

[0049] in Is a left singular vector matrix, the column vector is an orthogonal basis of A; is a diagonal matrix containing the singular values ​​of A; is the right singular vector matrix, and the column vector is the orthogonal basis of A. The modal coefficients, modes, and energy required to restore the flow field are calculated through these three matrices. is the modal coefficient; is modal; is the energy of the mode. This completes the POD decomposition information. By multiplying the modal coefficients by the modes and then superimposing them, the flow field can be restored. The modes are fixed, and different operating conditions correspond to different modal coefficients.

[0050] There are currently existing POD_RBF (Proper Orthogonal Decomposition_Radial Basis Function) and POD_RBFN (Proper Orthogonal Decomposition_Radial Basis Function Neural Network) algorithms.

[0051] The POD_RBF and POD_RBFN algorithms are similar. After performing POD modal decomposition, the modal coefficients obtained by decomposition are used as y values ​​and the boundary conditions are used as x values ​​for RBF interpolation or RBF neural network training. After that, a new boundary condition needs to be input, and RBF interpolation or RBFN neural network can be used to obtain the predicted modal coefficients. The modal coefficients are multiplied by the modes and superimposed to restore the predicted physical quantity information. The specific process is as follows Figure 3 As shown: Currently, the POD_RBF and POD_RBFN algorithms perform well in predicting data of different blade height sections when the speed remains unchanged and only the back pressure changes. However, the prediction effect is poor when both the speed and back pressure change simultaneously.

[0052] Based on the above principles, the embodiment of the present invention provides a POD-SVD random forest order reduction method for digital twin engines. The POD-SVD random forest algorithm is used to read simulation data to generate a reduced-order model, and the reduced-order model is used to quickly predict the distribution of physical quantities. The Random Forest algorithm is a widely used ensemble learning method in machine learning, widely used for classification and regression problems. The basic idea of ​​Random Forest is to integrate multiple prediction results based on multiple decision trees to form a final prediction value. For regression tasks, assuming the model integrates K decision trees, its prediction value can be expressed as:

[0053] in, Represents the prediction result of the kth decision tree, x is the feature vector of the input sample, Is the final predicted value. Each decision tree It is obtained by training on a subset generated by bootstrap sampling (i.e., bagging) of the training set.

[0054] When training a decision tree, each tree not only randomly samples the sample data, but also randomizes the feature selection. For example, for a dataset with M features, each tree randomly selects m features (usually or , and select split features in this subset to ensure the differences between trees.

[0055] The optimization goal of random forest regression is to minimize the mean squared error:

[0056] Where N is the number of samples, is the actual value, is the predicted value. Random Forest reduces the variance and bias by constructing multiple decision trees and taking the average, making the model have better prediction results.

[0057] from Figure 4 As you can see, the random forest algorithm randomly generates different data subsets from the original dataset D (duplicate data is allowed in different subsets). The data in each subset is not exactly the same. Different data subsets are used to train different decision trees. When parameters are input, each decision tree is calculated separately. The combiner combines multiple decision trees. In regression problems, the average of the multiple regression results is used as the final result.

[0058] The method comprises the following steps: step 1, adjusting a plurality of simulation data generated under different operating boundary conditions of the digital twin engine; Step 2: extracting physical quantities of the area of ​​interest from the multiple simulation data in step 1 through a data acquisition code, wherein the physical quantities of the area of ​​interest include extracting data such as static pressure, total pressure, total temperature, and blade shape of the compressor blade surface, and extracting the vertex coordinates x, y, and z of the blade shape; Step 3: Combine the physical quantities extracted in step 2 to form the original matrix required for POD_SVD decomposition; Step 4: Use the POD_SVD algorithm to decompose the original matrix in step 3 into three matrices, and convert the three matrices into modal coefficient matrix, modal matrix and energy matrix through the calculation formula; Step 5: The engine boundary conditions (such as rotor speed, outlet back pressure) and modal coefficient matrix are used as training data and fed into the random forest algorithm for training; Step 6: Input the boundary condition parameters corresponding to the working condition to be predicted into the random forest model trained in step 5, and the random forest model will generate the predicted modal coefficients; Step 7: Ensure that the modes and modal coefficient matrices with energy proportions greater than 99% are calculated to restore the physical quantities that need to be predicted; That is, each mode corresponds to an energy, and the sum of all the modes' energies is 100%. The more modes you select, the greater the energy contribution. Select the first n modes, ensuring that the total energy of the selected modes is greater than 99%. If the total energy of the first n modes is less than 99%, select the first n+1 modes, and so on until the energy contribution is greater than 99%.

[0059] Step 8: Add grid information to the restored physical quantity obtained in step 7 and generate a .dat file for post-processing.

[0060] Furthermore, the step 1 specifically includes assuming that the two parameters of the blade speed and back pressure are changed, and simulating to obtain the blade surface load under m working conditions; Specifically, step 2 is as follows: there are n static pressure data on the surface of each blade under each working condition, so the original data set matrix is ​​constructed , which means that each line corresponds to the static pressure data of the blade surface for one working condition, and the corresponding boundary condition value ; Each working condition has two parameters: speed and back pressure ; The working conditions that need to be predicted are .

[0061] Furthermore, the step 3 is specifically that the original data set needs to first calculate the average value of m working conditions, and the formula is as follows:

[0062] Among them, t represents the row of the matrix (working condition) and x represents the column of the matrix (space point). This means that the average value of the static pressure at each point in space is calculated by adding up each row of data. ; Each row of the original data set is subtracted from the mean matrix to obtain the matrix for SVD decomposition , the formula is as follows: .

[0063] Furthermore, the step 4 is specifically as follows: Perform SVD decomposition to obtain three matrices U, S, and V:

[0064] in, , the elements on the diagonal of S are The singular values ​​of , V is the modal matrix of POD.

[0065] Furthermore, the step 5 is specifically as follows: the modal coefficient matrix , the modal energy matrix , thus obtaining the modal coefficient matrix, modal matrix and energy matrix, the modal coefficient An obtained by POD_SVD decomposition is used as the y value, and the boundary condition B is used as the x value to construct the data set (the first two columns are the working condition parameters, and the last m columns are the m modal coefficients corresponding to the working condition) As shown below:

[0066] Use the random forest algorithm to train the model; assuming that the model integrates K decision trees, K different training data are generated based on the training data D , use this data subset to train K decision trees separately.

[0067] Furthermore, the step 6 is specifically as follows: after the training is completed, the operating parameters to be predicted are input. , calculate the mean of the results predicted by K decision trees. The calculation formula is as follows:

[0068] Among them, the generated is the modal coefficient corresponding to the predicted working condition.

[0069] Furthermore, the step 7 is specifically to restore the static pressure on the corresponding blade surface using the following formula according to the modal coefficient matrix, modal matrix and energy matrix: .

[0070] Here, k is the number of modes required to achieve 99% energy.

[0071] Furthermore, the step 8 is specifically to restore the predicted static pressure Using the following formula, the average static pressure matrix plus the modal multiplied by the modal coefficient is superimposed, where the restored : .

[0072] The corresponding core algorithm process is as follows Figure 6 shown.

[0073] When using the POD_SVD_Random Forest method to predict the fluid-structure interaction process of compressor and turbine blades, it is necessary to add a time variable to predict the blade shape and blade surface load at different times.

[0074] Considering that the blade shape changes over time, when extracting data, in addition to extracting physical quantities such as static pressure, it is also necessary to extract the blade shape (x, y, z coordinates of the vertex). At runtime, the x, y, z, and static pressure data need to be predicted separately and saved in a .dat file.

[0075] Similarly, when constructing the original training data D for training random forests, it is necessary to add the time variable T. .

[0076]

[0077] Figure 8 shows the temporal evolution of the vertex coordinates and static pressure modal coefficients. The vertex coordinates xyz and the first two modal coefficients of static pressure exhibit a regularity similar to that of blade vibration.

[0078] As shown in Figure 9, the maximum error of the blade surface static pressure prediction result is 1.1%. The error is concentrated in a small area of ​​the blade tip, and the error in most areas is less than 0.5%. The reduced-order model has a high accuracy in predicting the blade surface static pressure.

[0079] Implementation Method 2 An embodiment of the present invention provides a POD-SVD random forest order reduction system for a digital twin engine. The system uses the POD-SVD random forest order reduction method for a digital twin engine as described in embodiment 1. The system includes: Physical quantity extraction module: adjusts multiple simulation data generated under different operating boundary conditions of the digital twin engine; Extracting physical quantities of a region of interest from a plurality of simulation data using a data acquisition code, wherein the physical quantities of the region of interest include extracting data such as static pressure, total pressure, total temperature, and blade shape of a compressor blade surface, and extracting vertex coordinates x, y, and z of the blade shape; Matrix decomposition module: combines the extracted physical quantities into the original matrix required for POD_SVD decomposition; Use the POD_SVD algorithm to decompose the original matrix into three matrices, and convert the three matrices into modal coefficient matrix, modal matrix and energy matrix through the calculation formula; Training module: The engine boundary conditions (e.g., rotor speed, outlet back pressure) and modal coefficient matrix are used as training data to feed the random forest algorithm for training; Restoration prediction module: Input the boundary condition parameters corresponding to the working condition to be predicted into the trained random forest model, and the random forest model will generate the predicted modal coefficients; Ensure that the modes and modal coefficient matrices with energy proportions greater than 99% are calculated to restore the physical quantities that need to be predicted; That is, each mode corresponds to an energy, and the sum of all the modes' energies is 100%. The more modes you select, the greater the energy contribution. Select the first n modes, ensuring that the total energy of the selected modes is greater than 99%. If the total energy of the first n modes is less than 99%, select the first n+1 modes, and so on until the energy contribution is greater than 99%.

[0080] .dat file generation module: adds grid information to the restored physical quantities and generates .dat files for post-processing.

[0081] As can be seen from the above, the embodiment of the present invention uses POD_SVD to decompose the original matrix into modes, modal coefficients, and energy. The boundary condition matrix is ​​combined with the modal matrix to construct the original training data D. The boundary conditions are used as independent variables and the modal coefficients are used as dependent variables to train and generate a random forest model. The boundary conditions of the working conditions to be predicted are input, and the random forest will output the predicted modal coefficients. The first k predicted modal coefficients are multiplied by the first k modes and accumulated to restore the predicted data. Experimental results show that this system improves the simulation efficiency of components such as compressors and turbines, facilitates rapid analysis of component performance, and accelerates product development.

[0082] Implementation Method 3 An embodiment of the present invention provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. The memory is used to store software programs and modules, and the processor executes various functional applications and data processing by executing the software programs and modules stored in the memory. The memory and processor are connected via a bus. Specifically, the processor implements any step of the first embodiment described above by executing the computer program stored in the memory.

[0083] It should be understood that in the embodiments of the present invention, the processor referred to may be a central processing unit (CPU), and the processor may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc.

[0084] The memory may include a read-only memory, a flash memory, and a random access memory, and provides instructions and data to the processor. A portion or all of the memory may also include a non-volatile random access memory.

[0085] It should be understood that if the above-mentioned integrated modules / units are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the present invention can also implement all or part of the processes in the above-mentioned method embodiments by using a computer program to instruct the relevant hardware. The above-mentioned computer program can be stored in a computer-readable storage medium. When executed by a processor, the computer program can implement the steps of each of the above-mentioned method embodiments. The above-mentioned computer program includes computer program code, which can be in source code form, object code form, executable file, or some intermediate form. The above-mentioned computer-readable medium can include: any entity or device capable of carrying the above-mentioned computer program code, recording medium, USB flash drive, mobile hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal, and software distribution medium. It should be noted that the content contained in the above-mentioned computer-readable storage medium can be appropriately increased or decreased based on the requirements of legislation and patent practice in a jurisdiction.

[0086] The above description of the disclosed embodiments will enable one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not limited to the embodiments shown herein, but is to be construed in the widest manner consistent with the principles and novel features disclosed herein.

[0087] Those skilled in the art can clearly understand that, for the convenience and brevity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In actual applications, the above-mentioned functions can be distributed and completed by different functional units and modules as needed, that is, the internal structure of the above-mentioned device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the implementation method can be integrated into one processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of the present invention. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method implementation method, and will not be repeated here.

[0088] It should be noted that the methods and detailed examples provided in the above embodiments can be combined with the devices and equipment provided in the embodiments, and references can be made to each other, and no further details will be given.

[0089] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present invention.

[0090] In the embodiments provided by the present invention, it should be understood that the disclosed apparatus / terminal equipment and methods can be implemented in other ways. For example, the apparatus / device embodiments described above are merely illustrative. For example, the division of the modules or units described above is merely a logical functional division. In actual implementation, other division methods may be used. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not implemented.

[0091] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or make equivalent replacements for some of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention, and should all be included in the scope of protection of the present invention.

Claims

1. A POD-SVD random forest order reduction method for digital twin engines, characterized by: The method comprises the following steps: Step 1: Adjust multiple simulation data generated under different operating boundary conditions of the digital twin engine; Step 2: extracting physical quantities of the area of ​​interest from the multiple simulation data in step 1 through a data acquisition code, wherein the physical quantities of the area of ​​interest include extracting static pressure, total pressure and total temperature data of the compressor blade surface; Step 3: Combine the physical quantities extracted in step 2 to form the original matrix required for POD_SVD decomposition; Step 4: Use the POD_SVD algorithm to decompose the original matrix in step 3 into three matrices, and convert the three matrices into modal coefficient matrix, modal matrix and energy matrix through the calculation formula; Step 5: The boundary conditions and modal coefficient matrix of the engine are used as training data and input into the random forest algorithm for training; Step 6: Input the boundary condition parameters corresponding to the working condition to be predicted into the random forest model trained in step 5, and the random forest model will generate the predicted modal coefficients; Step 7: Ensure that the modes and modal coefficient matrices with energy proportions greater than 99% are calculated to restore the physical quantities that need to be predicted; Step 8: Add grid information to the restored physical quantity obtained in step 7 and generate a .dat file for post-processing.

2. The method according to claim 1, characterized in that Specifically, step 1 includes assuming that the rotation speed and back pressure of the blade are changed, and simulating to obtain the blade surface load under m working conditions; Specifically, step 2 is as follows: there are n static pressure data on the surface of each blade under each working condition, so the original data set matrix is ​​constructed , the corresponding boundary condition value ; Each working condition has two parameters: speed and back pressure ; The working conditions that need to be predicted are .

3. The method according to claim 2, characterized in that Specifically, step 3 is to first calculate the average value of m working conditions of the original data set: Where t represents the row of the matrix and x represents the column of the matrix. This means that the average value of the static pressure at each point in space is calculated by adding up each row of data. ; Each row of the original data set is subtracted from the mean matrix to obtain the matrix for SVD decomposition : 。 4. The method according to claim 3, characterized in that The step 4 is specifically as follows: Perform SVD decomposition to obtain three matrices U, S, and V: in, , the elements on the diagonal of S are The singular values ​​of , V is the modal matrix of POD.

5. The method according to claim 4, characterized in that: The step 5 is specifically as follows: , the modal energy matrix , the modal coefficient An obtained by POD_SVD decomposition is used as the y value and the boundary condition B is used as the x value to construct the data set , as shown below: Use the random forest algorithm to train the model; assuming that the model integrates K decision trees, K different training data are generated based on the training data D , use the data subset to train K decision trees separately.

6. The method according to claim 5, characterized in that The step 6 is specifically as follows: after the training is completed, input the operating condition parameters to be predicted , calculate the mean of the results predicted by K decision trees. The calculation formula is as follows: Among them, the generated is the modal coefficient corresponding to the predicted working condition.

7. The method according to claim 6, characterized in that Specifically, step 7 uses the following formula to restore the static pressure on the corresponding blade surface: Here, k is the number of modes required to achieve 99% energy.

8. The method according to claim 7, characterized in that: The step 8 is specifically to restore the predicted static pressure Using the following formula, the average static pressure matrix plus the modal multiplied by the modal coefficient is superimposed, where the restored : 。 9. A POD-SVD random forest order reduction system for digital twin engines, characterized by: The system uses the POD-SVD random forest order reduction method for digital twin engines according to any one of claims 1 to 8, and the system includes: Physical quantity extraction module: adjusts multiple simulation data generated under different operating boundary conditions of the digital twin engine; Extracting physical quantities of a region of interest from a plurality of simulation data using a data acquisition code, wherein the physical quantities of the region of interest include extracting static pressure on a compressor blade surface and blade shape data, and extracting vertex coordinates x, y, and z of the blade shape; Matrix decomposition module: combines the extracted physical quantities into the original matrix required for POD_SVD decomposition; Use the POD_SVD algorithm to decompose the original matrix into three matrices, and convert the three matrices into modal coefficient matrix, modal matrix and energy matrix through the calculation formula; Training module: The engine boundary conditions and modal coefficient matrix are used as training data to input into the random forest algorithm for training; Restoration prediction module: Input the boundary condition parameters corresponding to the working condition to be predicted into the trained random forest model, and the random forest model will generate the predicted modal coefficients; Ensure that the modes and modal coefficient matrices with energy proportions greater than 99% are calculated to restore the physical quantities that need to be predicted; .dat file generation module: adds grid information to the restored physical quantities and generates .dat files for post-processing.

10. A computer device, characterized in that: The method comprises a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the method according to any one of claims 1 to 7 is implemented.

Images