Physical field prediction software architecture and prediction method based on modal decomposition

Abstract

The invention belongs to the field of digital twinning and physical field prediction, and discloses a physical field prediction software architecture and prediction method based on modal decomposition, and the physical field prediction software architecture comprises a preprocessing module and a prediction module; the preprocessing module is used for reading original data according to a preset data reading method and integrating the read data into a snapshot matrix; performing modal decomposition on the snapshot matrix according to a predetermined modal decomposition algorithm to obtain an expert evaluation database; the prediction module is used for obtaining a primary function coefficient according to a defined primary function coefficient calculation method; obtaining a physical field prediction result according to the primary function coefficient and the expert evaluation database; the method does not need to depend on grids of specific structures or complex self-programming codes, can adapt to physical field data of different types and different sources, and improves the universality of the modal decomposition technology in actual engineering application; through accurate primary function coefficient calculation and reasonable expert evaluation database application, the calculation precision of physical field prediction can be effectively improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of digital twinning and physical field prediction, and particularly relates to a physical field prediction software architecture and prediction method based on modal decomposition. BACKGROUND

[0002] With the digital transformation of industry and the rapid development of digital twinning technology, the requirements for timeliness and accuracy of physical field simulation in industrial scenarios are becoming increasingly urgent. Modal decomposition technology is an effective physical field fast and accurate construction method based on order reduction calculation, which has been relatively widely applied in the academic field. The physical field fast simulation method based on modal decomposition technology provides a highly potential technical means for the fast construction and real-time interaction of digital twinning models, and is widely used in fields such as signal processing, big data analysis, image processing and financial market analysis.

[0003] As a data analysis method, modal decomposition technology usually decomposes complex signals or data into a linear combination of a set of basic modes (modal functions). The basic mode is a local feature or vibration mode in the data, and its linear combination can reconstruct the original data. When the physical field information is selected as the data to be analyzed, modal decomposition can realize the fast construction of the physical field. The basic process is as follows: first, a set of representative physical fields (snapshots) is generated as an initial database through experimental measurement, numerical calculation and other methods; then, a suitable modal decomposition technology is used to decompose a set of modal functions based on the initial database to form an expert evaluation database; secondly, the weight coefficients corresponding to each modal function are calculated under given or input actual working conditions; finally, the physical field is quickly constructed through the linear combination of the modal function and the corresponding weight coefficient.

[0004] At present, the application of existing modal decomposition technology is mostly realized through self-programming, for example, the modal decomposition technology based on proper orthogonal decomposition (POD) algorithm, which is mainly aimed at structured grids, and the algorithm and code are extremely complex, making it difficult to ensure the universality when applied in actual engineering. Secondly, the existing order reduction model software can provide fewer modal decomposition methods and weight coefficient calculation methods, making it difficult to ensure the accuracy of the calculation results. SUMMARY

[0005] In view of the technical problems existing in the prior art, the application provides a physical field prediction software architecture and prediction method based on modal decomposition, to solve the technical problems that the application of existing modal decomposition technology is difficult to ensure the universality when applied in actual engineering and is difficult to ensure the accuracy of the calculation results.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows: This invention provides a physical field prediction software architecture based on mode decomposition, including a preprocessing module and a prediction module; The preprocessing module is used to read raw data according to a preset data reading method and integrate the read data into a snapshot matrix; and to perform mode decomposition on the snapshot matrix according to a predetermined mode decomposition algorithm to obtain an expert evaluation database; wherein, the expert evaluation database includes several basis functions; The prediction module is used to obtain basis function coefficients according to the defined basis function coefficient calculation method; and to obtain the physical field prediction result based on the basis function coefficients and the expert evaluation database.

[0007] Furthermore, the preprocessing module includes a first enumeration variable module, a data reading module, a snapshot storage module, and a mode decomposition module; The first enumeration variable module is used to determine the data format of the original data; The data reading module is used to read the original data according to a preset data reading method based on the data format discrimination result of the original data, and integrate the read data information into a snapshot matrix; The snapshot storage module is used to store the snapshot matrix; The mode decomposition module is used to perform mode decomposition on the snapshot matrix according to a pre-written mode decomposition algorithm to obtain an expert evaluation database.

[0008] Furthermore, the first enumeration variable module uses the enumeration variable ResultDataType, the data reading module is a class CASE, the snapshot storage module is a two-dimensional container variable SnapshotMatrix, and the modality decomposition module is a class DECOMPOSE.

[0009] Furthermore, the pre-written mode decomposition algorithm includes singular value decomposition algorithm, fast singular value decomposition algorithm, intrinsic orthogonal decomposition algorithm, or dynamic mode decomposition algorithm.

[0010] Furthermore, the prediction module includes a second enumerated variable module, a third enumerated variable module, a basis function coefficient calculation module, a linear combination module, and a prediction result output module; The second enumerated variable module is used to define the calculation method for the basis function coefficients; The third enumeration variable module is used to define the file output format of the physical field prediction results; The basis function coefficient calculation module is used to calculate the basis function coefficients according to the defined basis function coefficient calculation method and in conjunction with the expert evaluation database. The linear combination module is used to linearly combine the basis functions in the expert evaluation database with the basis function coefficients according to the defined class function for constructing the physical field, so as to obtain the predicted physical field. The prediction result output module is used to output the predicted physical field according to the defined physical field prediction result file output format based on the defined class function for outputting physical field information, so as to obtain the physical field prediction result.

[0011] Furthermore, the second enumeration variable module is the enumeration variable PredictType, and the third enumeration variable module is the enumeration variable OutputFileType; the basis function coefficient calculation module, the linear combination module, and the prediction result output module are all of the PREDICT class.

[0012] Furthermore, the prediction module also includes a file directory specification module; The file target specification module is used to specify the file directory of the expert evaluation database and the file directory for storing the physical field prediction results to the basis function coefficient calculation module; wherein, the file directory specification module is a string variable module.

[0013] Furthermore, the methods for calculating the defined basis function coefficients include Galerkin projection, interpolation, or machine learning methods.

[0014] Furthermore, the defined output formats for physical field prediction results include Tecplot, VTK, Fluent, CFX, or MHT formats.

[0015] The present invention also provides a physical field prediction method based on mode decomposition, utilizing the aforementioned physical field prediction software architecture based on mode decomposition; Physical field prediction methods based on mode decomposition include: The original data is read according to a preset data reading method, and the read data is integrated into a snapshot matrix; the snapshot matrix is ​​then decomposed according to a predetermined mode decomposition algorithm to obtain an expert evaluation database; The basis function coefficients are obtained according to the defined method for calculating basis function coefficients; the physical field prediction results are obtained based on the basis function coefficients and the expert evaluation database.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: The physical field prediction software architecture and prediction method based on mode decomposition provided by this invention integrates the mode decomposition process into the software. A preprocessing module reads raw data according to a preset data reading method and integrates it into a snapshot matrix. Then, mode decomposition is performed on the snapshot matrix according to a predetermined mode decomposition algorithm. This invention does not rely on a specific mesh structure or complex self-programmed code, and can adapt to different types and sources of physical field data, greatly improving the versatility of mode decomposition technology in practical engineering applications. By encapsulating the complex mode decomposition algorithm in the preprocessing module, users only need to provide the raw data according to the software's operation flow, without needing to close the software. The implementation details of the underlying algorithm lower the barrier to using mode decomposition (MDD) technology for physical field prediction. Specifically, in the preprocessing module, the snapshot matrix is ​​decomposed according to a pre-determined MMD algorithm to obtain an expert evaluation database. This allows for flexible selection and optimization of the MMD algorithm to adapt to the characteristics of different physical fields, thereby obtaining more accurate basic mode information. Simultaneously, the prediction module obtains basis function coefficients according to a defined basis function coefficient calculation method, and then combines this with the expert evaluation database to obtain the physical field prediction results. Through accurate basis function coefficient calculation and reasonable application of the expert evaluation database, the computational accuracy of physical field prediction can be effectively improved. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application 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 this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 The structural block diagram of the physical field prediction software architecture based on mode decomposition provided by the present invention; Figure 2 This is a schematic diagram of the class modules in this invention; Figure 3 A schematic diagram of the physical field prediction method based on mode decomposition provided by the present invention; Figure 4 This is a schematic diagram of the vertical framework structure of the physical field prediction software architecture in this invention; Figure 5 This is a schematic diagram illustrating the application of the present invention in a rapid velocity field prediction example of a standard three-way pipe calculation. Figure 6 This is a schematic diagram illustrating the application of the present invention in an example of predicting the thermal insulation characteristics of a nuclear reactor pressure vessel; Figure 7 This is a schematic diagram illustrating the application of the present invention in an example of predicting the hydrothermal management characteristics of a fuel cell. Detailed Implementation

[0019] To make the technical problems, technical solutions, and beneficial effects solved by this application clearer, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0020] This invention provides a physical field prediction software architecture based on mode decomposition, including a preprocessing module and a prediction module. The preprocessing module is used to read raw data according to a preset data reading method and integrate the read data into a snapshot matrix; perform mode decomposition on the snapshot matrix according to a predetermined mode decomposition algorithm to obtain an expert evaluation database; wherein, the expert evaluation database includes several basis functions; the prediction module is used to obtain basis function coefficients according to a defined basis function coefficient calculation method; and obtain physical field prediction results based on the basis function coefficients and the expert evaluation database.

[0021] The physical field prediction software architecture based on modal decomposition described in this invention reads and integrates data according to a preset method in the preprocessing module and performs modal decomposition according to a pre-determined algorithm to obtain an expert evaluation database. The prediction module calculates the basis function coefficients according to a defined method and combines them with the expert evaluation database to obtain the physical field prediction results. This enriches the modal decomposition method, helps to improve the accuracy of the calculation results, and provides more efficient and accurate technical support for physical field simulation in industrial scenarios.

[0022] The following specific embodiments further explain the physical field prediction software architecture based on mode decomposition provided by the present invention: Example As attached Figure 1 As shown, this embodiment provides a physical field prediction software architecture based on mode decomposition, including a preprocessing module and a prediction module. The preprocessing module is used to read raw data according to a preset data reading method and integrate the read data into a snapshot matrix; perform mode decomposition on the snapshot matrix according to a predetermined mode decomposition algorithm to obtain an expert evaluation database; wherein, the expert evaluation database includes several basis functions; the prediction module is used to obtain basis function coefficients according to a defined basis function coefficient calculation method; and obtain physical field prediction results based on the basis function coefficients and the expert evaluation database.

[0023] The preprocessing module includes a first enumeration variable module, a data reading module, a snapshot storage module, and a mode decomposition module. The data reading module reads the original data according to a preset data reading method based on the data format discrimination result of the original data, and integrates the read data information into a snapshot matrix. The snapshot storage module stores the snapshot matrix. The mode decomposition module performs mode decomposition on the snapshot matrix according to a pre-written mode decomposition algorithm to obtain several basis functions. Based on these basis functions, an expert evaluation database is constructed. The pre-written mode decomposition algorithm includes singular value decomposition (SVD), fast singular value decomposition (SSD), intrinsic orthogonal decomposition (IOD), or dynamic mode decomposition (VMD).

[0024] The prediction module includes a second enumerated variable module, a third enumerated variable module, a basis function coefficient calculation module, a file target specification module, a linear combination module, and a prediction result output module. The second enumerated variable module defines the calculation method for the basis function coefficients. The third enumerated variable module defines the file output format for the physics prediction results. Preferably, the defined file output format for the physics prediction results includes Tecplot, VTK, Fluent, CFX, or MHT formats. The basis function coefficient calculation module calculates the basis function coefficients according to the defined calculation method and in conjunction with the expert evaluation database. The defined calculation method for the basis function coefficients... This includes Galerkin projection, interpolation, or machine learning methods; the file target specification module is used to specify the file directory of the expert evaluation database and the file directory for storing the physical field prediction results to the basis function coefficient calculation module; wherein, the file directory specification module is a string variable module; the linear combination module is used to linearly combine the basis functions in the expert evaluation database with the basis function coefficients according to the defined class function for constructing the physical field to obtain the predicted physical field; the prediction result output module is used to output the predicted physical field according to the defined file output format of the physical field prediction results according to the defined class function for outputting physical field information to obtain the physical field prediction result.

[0025] In this embodiment, both the preprocessing module and the prediction module are implemented based on object-oriented programming methods. Corresponding functions are achieved through several class modules, thereby enhancing the modularity of the physical field prediction software architecture and facilitating collaboration and maintenance among developers. Preferably, the object-oriented programming method is implemented using C++ or Python. (See attached...) Figure 2As shown, the enumeration variable module uses the enumeration variable ResultDataType, the data reading module is a CASE class, the snapshot storage module is a two-dimensional container variable SnapshotMatrix, and the mode decomposition module is a DECOMPOSE class; the second enumeration variable module is the enumeration variable PredictType, and the third enumeration variable module is the enumeration variable OutputFileType; the basis function coefficient calculation module, the linear combination module, and the prediction result output module are all PREDICT class.

[0026] It should be noted that the CASE class can read raw data from the initial database to extract the required snapshot physics information and integrate the read data into a snapshot matrix. The CASE.read_data method in the CASE class has a preset data reading method to read the initial database in formats such as Fluent, Tecplot, MHT, or VTK, and output the required physics information such as geometry, computational networks, or physics fields in matrix form.

[0027] The two-dimensional container variable SnapshotMatrix is ​​a class that provides matrix processing functionality. Instances of this class can perform matrix copying, transposition, and addition, subtraction, multiplication, and division operations. Specifically, MATRIX.copy performs matrix copying; MATRIX.transposition performs matrix transposition; MATRIX.resize performs resizing of the number of rows and columns; and MATRIX.toString performs converting a matrix to a string. It can perform matrix addition, subtraction, multiplication, and division operations.

[0028] The DECOMPOSE class provides various mode decomposition algorithms to perform mode decomposition on snapshot matrices and establish an expert evaluation database. Specifically, the DECOMPOSE.solve method within the DECOMPOSE class is configured with algorithms such as singular value decomposition, fast singular value decomposition, eigenorthogonal decomposition, or dynamic mode decomposition, which can perform mode decomposition on the snapshot matrix according to user needs and preset truncation factors. The DECOMPOSE.write method within the DECOMPOSE class can store the obtained basis functions in a specific file format in the expert evaluation database.

[0029] The PREDICT class is used to solve for the basis function coefficients and construct the physical field under the prediction conditions based on different calculation methods for basis function coefficients predefined in the enumeration variable PredictType. Specifically, the PREDICT.solve method in the PREDICT class can read data from the expert evaluation database and the prediction condition data stream, and combine various basis function coefficient calculation methods to calculate the basis function coefficients under the set conditions according to the user's selection. The PREDICT.solve method can also construct the physical field under the prediction conditions based on the basis functions in the expert evaluation database and the calculated coefficients. The PREDICT.output method in the PREDICT class provides algorithms for generating various data file formats, allowing the obtained physical field prediction structure to be output in different file formats according to user needs.

[0030] Optionally, the preprocessing module further includes a module for reading and storing mesh information, used to read and store preset mesh information, and to write class functions for previewing and reading the mesh; wherein, the module for reading and storing mesh information is a MESH class; specifically, the MESH class can read and store mesh information including overall mesh information, node information, cell information, surface information, and topology information; wherein, the MESH.scan_Mesh_File method in the MESH class has a built-in method for previewing mesh files, and the MESH.read_Mesh_File method in the MESH class has a built-in method for reading mesh files.

[0031] In this embodiment 1, the construction process of the preprocessing module is as follows: Step 101: Construct a MESH class as a module for reading and storing mesh information, which is used to read and store preset mesh information, and write class functions to preview the mesh and read the mesh.

[0032] Step 102: Define the enumeration variable ResultDataType as the first enumeration variable module to realize the discrimination of different preset data formats.

[0033] Step 103: Construct the class CASE as the data reading module; wherein, the class CASE uses the defined enumeration variable ResultDataType as the input parameter of the construction function.

[0034] Step 104: Define the required data format reading methods in the CASE class; the defined required data format reading methods correspond to the data format identified in the defined enumeration variable ResultDataType, so as to provide the corresponding data reading methods.

[0035] Step 105: Define a two-dimensional container variable SnapshotMatrix as a snapshot storage module to store the read snapshot physical field, specifically for storing the snapshot matrix integrated through the CASE class.

[0036] Step 106: Define class DECOMPOSE as the modal decomposition module; the inputs to the constructor of class DECOMPOSE include snapshot matrix, truncated residuals, and decomposition method.

[0037] Step 107: In the DECOMPOSE class, write class functions to implement mode decomposition algorithms, such as singular value decomposition, fast singular value decomposition, intrinsic orthogonal decomposition, or dynamic mode decomposition. The pre-written mode decomposition algorithms include mesh information stored in the MESH class, and the basis functions obtained after decomposition are selected based on the truncated residuals input in the DECOMPOSE class. After software release, if new mode decomposition methods need to be added, simply add the new mode decomposition algorithm class functions to the DECOMPOSE class.

[0038] Step 108: In the DECOMPOSE class, write the data saved by adding the expert evaluation database, such as the basis function coefficients of the basis function and the snapshot, for reading in the physical field prediction part; at this point, the preprocessing module is completed.

[0039] In this embodiment 1, the construction process of the prediction module is as follows: Step 201: Define the enumeration variable PredictType as the second enumeration variable to specify the calculation method of different basis function coefficients for methods such as Galerkin projection, interpolation, and machine learning.

[0040] Step 202: Define the enumeration variable OutputFileType as the third enumeration variable to specify different output physics file formats, such as Tecplot, Fluent, CFX, MHT, and VTK.

[0041] Step 203: Define two string variables as the file directory specification module, which are used to specify the file directory of the expert evaluation database generated by the preprocessing part and the file directory to be output as the final physical field file. Specify the file directory of the expert evaluation database and the file directory to be used to store the physical field prediction results to the basis function coefficient calculation module.

[0042] Step 204: Class PREDICT, used for constructing the physical field of the basis function computation domain; wherein, the enumeration variables PredictType and OutputFileType, as well as two string variables, are defined as input parameters of the construction function to transmit the required basis function coefficient calculation method, output file format, expert evaluation database address, and output file directory.

[0043] Step 205: In the PREDICT class, write class functions to implement various basis function calculation methods, and transmit the calculated basis function coefficients in matrix form. After the software is released, if it is necessary to add new basis function coefficient calculation methods, simply add class functions to the PREDICT class to implement the new algorithms; the rest of the program framework does not need to be modified.

[0044] Step 206: Define a class function in class PREDICT to construct the physical field. The physical field is constructed by a linear combination of the basis functions and the coefficients of the solved basis functions.

[0045] Step 207: Define a class function in class PREDICT for outputting physical field information, which serves as the interface for software interaction with users / industrial equipment. The physical field output can utilize the mesh information contained in class MESH. It should be noted that different data file formats or physical field information correspond to a separate class function to facilitate modular software development and later operation and maintenance.

[0046] It is worth noting that in the MESH class, Domain is used to store mesh-related information, such as the number of regions, nodes, cells, and faces; Node, Cell, and Face are one-dimensional arrays used to store specific information about each node, cell, and face within the mesh; Topology is used to store the topological information between nodes, cells, and faces, with the main basic information including cells adjacent to a face and nodes at face corners, and optional extended information including cells and faces adjacent to a node, faces adjacent to a cell, and nodes at cell corners; scan Mesh File is used to browse and extract information from the mesh file; read Mesh File is used to read the mesh file into a MESH class object.

[0047] In the CASE class, Mesh is treated as an object of the MESH class, storing the grid information upon which the working condition is based; Result datatype is used to store the storage format type of information such as physical fields within the working condition; Field number is used to store the number of physical fields to be predicted; Physical field is used to store the specific physical field data; read Data is used to read the working condition file into the CASE class object according to the input format type stored in Result datatype; among them, the number of physical fields is stored in Field number, and the physical field information is stored in Physical field, which is used as raw snapshot data in the preprocessing part after reading.

[0048] In the MATRIX class, Size stores the number of rows and columns of the matrix; Data stores the entire matrix; copy, transposition, resize, and toString are used for operations on the matrix, namely copying, transposing, and converting to a string, respectively. The class also includes operations such as matrix addition, subtraction, matrix multiplication, and matrix division (for solving systems of algebraic equations).

[0049] The `DECOMPOSE` class, which inherits from `MATRIX`, stores the snapshot matrix to be decomposed, read from the `Case` object. `Mesh` is an object of the `MESH` class, storing the mesh information underlying the working case. `DecomposeAlgorithm` enumerates the modal decomposition methods. `Residual` sets the maximum acceptable residual in the modal decomposition algorithm. `Basic physical field` stores the baseline physical field. `Mode` stores the basis functions of the modal space. `Energycaptured` stores the energy captured by each basis function. `BMatrix` stores the weight coefficients of each snapshot in each basis function direction. `solve` performs the decomposition operation, based on the snapshot matrix in `Data` and the specific modal decomposition method stored in `DecomposeAlgorithm`. Modal decomposition is considered converged when the residual falls below `Residual`. The decomposition yields the basic physical field, the basis functions `Mode`, and the weight coefficients `BMatrix` of each snapshot in each basis function direction. The mesh information from the `Mesh` object may be used during the modal decomposition process. Finally, the energy captured by each basis function is statistically obtained. `write` is used to output the results generated in the preprocessing part to the database, and the part to be predicted is read, including the Basic physical field, Mode, BMatrix and Energy captured.

[0050] In the class `PREDICT`, `Decompose` is an array of pointers, each pointing to an object of class `DECOMPOSE`, used to store the basis function database for each physics field; `Snapshot condition` stores the load condition information for each snapshot; `Factor level` stores the level values ​​of each factor; `Predict type` enumerates the prediction methods; `Predict condition` stores the load condition information to be predicted; `Truncation error` stores the acceptable truncation error; `Predict BMatrix` stores the weight coefficient matrix of the predicted load condition; `Predict field` stores the predicted physics field; `Output file type` enumerates the output file formats; and `solve` performs the prediction operation. Specifically, first, based on the `Truncation error` and `Energy captured` in `Decompose`, `Mode` and `BMatrix` are truncated; then, using the methods stored in `Predict type`, the weight coefficients of the predicted load condition are predicted based on `Snapshot condition`, `Factor level`, `Predict condition`, `Mesh`, and `Decompose`, and stored in `Predict BMatrix`; then, based on `Predict BMatrix` and `Mode`, the predicted physics field is obtained and stored in `Predict field`; `output result` outputs the predicted physics field, based on the input and stored in `Outputfile`. The required physical field format in the type field, combined with the mesh information stored in the Mesh field in Decompose, outputs the predicted physical field as a file.

[0051] It should also be noted that, vertically, the aforementioned physical field prediction software architecture includes an interaction layer, a business layer, and a data layer, as shown in the appendix. Figure 3 As shown; specifically, the interaction layer is used to provide the human-machine interface and interaction interface for interaction with users or industrial equipment, and is responsible for data format conversion and calculation parameter setting; the business layer is responsible for the implementation of the algorithm: modal decomposition, weight coefficient calculation, and physical field construction; the data layer is used for the encapsulation and storage of software data; the main data includes the original snapshot database, expert evaluation database, basis function coefficient matrix, equation coefficient matrix, etc.; among them, the interaction layer, business layer and data layer communicate with each other through data flow.

[0052] This invention also provides a physics field prediction method based on mode decomposition, utilizing the aforementioned physics field prediction software architecture; as shown in the appendix. Figure 4As shown, the physical field prediction method based on mode decomposition includes: Using a preprocessing module, raw data is read according to a preset data reading method, and the read data is integrated into a snapshot matrix; the snapshot matrix is ​​then subjected to mode decomposition according to a predetermined mode decomposition algorithm to obtain an expert evaluation database; wherein, the expert evaluation database includes several basis functions.

[0053] Using the prediction module, the basis function coefficients are obtained according to the defined basis function coefficient calculation method; based on the basis function coefficients and the expert evaluation database, the physical field prediction results are obtained.

[0054] Example explanation: The following examples demonstrate the prediction performance of the mode decomposition-based physical field prediction software architecture described in the above embodiments, using the standard calculation example of a three-way pipe, the example of thermal insulation characteristics prediction of a nuclear reactor pressure vessel, and the example of hydrothermal management characteristics prediction of a fuel cell as examples. This is to verify the effectiveness and versatility of the software architecture described in the above embodiments in the rapid prediction of physical problems in various scenarios.

[0055] Example 1: A prediction example of the flow field within a standard T-junction; where the scene mesh diagram, the prediction results using computational fluid dynamics methods, and the prediction results using the software architecture of the embodiment are shown below. Figure 5 As shown; in this Example 1, the mesh used is a triangular mesh with quadrilateral boundary layers, with a total of approximately 2000 meshes. The CFD simulation solver used is MHT, and the original data format is MHT. The mode decomposition algorithm selected is singular value decomposition; the basis function coefficients are calculated using interpolation; the output format for the physics field prediction results is TECPLOT format; from the attached... Figure 5 As can be seen from the example, the velocity field predicted by the software architecture described in the embodiment is almost completely consistent with the results of the standard CFD method, and the simulation speed is increased by more than 100 times.

[0056] Example 2: A prediction example of thermal insulation characteristics in a nuclear reactor pressure vessel; wherein, a scenario diagram, predictions using computational fluid dynamics methods, and predictions using the software architecture of the embodiment are shown, such as... Figure 6 As shown; in this Example 2, the mesh used is a hexahedral multi-block structured mesh with interfaces, with a total mesh count of approximately 100 million. The CFD simulation solver used is FLUENT, and the raw data format is TECPLOT. The mode decomposition algorithm selected is Fast Singular Value Decomposition (FSVDe); the basis function coefficients are calculated using interpolation; the output file format for the physics field prediction results is TECPLOT. (See attached...) Figure 6As can be seen from the example, the software architecture described in the embodiment uses a reduced-order model to perform digital twins, and the obtained temperature field and heat flow field have very small errors compared with the results of the standard CFD method; the software architecture described in the embodiment can obtain multiphysics fields in 1 second, while the standard CFD method requires more than 1 day.

[0057] Example 3: A prediction example of hydrothermal management characteristics in a fuel cell; wherein the mesh diagram, simulation error diagram, prediction results using computational fluid dynamics methods, and prediction results using the software architecture of the embodiment are shown below. Figure 7 As shown; in this Example 3, the mesh used is an 11-region hexahedral structured mesh with a total of approximately 500,000 meshes. The CFD simulation solver used is FLUENT, and the original data format is FLUENT. The mode decomposition algorithm selected is the intrinsic orthogonal decomposition algorithm; the basis function coefficients are calculated using a machine learning method; the output format of the physics field prediction results is TECPLOT format; from the appendix... Figure 7 As can be seen from the example, the software architecture described in the embodiment, which uses a reduced-order model to perform digital twins, obtains temperature fields, membrane water content fields, and liquid water saturation fields with smaller errors compared to the results of standard CFD methods. For 10 random operating conditions, the maximum error is only 3%. The software architecture described in the embodiment can obtain multiphysics fields within 1 second, while the standard CFD method requires 1 to 2 days.

[0058] The physical field prediction software architecture and prediction method based on modal decomposition described in this embodiment introduces and constructs an unstructured mesh data structure. This data structure includes the topological relationships between different elements in the space, ensuring the effectiveness of the architecture when applied to general problems in practical engineering. Relying on the unstructured mesh data structure, it can provide more comprehensive modal decomposition methods and methods for calculating weight coefficients. Furthermore, since classical modal decomposition techniques are highly accurate but lack versatility for coupled problems, this embodiment also introduces highly versatile machine learning methods and combines classical modal decomposition techniques with machine learning methods to ensure that users can freely choose the solution method according to their needs for different application scenarios.

[0059] The physical field prediction software architecture described in this invention has a clear structure, is easy to use, and has high versatility. It can support multiple mainstream numerical data formats, provide a choice of multiple mode decomposition methods and weight coefficient calculation methods, and improves versatility and calculation accuracy.

[0060] The above embodiments are merely one of the implementation methods for achieving the technical solution of the present invention. The scope of protection claimed by the present invention is not limited to this embodiment, but also includes any variations, substitutions and other implementation methods that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention.

Claims

1. A physical field prediction software architecture based on mode decomposition, characterized in that, Includes a preprocessing module and a prediction module; The preprocessing module is used to read raw data according to a preset data reading method and integrate the read data into a snapshot matrix; and to perform mode decomposition on the snapshot matrix according to a predetermined mode decomposition algorithm to obtain an expert evaluation database; wherein, the expert evaluation database includes several basis functions; The prediction module is used to obtain basis function coefficients according to the defined basis function coefficient calculation method; and to obtain the physical field prediction result based on the basis function coefficients and the expert evaluation database.

2. The physical field prediction software architecture based on mode decomposition according to claim 1, characterized in that, The preprocessing module includes a first enumeration variable module, a data reading module, a snapshot storage module, and a mode decomposition module; The first enumeration variable module is used to determine the data format of the original data; The data reading module is used to read the original data according to a preset data reading method based on the data format discrimination result of the original data, and integrate the read data information into a snapshot matrix; The snapshot storage module is used to store the snapshot matrix; The mode decomposition module is used to perform mode decomposition on the snapshot matrix according to a pre-written mode decomposition algorithm to obtain an expert evaluation database.

3. The physical field prediction software architecture based on mode decomposition according to claim 2, characterized in that, The first enumeration variable module uses the enumeration variable ResultDataType, the data reading module is a class CASE, the snapshot storage module is a two-dimensional container variable SnapshotMatrix, and the modality decomposition module is a class DECOMPOSE.

4. The physical field prediction software architecture based on mode decomposition according to claim 2, characterized in that, The pre-written mode decomposition algorithm includes singular value decomposition algorithm, fast singular value decomposition algorithm, intrinsic orthogonal decomposition algorithm, or dynamic mode decomposition algorithm.

5. The physical field prediction software architecture based on mode decomposition according to claim 1, characterized in that, The prediction module includes a second enumeration variable module, a third enumeration variable module, a basis function coefficient calculation module, a linear combination module, and a prediction result output module; The second enumerated variable module is used to define the calculation method for the basis function coefficients; The third enumeration variable module is used to define the file output format of the physical field prediction results; The basis function coefficient calculation module is used to calculate the basis function coefficients according to the defined basis function coefficient calculation method and in conjunction with the expert evaluation database. The linear combination module is used to linearly combine the basis functions in the expert evaluation database with the basis function coefficients according to the defined class function for constructing the physical field, so as to obtain the predicted physical field. The prediction result output module is used to output the predicted physical field according to the defined physical field prediction result file output format based on the defined class function for outputting physical field information, so as to obtain the physical field prediction result.

6. The physical field prediction software architecture based on mode decomposition according to claim 5, characterized in that, The second enumerated variable module is the enumerated variable PredictType, and the third enumerated variable module is the enumerated variable OutputFileType; the basis function coefficient calculation module, the linear combination module, and the prediction result output module are all of the PREDICT class.

7. The physical field prediction software architecture based on mode decomposition according to claim 5, characterized in that, The prediction module also includes a file directory specification module; The file target specification module is used to specify the file directory of the expert evaluation database and the file directory for storing the physical field prediction results to the basis function coefficient calculation module; wherein, the file directory specification module is a string variable module.

8. The physical field prediction software architecture based on mode decomposition according to claim 5, characterized in that, The methods for calculating the coefficients of the defined basis functions include Galerkin projection, interpolation, or machine learning.

9. The physical field prediction software architecture based on mode decomposition according to claim 5, characterized in that, The defined output formats for physical field prediction results include Tecplot, VTK, Fluent, CFX, or MHT formats.

10. A physical field prediction method based on mode decomposition, characterized in that, The physical field prediction software architecture based on mode decomposition as described in any one of claims 1-9 is utilized; Physical field prediction methods based on mode decomposition include: The original data is read according to a preset data reading method, and the read data is integrated into a snapshot matrix; the snapshot matrix is ​​decomposed according to a predetermined mode decomposition algorithm to obtain an expert evaluation database; wherein, the expert evaluation database includes several basis functions; The basis function coefficients are obtained according to the defined method for calculating basis function coefficients; the physical field prediction results are obtained based on the basis function coefficients and the expert evaluation database.

Images