Digital twinborn simulation prediction method for transient operation process of nuclear reactor

Abstract

The invention belongs to the technical field of nuclear reactor simulation, and particularly relates to a digital twinborn simulation prediction method for a transient operation process of a nuclear reactor. Comprising the following steps: step 1, establishing a reactor three-dimensional model based on an OpenMC program; step 2, carrying out post-processing on an OpenMC calculation result; 3, carrying out reactor core-heat-force coupling simulation on the basis of a multi-working-condition reaction section set provided by a Monte Carlo program; 4, dynamically modifying the parameter settings of the model and executing program operation; 5, integrating and outputting related multi-physics field parameters to form a multi-working-condition reactor operation state database; 6, decomposing the data matrix into a product of three matrixes; and 7, obtaining the physical field distribution of the working condition needing to be predicted. The method has the advantages that the reactor multi-physical-field multi-working-condition operation state database is successfully constructed, numerical simulation of reactor fuel deformation is added, and the database structure is further improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of nuclear reactor simulation, and particularly relates to a digital twin simulation prediction method for the transient operation process of a nuclear reactor. Background Art

[0002] In a reactor, the core interior is in a harsh operating environment such as high temperature, high pressure, and strong irradiation. To know the state of the physical field inside the core, only conventional measurement methods can be used, including setting thermocouples at the inlet and outlet of the core to measure the temperature of the coolant flowing into and out of the core, or inserting a neutron fluence rate measurement conduit from the bottom of the core to measure the neutron flux at a certain point in the core, etc. However, the measurement effects of such measurement methods are relatively limited, and it is impossible to accurately know the physical field state inside the reactor, especially the deformation behavior of the core fuel. Therefore, it is necessary to carry out numerical simulation of the reactor core based on these measurable parameters. However, since the operation process of the reactor involves various complex physical phenomena, traditional single physical field analysis methods cannot accurately capture the complexity of these interactions and cannot comprehensively provide a safe and effective assessment of the reactor operation. Therefore, relying on traditional single physical field analysis methods can no longer fully meet the high standards required for the safe, stable, and efficient operation of modern nuclear power plants. Therefore, it is urgent to develop multi-physical field coupling analysis technology to achieve a more accurate and comprehensive analysis of the reactor.

[0003] The existing simulation methods are as follows:

[0004] A divide-and-conquer nuclear reactor cross-scale digital twin POD modeling method (application number: 202410557662.2). This method is based on a system-level one-dimensional numerical model to carry out thermal-hydraulic simulation to obtain parameters such as inlet temperature, pressure, and flow rate (these parameters are the inlet parameters calculated by the one-dimensional program); based on the calculation results of the one-dimensional program, a multi-condition physical thermal-hydraulic calculation model (nuclear-thermal coupling calculation) is established, and the physical field state distribution is decomposed by POD to obtain modes and mode coefficients; the least-squares interpolation method is used to obtain new mode coefficients (finally, the least-squares difference method is used for the modal fitting calculation of the mode coefficients). Ultimately, the distribution of the physical field can be obtained.

[0005] A nuclear reactor digital twin key parameter autonomous optimization data inversion method (application number: 202211299608.X). This method only focuses on the digital twin process and does not mention how the data is obtained. It only preprocesses the data to be digitally twinned, and focuses on the training effects of different neural network algorithms and the reconstruction accuracy of the physical field.

[0006] Digital Twin Optimization Method for Nuclear Reactor Core Based on Data Fusion (Application No.: 202310869958.3) performs nuclear-thermal coupling calculation on the reactor based on in-core detector measurement data and other initial conditions; subsequently, the physical model is optimized according to the real-time detection data of sensors.

[0007] Research on Digital Twin of Reactor Physics Operation Based on Physics-Guided and Data-Augmented (Article No.: 0258-0926(2021)S2-0048-06; doi:10.13832 / j.jnpe.2021.S2.0048) realizes model order reduction based on the POD method and also uses the KNN method to calculate modal coefficients, but the simulation software used is only the core physics calculation software, and the distribution of the physical field is not seen in the article.

[0008] Research on Model Order Reduction and Ultra-Real-Time Prediction Technology for Digital Twin of Lead-Cooled Fast Reactor (Article No.: 2023:9. DOI: 10.26914 / c.cnkihy.2023.103828.) Conference paper of the 2023 Annual Conference of the Chinese Nuclear Society. The coupling model used is two-dimensional and only nuclear-thermal coupling, without mechanics-related content. Summary of the Invention

[0009] The object of the present invention is to provide a digital twin simulation prediction method for the transient operation process of a nuclear reactor, which can quickly and accurately calculate and predict the distribution of multi-physical fields in the reactor based on the operation state database accumulated by the multi-field coupling simulation of the reactor core.

[0010] The technical solution of the present invention is as follows: A digital twin simulation prediction method for the transient operation process of a nuclear reactor includes the following steps:

[0011] Step 1: Establish a three-dimensional model of the reactor based on the OpenMC program, and generate the key input files required for simulation in Python language through the Python API, including material definition files, geometric structure files, and simulation setting files;

[0012] Step 2: After the simulation is completed, post-process the calculation results of OpenMC;

[0013] Step 3: Based on the multi-condition reaction cross-section set provided by the Monte Carlo program, perform core nuclear-thermal-mechanical coupling simulation;

[0014] Step 4: Implement the automated operation of the GeN-Foam coupling program through Python code. For the input parameters of different powers, coolant flow rates, and coolant inlet temperatures, dynamically modify the above parameter settings of the model and execute the program operation, so as to complete the calculation of the core nuclear-thermal-mechanical coupling multi-condition simulation;

[0015] Step 5: Post-process based on the multi-condition numerical simulation results, integrate and output relevant multi-physical field parameters to form a multi-condition reactor operation state database. This database contains the distribution parameters of the neutron flux density field, coolant temperature field, and fuel deformation field in the core, and consists of a training set and a test set;

[0016] Step 6: Based on the reactor operation state database obtained from the multi-physical field coupling numerical simulation, use singular value decomposition in the POD method to decompose the data matrix into the product of three matrices:

[0017] X = UΣV T (2)

[0018] In the formula, U is the left singular vector matrix representing the POD mode of the physical field; Σ is the singular value matrix representing the energy corresponding to the physical field mode; V T is the right singular vector matrix representing the measurable parameters of the core under the corresponding physical field;

[0019] Step 7: Based on the measured parameters of the relevant sensors of the reactor, combined with the previously trained fast prediction model of the reactor digital twin, realize the real-time reconstruction prediction of the steady-state and transient physical fields of the reactor, and obtain the physical field distribution of the required prediction conditions.

[0020] In the described Step 1, the material definition file defines the materials required for the simulation and their constituent nuclides; the geometric structure file describes the geometric structure of the model, including the spatial position distribution of fuel assemblies with different enrichments, control rod assemblies, reflector assemblies, and the active zones, helium layers, axial reflectors, etc. of different assemblies inside the core, and at the same time specifies the behavior of particles when moving to the boundary; the simulation settings file sets the operation mode of the simulation, the number of particles participating in the simulation, batches, and the specific parameters of the particle source.

[0021] In the described Step 2, according to the order of different interpolation tables of different components in the nuclear-thermal-mechanical coupling program, organize the core cross-section parameters, and standardize the cross-section units to generate a multi-condition reaction cross-section set of the core to meet the input requirements of subsequent nuclear-thermal-mechanical coupling calculations. Normalize the neutron flux density of the core according to Equation (1) to obtain the neutron flux density distribution inside the core

[0022]

[0023] In the formula, P is the reactor power, φ is the total flux, H is the total nuclear heat, and V is the total volume of the reactor.

[0024] In step 3, the GeN-Foam program is used to establish a reactor geometry model consistent with the Monte Carlo program, and relevant thermal-hydraulic and mechanical parameters are set, including thermal-hydraulic parameters such as coolant density and dynamic viscosity, as well as Young's modulus and linear expansion coefficient. The reaction cross-section sets of each component generated above are used as key physical parameters, the boundary conditions of the core are set, and the program operation mode is clarified, including time step and number of iterations. After completing the core neutronics-thermal-hydraulics-mechanics coupled numerical simulation, the simulation results are compared and verified with the previously calculated core neutron flux density distribution to evaluate the accuracy and consistency of the model.

[0025] In step 5, the database is stored in the form of a multi-dimensional matrix, where each column represents a specific operating condition, covering the operating conditions of the core at different powers, coolant inlet temperatures, and flow rates, and can comprehensively characterize the core operating state, providing a solid data basis for the construction and verification of the subsequent reactor digital twin model.

[0026] In step 6, based on the training set, the physical field modes, their corresponding energies, and modal coefficients under each condition are extracted by singular value decomposition to construct a reduced-order model, and a reconstruction test is carried out through the test conditions of the training set. The number of modes participating in the reconstruction is adjusted according to the error of the reconstructed structure to determine the optimal reconstruction modes. Using the test set, the fitting relationship between the input parameters and the modal coefficients is verified and optimized through the KNN algorithm, and the physical field is reconstructed through a certain test condition of the test set. The K-nearest neighbor parameters used for reconstruction are continuously adjusted to determine the optimal K value, and a KNN regression model of the modal coefficients is established. Finally, the reduced-order model and the regression model are combined, and the LSTM neural network is introduced for training to construct a fast prediction model of the reactor digital twin. Subsequently, a certain test condition is selected, and based on the input reactor power, coolant inlet temperature, and flow rate, a fast prediction of the corresponding physical field reconstruction is carried out through the trained fast prediction model of the reactor digital twin, and an R 2 goodness-of-fit analysis is carried out according to Equation (3) to evaluate the degree of agreement between the predicted values and the true values of the model, and at the same time, the original numerical simulation results, the comparison chart of the prediction results, and the prediction error are given to evaluate the prediction performance of the prediction model;

[0027]

[0028] In the formula, represents the true numerical simulation result of the j-th node sample, represents the fast prediction result of the digital twin of the j-th node sample, represents the average value of all true results, and each condition of each physical field contains N nodes.

[0029] The beneficial effects of the present invention are as follows: (1) Three-dimensional nuclear-thermal-mechanical coupling simulations are carried out, and a database of the operating states of multiple physical fields of the reactor under multiple conditions is successfully constructed. In addition to the conventional nuclear-thermal coupling simulations, numerical simulations of the deformation of reactor fuel are added, further improving the database structure. (2) Based on the database of operating states under multiple conditions and model order reduction, on the basis of using the KNN algorithm for predicting the conventional steady-state physical fields, the physical field distribution prediction of the transient conditions of the reactor is carried out using the LSTM neural network based on the measurement parameters of the relevant sensors of the reactor, assisting the operators in judging the state of the reactor core. (3) By combining the three-dimensional nuclear-thermal-mechanical coupling simulation at the full-reactor level and the digital twin technology, the limitations of traditional single simulations are broken through, ensuring both the accuracy of multi-physical field coupling and taking into account the advantages of the digital twin technology in rapid prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 FIG. is a schematic flow chart of a digital twin simulation prediction method for the transient operation process of a nuclear reactor provided by the present invention;

[0031] Figure 2 FIG. is a schematic diagram for calculating cross-sectional parameters;

[0032] Figure 3 FIG. is a schematic diagram of the formation of nuclear-thermal-mechanical coupling simulation and operating state database;

[0033] Figure 4 FIG. is a schematic diagram of digital twin physical field reconstruction;

[0034] Figure 5 FIG. is a schematic diagram of the fuel deformation field of multi-physical field coupling under a certain condition;

[0035] Figure 6 FIG. is a schematic diagram of the digital twin fuel deformation field under a certain condition;

[0036] Figure 7 FIG. is a schematic diagram of the error analysis of the digital twin neutron flux density field under a certain condition;

[0037] Figure 8 FIG. is a schematic diagram of the error distribution of the digital twin fuel deformation field under a certain condition. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0038] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0039] A digital twin simulation prediction method for the transient operation process of a nuclear reactor makes full use of simulation technologies such as physical models and sensors, and can perform multi-physical field coupling simulations through model reduction techniques and machine learning algorithms. It completes the dimension reduction of the parameters in the high-dimensional operation state database of the reactor and the calculation of key modal coefficients, realizing the rapid and accurate prediction of the multi-physical field distribution of the reactor. Compared with traditional core multi-physical field coupling calculation software, it significantly reduces the solution time, assists nuclear power plant operators in judging the state of the core physical field, and ensures the safe operation of the reactor.

[0040] By combining the reactor multi-physical field coupling simulation technology with the digital twin rapid prediction technology, the present invention proposes an efficient and accurate reactor operation state prediction method. First, the multi-physical field numerical simulation technology under multiple working conditions is used to conduct a detailed analysis of the operation process of the reactor, and a running state database containing the multi-physical field distribution of the reactor under different working conditions and operating conditions is accumulated. Based on this database, the present invention further introduces the digital twin technology, simplifies the complex multi-physical field model through the model reduction method, and greatly reduces the computational complexity on the premise of ensuring accuracy. Finally, the rapid and accurate prediction of the multi-physical field distribution of the reactor is realized, significantly reducing the calculation time, and at the same time improving the efficiency and accuracy of the reactor operation state evaluation.

[0041] By combining the reactor multi-physical field coupling simulation technology with the digital twin rapid prediction technology, the present invention proposes a method that combines the nuclear-thermal-mechanical coupling transient simulation of the reactor and the digital twin rapid prediction technology. First, based on the measurement parameters of reactor-related sensors such as coolant flow rate sensors, temperature sensors, and neutron flux density sensors and other reactor operation parameters, nuclear-thermal-mechanical coupling transient simulations are carried out to accumulate multi-physical field parameters under multiple working conditions of the reactor and form a running state database. Subsequently, based on this database, the present invention further introduces the digital twin technology, and simplifies the dimension of the complex multi-physical field running state database through the POD (Proper Orthogonal Decomposition) model reduction method, and greatly reduces the computational complexity on the premise of ensuring accuracy. Finally, based on the reduced-order model, the KNN (K-Nearest Neighbors) algorithm and the LSTM (Long Short-Term Memory) neural network are used to realize the rapid and accurate prediction of the steady-state and transient multi-physical field distribution of the reactor, significantly reducing the calculation time, improving the efficiency and accuracy of the reactor operation state evaluation, and being applicable to multiple fields such as reactor design optimization, operation state monitoring, and prediction.

[0042] As Figure 1 shown, a digital twin simulation prediction method for the transient operation process of a nuclear reactor includes the following steps:

[0043] Step 1: As Figure 2 shown, first, establish a three-dimensional model of the reactor based on the OpenMC program, and generate the key input files required for the simulation in Python language through the Python API, including the material definition file (materials.xml), the geometry file (geometry.xml), and the simulation settings file (settings.xml). Among them, the material definition file defines the materials required for the simulation (such as UN fuel, B 4 C control rods, etc.) and their constituent nuclides; the geometry file describes the geometric structure of the model, including the spatial position distribution of fuel assemblies with different enrichments, control rod assemblies, reflector assemblies inside the core, and the active zones, helium layers, axial reflectors, etc. of different components, and at the same time specifies the behavior of particles when moving to the boundary; the simulation settings file sets the operating mode of the simulation, the number of particles participating in the simulation, the batch, and the specific parameters of the particle source. On this basis, carry out the Monte Carlo neutron transport numerical simulation of the core, calculate the key physical parameters such as the particle motion speed, neutron fission cross-section, neutron effective fission cross-section, neutron absorption cross-section, scattering cross-section, and diffusion coefficient, and consider the influencing factors under multiple working conditions, including different temperatures, different expansion degrees of components, and different densities of coolants, etc., to comprehensively evaluate the neutron physical characteristics of the core.

[0044] Step 2: As Figure 2 shown, after the simulation is completed, post-process the OpenMC calculation results. First, according to the order of different interpolation tables of different components in the nuclear-thermal-mechanical coupling program, sort out the core cross-section parameters, and standardize the cross-section units to generate a set of core multi-condition reaction cross-sections to meet the input requirements of subsequent nuclear-thermal-mechanical coupling calculations.

[0045] At the same time, normalize the core neutron flux density according to Equation (1) to obtain the neutron flux density distribution inside the core. This result provides an important reference for the verification of the neutron flux density in the reactor nuclear-thermal-mechanical coupling program, ensuring the accuracy and consistency of subsequent multi-physics field coupling simulations.

[0046]

[0047] In the formula, P is the reactor power, φ is the total flux, H is the total nuclear heat, and V is the total volume of the reactor.

[0048] Step 3: As Figure 3As shown, based on the multi - condition reaction cross - section set provided by the above - mentioned Monte Carlo program, core nuclear - thermal - hydraulic coupling simulation is carried out. First, the GeN - Foam program is used to establish a reactor geometry model consistent with the Monte Carlo program, and relevant thermal - hydraulic and mechanical parameters are set, including thermal - hydraulic parameters such as coolant density and dynamic viscosity, and mechanical parameters such as Young's modulus and linear expansion coefficient. At the same time, the reaction cross - section sets of each component generated previously are used as key physical parameters to set the boundary conditions of the core, such as the albedo of neutrons at the core boundary and the temperature of the coolant when it flows in from the bottom. In addition, the program operation mode is clarified, including calculation control parameters such as time step and number of iterations. After completing the core nuclear - thermal - hydraulic coupling numerical simulation, the simulation results are compared and verified with the previously calculated core neutron flux density distribution to evaluate the accuracy and consistency of the model and ensure the reliability of the numerical simulation.

[0049] Step 4: As Figure 3 shown, the automation of the GeN - Foam coupling program is realized through Python code. For different input parameters such as power, coolant flow rate, and coolant inlet temperature (all relevant parameters can be measured in the reactor by sensors), the above - mentioned parameter settings of the model are dynamically modified and the program is executed to complete the efficient calculation of the core nuclear - thermal - hydraulic coupling multi - condition simulation.

[0050] Step 5: Based on the multi - condition numerical simulation results as Figure 5 shown, post - processing is carried out to integrate and output relevant multi - physical - field parameters to form a multi - condition reactor operation state database. This database contains distribution parameters of key physical fields such as the neutron flux density field, coolant temperature field, and fuel deformation field of the core, and is composed of a training set and a test set. The database is stored in the form of a multi - dimensional matrix, where each column represents a specific operation condition, covering the operation conditions of the core at different powers, coolant inlet temperatures, and flow rates, and can comprehensively characterize the core operation state, providing a solid data basis for the construction and verification of the subsequent reactor digital twin model. The parameter variation ranges of the training set are shown in Table 1, and the parameter variation ranges of the test set are shown in Table 2.

[0051] Table 1 Setting of the parameter ranges of the training set working conditions

[0052]

[0053] Table 2 Setting of the parameter ranges of the test set working conditions

[0054]

[0055] Step 6: As Figure 4 shown, based on the reactor operation state database obtained from the multi - physical - field coupling numerical simulation, the data matrix is decomposed into the product of three matrices by singular value decomposition (SVD) in the POD method:

[0056] X = UΣV T (2)

[0057] where U is the left singular vector matrix representing the POD modes of the physical field; Σ is the singular value matrix representing the energy corresponding to the physical field modes; V T is the right singular vector matrix representing the measurable parameters of the reactor core under the corresponding physical field.

[0058] Based on the training set, the physical field modes, their corresponding energies, and modal coefficients under each working condition are extracted through singular value decomposition, a reduced-order model is constructed, and a reconstruction test is carried out through the test working conditions of the training set. The number of modes participating in the reconstruction is continuously adjusted according to the error of the reconstruction structure to determine the optimal reconstruction modes. Using the test set, the fitting relationship between the input parameters and the modal coefficients is verified and optimized through the KNN algorithm, and the physical field reconstruction is carried out through the test working conditions of the test set. The K-nearest neighbor parameters for reconstruction are continuously adjusted to determine the optimal K value, and a KNN regression model of the modal coefficients is established. Finally, the reduced-order model and the regression model are combined, and an LSTM neural network is introduced for training to construct a fast prediction model for the reactor digital twin. Subsequently, select a certain test working condition, based on the input reactor power, coolant inlet temperature, and flow rate, carry out a rapid prediction of the corresponding physical field reconstruction through the trained fast prediction model of the reactor digital twin, and perform an R 2 goodness-of-fit analysis based on Equation (3) to evaluate the degree of agreement between the predicted values and the true values of the model. At the same time, the original numerical simulation results, comparison charts of the fast prediction results of the digital twin, and prediction errors are given to evaluate the prediction performance of the prediction model.

[0059]

[0060] where represents the true numerical simulation result of the j-th node sample, represents the fast prediction result of the digital twin of the j-th node sample, represents the average value of all true results, and each working condition of each physical field contains N nodes.

[0061] Step 7: As Figure 4 shown, finally, based on the measured parameters of the reactor-related sensors, combined with the previously obtained fast prediction model of the reactor digital twin, real-time reconstruction prediction of the steady-state and transient physical fields of the reactor is realized, and the physical field distribution of the required prediction working condition is obtained as Figure 6 shown, significantly reducing the calculation time, saving computing resources, and realizing fast prediction of the reactor physical field.

[0062] Example 1:

[0063] As Figure 1As shown, this embodiment relates to a digital twin simulation prediction system for the transient operation process of a nuclear reactor. As Figure 2 shown, a three-dimensional model of a 50MW th integral small lead-cooled fast reactor is established based on the OpenMC program, and relevant parameters and boundary conditions are set. Specifically, it includes the construction of the material composition and geometric models of the in-core fuel assembly, out-core fuel assembly, control rod assembly, and reflector assembly. A total of 200 cycles are set in the simulation process. Each single cycle includes 300,000 particles, and the first 50 cycles with relatively large errors are ignored. After the simulation is completed, the cross-section sets of the reactor fuel assembly, control rod assembly, reflector assembly, and each axial part of all components under different working conditions are output as the input parameters of the GeN-Foam program.

[0064] As Figure 3 shown, based on the cross-section sets of each component of the reactor output in the previous step, a geometric model identical to the previous program is established using the GeN-Foam program. Relevant thermal-hydraulic parameters, structural mechanics parameters, neutron physics parameters (cross-section sets output by the OpenMC program), boundary conditions, and the operating mode of the program are set for the full-core nuclear-thermal-mechanical multi-physical field coupling numerical simulation, including different reactor powers, coolant inlet temperatures, and coolant flow rates. After the simulation is completed, relevant multi-physical field parameters are output, and the neutron flux density field is verified. It is found that the maximum error is only 3.77%, meeting the requirements. Subsequently, the multi-condition parameters are post-processed and integrated into a training set and a test set to form a reactor multi-condition operation state database.

[0065] As Figure 4 shown, based on the training set data, the POD method is used to reduce the dimension of a large amount of high-dimensional physical field data in the multi-condition database, calculate the various order modes and mode coefficients of the corresponding physical field to construct a reduced-order model. Based on the test set, the K-nearest neighbor parameters are verified and adjusted to construct a regression model for predicting the working conditions. Combining the reduced-order model and the regression model, the final reactor digital twin rapid prediction model is trained through the LSTM neural network.

[0066] Based on a certain test condition, the maximum errors of the numerical simulation results and prediction results of multiple physical fields are given. Combining the R 2 goodness of fit to evaluate the prediction performance of the model, it is found that: in the test conditions of the training set, by determining the optimal reconstruction mode and initial K-nearest neighbor parameters, the coolant temperature field, core neutron flux density field, and fuel deformation field are reconstructed. The results show that the maximum errors are 0.56%, 0.0136%, and 0.4% respectively, and the R 2 goodness of fit is 0.994185, 0.999756, and 0.999986 respectively. The errors are all less than 1%, proving that the digital twin model can accurately reproduce the real-time multi-physical field distribution inside the core that cannot be measured.

[0067] In the test set verification, the optimal K-nearest neighbor parameters were adjusted to predict and reconstruct the multi-physical fields of the reactor core. It was found that the maximum errors of the coolant temperature field, the core neutron flux density field, and the fuel deformation field were 0.24%, 0.0104%, and 0.20% respectively, and the R 2 goodness-of-fit values were 0.999129, 0.999859, and 0.999997 respectively, and the errors were all less than 0.5%. The error analysis of the neutron flux density field is as follows Figure 7 shown. The results show that the digital twin rapid prediction model still exhibits a high-precision prediction ability when facing working conditions within the training range but not trained.

[0068] Finally, when predicting new working conditions, after inputting the core power, coolant flow rate, and coolant inlet temperature of a required prediction working condition measured by relevant sensors, the physical field prediction and reconstruction are realized through the above algorithm, as shown in Figure 8 the predicted error distribution map of the fuel deformation field of this prediction working condition is obtained. It is found that the maximum error is only 1.05%, and the prediction time is only 1.04 s, significantly reducing the calculation time, and successfully realizing the rapid and accurate prediction of the fuel deformation field under this working condition.

[0069] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: modifying the technical solutions recorded in the foregoing embodiments, or equivalently replacing some or all of the technical features therein, does not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A digital twin simulation prediction method for the transient operation process of a nuclear reactor, characterized in that: The following steps are involved: Step 1: Build a 3D reactor model based on the OpenMC program, and generate key input files required for simulation in Python language through the Python API, including material definition files, geometry files, and simulation setting files; Step 2: After the simulation is completed, post-process the OpenMC calculation results; Step 3: Based on the multi-condition reaction cross-section set provided by the Monte Carlo program, perform core nuclear-thermal-mechanical coupling simulation; Step 4: The Python code is used to automate the operation of the GeN-Foam coupling program. According to the input parameters of different powers, coolant flow rates, and coolant inlet temperatures, the above parameter settings of the model are dynamically modified and the program is executed to complete the calculation of the core nuclear-thermal-mechanical coupling multi-condition simulation. Step 5: Post-process the multi-condition numerical simulation results, integrate and output the relevant multi-physical field parameters to form a multi-condition reactor operation status database, which contains the distribution parameters of the core neutron flux density field, coolant temperature field and fuel deformation field, and consists of a training set and a test set; Step 6: Based on the reactor operation status database obtained from the multi-physics field coupling numerical simulation, the singular value decomposition in the POD method is used to decompose the data matrix into the product of three matrices: X=UΣV T (2) Where U is the left singular vector matrix representing the POD mode of the physical field; Σ is the singular value matrix representing the energy corresponding to the physical field mode; V T is the right singular vector matrix representing the measurable parameters of the core under the corresponding physical field; Step 7: Based on the measured parameters of the reactor-related sensors and combined with the previously obtained reactor digital twin rapid prediction model, real-time reconstruction and prediction of the reactor's steady-state and transient physical fields are achieved to obtain the physical field distribution of the required predicted operating conditions.

2. A digital twin simulation prediction method for a nuclear reactor transient operation process according to claim 1, characterized in that: The material definition file in step 1 defines the materials and their constituent nuclides required for the simulation; the geometry file describes the geometry of the model, including the spatial distribution of fuel assemblies with different enrichments, control rod assemblies, reflector layer assemblies, and active areas, helium layers, axial reflector layers, etc. of different assemblies inside the core, and specifies the behavior of particles when they move to the boundary; the simulation setting file sets the simulation operation mode, the number of particles involved in the simulation, batches, and specific parameters of the particle source.

3. The digital twin simulation prediction method for the transient operation process of a nuclear reactor according to claim 1, characterized in that: The step 2 is to sort out the core cross-sectional parameters according to the order of different interpolation tables of different components in the nuclear-thermal-mechanical coupling program, and to standardize the cross-sectional units to generate a core multi-condition reaction cross-sectional set to meet the input requirements of subsequent nuclear-thermal-mechanical coupling calculations. The core neutron flux density is normalized according to formula (1) to obtain the neutron flux density distribution inside the core. Where P is the reactor power, φ is the total flux, H is the total nuclear heat, and V is the total reactor volume.

4. The digital twin simulation prediction method for the transient operation process of a nuclear reactor according to claim 1, characterized in that: The step 3 uses the GeN-Foam program to establish a reactor geometry model consistent with the Monte Carlo program, and sets relevant thermal and mechanical parameters, including thermal parameters such as coolant density and dynamic viscosity, as well as Young's modulus and linear expansion coefficient. The reaction cross-section set of each component generated above is used as the key physical parameter, the boundary conditions of the core are set, and the program operation mode, including the time step and the number of iterations, is clarified. After completing the core nuclear-thermal-mechanical coupling numerical simulation, the simulation results are compared and verified with the core neutron flux density distribution calculated previously to evaluate the accuracy and consistency of the model.

5. The digital twin simulation prediction method for the transient operation process of a nuclear reactor according to claim 1, characterized in that: The database in step 5 is stored in the form of a multi-dimensional matrix, in which each column represents a specific operating condition, covering the operating conditions of the core at different powers, coolant inlet temperatures and flow rates. It can comprehensively characterize the operating status of the core and provide a solid data foundation for the subsequent construction and verification of the reactor digital twin model.

6. The digital twin simulation prediction method for the transient operation process of a nuclear reactor according to claim 1, characterized in that: In the step 6, based on the training set, the physical field modes and their corresponding energies and modal coefficients under each working condition are extracted by singular value decomposition, a reduced-order model is constructed, and a reconstruction test is performed through the training set test condition. The number of modes involved in the reconstruction is adjusted according to the error of the reconstructed structure to determine the optimal reconstructed mode; using the test set, the fitting relationship between the input parameters and the modal coefficients is verified and optimized by the KNN algorithm, and the physical field is reconstructed through a test set test condition, and the K nearest neighbor parameters used for reconstruction are continuously adjusted to determine the optimal K value, and a KNN regression model of the modal coefficient is established. The reduced-order model is combined with the regression model, and an LSTM neural network is introduced for training to construct a reactor digital twin rapid prediction model. Subsequently, a certain test condition is selected, and based on the input reactor power, coolant inlet temperature and flow rate, the reconstruction of the corresponding physical field is rapidly predicted through the trained reactor digital twin rapid prediction model, and the model is R-tested based on formula (3). 2 Goodness of fit analysis evaluates the degree of agreement between the model prediction value and the true value, and provides the original numerical simulation results, the comparison chart of the prediction results and the prediction error to evaluate the prediction performance of the prediction model; In the formula, represents the actual numerical simulation result of the j-th node sample, represents the digital twin quick prediction result of the j-th node sample, Represents the average value of all real results, and each load case of each physical field contains a total of N nodes.

Images