A method and apparatus for calibrating parameters of a CFD turbulence model of a nuclear reactor rod beam field.

By combining Latin hypercube sampling and ensemble Kalman filtering algorithms, efficient calibration of turbulence model parameters for nuclear reactor rod bundle flow fields was achieved. This solved the problems of large prediction deviations and high computational costs across operating conditions in existing technologies, and improved the accuracy and engineering reliability of flow field prediction.

CN123069974APending Publication Date: 2026-11-10HARBIN ENG UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611333858.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-31
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

Existing technologies for calibrating CFD turbulence model parameters in nuclear reactor rod beam flow fields suffer from problems such as large prediction deviations across operating conditions, high computational costs, and a lack of unified standards for manual parameter tuning, making it difficult to simultaneously address multiple flow characteristics and improve simulation accuracy.

Method used

Multiple sets of turbulence parameter sample sets are generated using the Latin hypercube sampling method. Combined with the ensemble Kalman filter algorithm and one-dimensional linear interpolation technology, the turbulence parameters are efficiently calibrated by error propagation and correction between CFD simulation and LDA experimental data.

Benefits of technology

It improves the calibration accuracy and generalization ability of turbulence model parameters, reduces computational overhead, and enhances the accuracy and engineering reliability of flow field prediction.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123069974A_ABST
    Figure CN123069974A_ABST
Patent Text Reader

Abstract

This invention discloses a method and apparatus for calibrating parameters of a CFD turbulence model for a nuclear reactor fuel rod bundle flow field, belonging to the field of thermal hydraulics and CFD simulation technology. The method includes: establishing a three-dimensional geometric model and computational grid for the fuel rod bundle with a stirring fin; selecting the SST k-ω turbulence model and determining the parameter vector to be assimilated; generating multiple sets of parameter samples using Latin hypercube sampling; batch-calling the CFD model to perform parallel steady-state calculations; extracting flow field data and aligning it with LDA experimental measurement points using linear interpolation; introducing LDA observation data; calculating the ensemble mean, prediction covariance matrix, and Kalman gain based on ensemble Kalman filtering; performing posterior update of the sample set; and quantifying the prediction bias using the root mean square error cost function; iteratively updating the parameter samples until convergence with the goal of minimizing this cost function; and outputting the optimal combination of turbulence parameters. This invention can effectively calibrate multiple turbulence parameters, improving the prediction accuracy and cross-condition generalization ability of complex flow fields in fuel rod bundles.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of nuclear reactor thermal-hydraulic and CFD simulation technology, specifically to a method and apparatus for calibrating parameters of a CFD turbulence model of a nuclear reactor rod beam flow field. Background Technology

[0002] Fuel rod bundles are the core structure of pressurized water reactor (PWR) fuel assemblies. The complex turbulent effects, such as lateral flow and vortices induced by the grid mixing wings, directly affect the coolant mixing characteristics and core heat transfer capacity. Accurate prediction of the flow field distribution inside the fuel rod bundle is crucial for reactor thermal-hydraulic design and safety analysis. Computational fluid dynamics (CFD) is an important tool for resolving the refined three-dimensional turbulent flow field of the fuel rod bundle, and the simulation accuracy is highly dependent on the empirical parameters of the selected turbulence model.

[0003] Most existing studies directly use default parameters of turbulence models to calculate the flow field of rod bundles. These default parameters are calibrated for general flow conditions and are poorly adapted to complex conditions such as strong swirling and flow separation within rod bundles with mixing blades. This can easily lead to prediction errors in key parameters such as flow velocity, turbulence intensity, and flow pressure drop. A few turbulence parameter optimization studies rely solely on manual trial and error to adjust parameters, lacking a systematic calibration framework. This makes it difficult to consider multiple flow characteristics and results in insufficient generalization ability of the calibration results. At the same time, traditional parameter optimization methods require a large number of costly CFD iterative calculations, resulting in high computational overhead and limiting the improvement of the accuracy of rod bundle turbulence simulations.

[0004] Therefore, there is an urgent need to develop an efficient and reliable parameter calibration scheme for the CFD turbulence model of the nuclear reactor rod bundle flow field, so as to improve the accuracy of numerical prediction of the complex turbulent flow field of the rod bundle and meet the needs of refined thermal-hydraulic simulation analysis of reactor fuel assemblies. Summary of the Invention

[0005] This specification provides a method and apparatus for calibrating the parameters of a CFD turbulence model of a nuclear reactor rod beam flow field, in order to solve at least one of the technical problems mentioned above.

[0006] To solve the above-mentioned technical problems, the embodiments in this specification are implemented as follows: According to a first aspect of the embodiments of this specification, a method for calibrating parameters of a CFD turbulence model of a nuclear reactor rod beam flow field is provided, comprising the following steps: A three-dimensional geometric model of the fuel rod bundle with a mixing wing was established and the computational mesh was generated. SST k- was selected. Turbulence model, determine the set of multi-dimensional turbulence empirical parameter vectors to be assimilated and their value ranges, set boundary conditions consistent with experimental conditions, unify the dimensionless conversion and physical quantity conversion standards between simulation output and experimental data, and build a CFD numerical simulation model of fuel rod bundle. The parameter space was hierarchically sampled using the Latin hypercube sampling method to generate multiple sets of turbulent parameter samples to be assimilated. The completed CFD simulation model was called in batches to carry out parallel steady-state calculations, obtain the rod bundle flow field simulation results corresponding to each set of parameter samples, and extract the simulation data at the experimental measurement location. The CFD grid data was projected onto the LDA experimental measurement points through one-dimensional linear interpolation to complete the spatial alignment. LDA experimental measurement data of rod bundle flow field is introduced. Based on the ensemble Kalman filter algorithm, the error propagation and correction relationship between CFD simulation results and experimental observation data is established. The ensemble mean, prediction error covariance matrix and Kalman gain are calculated. The posterior update of the parameter sample set is performed. The root mean square error (RMSE) cost function is constructed to quantify the prediction bias. With minimizing the RMSE cost function as the optimization objective, the updated posterior turbulence parameter sample set is re-substituted into the batch CFD simulation process, and the turbulence parameter sample set to be assimilated is updated iteratively. Output the optimal combination of turbulence parameters after iterative optimization convergence to complete the parameter calibration of the fuel rod bundle turbulence model.

[0007] In some optional implementations, the process of establishing a three-dimensional geometric model of the fuel rod bundle with a mixing wing and completing the computational mesh generation, selecting SST k- The steps for building a CFD numerical simulation model of a fuel rod bundle include: determining the set of multi-dimensional turbulence empirical parameter vectors to be assimilated and their value ranges; setting boundary conditions consistent with experimental conditions; and unifying the dimensionless conversion and physical quantity conversion standards between simulation output and experimental data. Construct the geometric computational domain of the rod bundle, and select SST k- under the RANS time-averaged system. The two-equation turbulence model serves as the core prediction model, and its turbulent kinetic energy transport equation and specific dissipation rate transport equation are as follows: , ; In the formula For fluid density, For dynamic viscosity, Where is the eddy viscosity coefficient. The viscosity coefficient of the eddy current is . Let Reynolds stress tensor be the stress tensor. For velocity components, For coordinate components, This represents the value of the strain rate. This is the vorticity value. For model constants, the mixture function The interpolation rule is Used to achieve near-wall k- With far-field k- The model exhibits a smooth transition; its eddy viscosity coefficient is calculated using the following formula: and In the formula The absolute value of the vorticity in the flow field. The second mixing function is defined as follows: The determined parameter vector to be assimilated is... Each parameter is used to solve for turbulent dissipation, eddy viscosity coefficient, and inlet / outlet turbulent boundary conditions.

[0008] In some optional implementations, the steps of standardizing the dimensionless conversion and physical quantity conversion of the unified simulation output and experimental data include: Extract the geometric coordinates along the measurement line and perform dimensionless grid pitch conversion. The coordinate transformation formula is as follows: ; In the formula, Y represents the original spanwise coordinate of the cross section, and Pitch represents the bar strand pitch. This formula is used to align the CFD mesh nodes with the abscissa system of the LDA test measurement points. The definition of the average axial velocity during the LDA test is as follows: ; In the formula The axial velocity during LDA testing, Instantaneous velocity Where N is the sampling time interval and N is the number of sampling points; the root mean square velocity of turbulent fluctuations is derived from turbulent kinetic energy, and the conversion formula is: ; In the formula The root mean square velocity of the turbulent fluctuations. The turbulent kinetic energy is then normalized to the cross-sectional average velocity to obtain the dimensionless physical quantity of the assimilated observation target. as well as In the formula, W is the axial velocity. The average flow velocity across the cross section is given.

[0009] In some optional implementations, the step of using the Latin hypercube sampling method to perform stratified sampling of the parameter space to generate multiple sets of turbulent parameter sample sets to be assimilated includes: The parameter value range in the parameter space to be assimilated is equally divided into N equally probable sub-intervals. For each sample group, a single parameter component is randomly selected from each sub-interval and then randomly paired. Here, N is a preset positive integer used to set different sample size levels. The multi-dimensional parameter vector obtained from each sampling is input into SST k- The turbulence model is used to perform batch steady-state CFD calculations to obtain the flow field state matrix corresponding to all samples in the set.

[0010] In some optional implementations, the step of extracting simulation data at the experimental measurement location and projecting the CFD mesh data onto the LDA experimental measurement points through one-dimensional linear interpolation to complete spatial alignment includes: One-dimensional linear interpolation is performed to align the CFD discrete mesh data with the experimental discrete measurement points. The interpolation is based on the abscissa of the experimental measurement points to obtain the simulation values ​​at the same location, thus eliminating the comparison deviation caused by the non-coincidence of the mesh and the measurement points. The interpolation calculation formula is as follows: ; In the formula For the first Simulated values ​​at each interpolation point The dimensionless abscissa of the test measurement point, , The discrete x-coordinates of adjacent nodes in the CFD mesh. , This represents the flow field value at the corresponding node location.

[0011] In some optional implementations, the steps of introducing LDA experimental measurement data of the rod bundle flow field, establishing the error propagation and correction relationship between CFD simulation results and experimental observation data based on the ensemble Kalman filter algorithm, and calculating the ensemble mean, prediction error covariance matrix, and Kalman gain include: The fundamental equations of the EnKF system are constructed, and the ensemble prediction process expression is as follows: ; In the formula For the first Prior predicted state of the group of samples, For the first The initial parameter set of the group, F(·), is SST k- CFD prediction model Represents the system noise term; the expression for the set state matrix is ​​as follows: ; In the formula For the first The flow field prediction vector corresponding to the group of samples For the first A set of multidimensional turbulent parameter vectors to be assimilated is obtained; the ensemble mean and prediction error covariance matrix are solved. The formula for calculating the ensemble mean is as follows: ; In the formula The average flow field and parameter states for all samples. The total number of samples in the current sample set; the expression for the prediction covariance matrix is ​​as follows: ; In the formula This is the prediction error covariance matrix, used to quantify the correlation between turbulence parameters and observed flow field variables.

[0012] In some optional implementations, the step of performing a posteriori update of the parameter sample set includes: Construct a diagonal form of the observation error covariance matrix. The single-point observation error constraint is as follows: ; In the formula For the first Standard deviation of observations at each measuring point The observation error covariance matrix for the LDA test observations at the measuring point is as follows: ; In the formula, R is the observation error covariance matrix; the Kalman gain matrix is ​​calculated to complete the error weight allocation, and the gain solution formula is as follows: ; In the formula, K is the Kalman gain matrix, used to assign correction weights to the turbulence parameters based on experimental observation data. To observe the operator matrix, Here is the matrix transpose notation; after performing ensemble sample posterior update, the parameter and flow field state correction formulas are as follows: ; In the formula For the updated version Group posterior state To standardize the LDA experiment observation vector, To observe the disturbance noise vector.

[0013] In some optional implementations, after the step of outputting the optimal combination of turbulent parameters after iterative optimization and completing the parameter calibration of the fuel rod bundle turbulence model, the method further includes a generalization capability verification and comprehensive evaluation step, including: The optimal parameter vector obtained from the output is re-imported into the CFD simulation model, and the flow field is recalculated under the first Reynolds number condition involving assimilation and the second Reynolds number condition not involving assimilation. The simulation output data is processed uniformly according to the coordinate dimensionless formula and the linear interpolation formula. Based on the RMSE calculation formula, the prediction errors of axial normalized velocity and turbulence intensity are calculated for the two types of conditions, and the error reduction of the default parameters before assimilation and the optimal parameters after assimilation is compared. The maximum radial and axial pressure drop data not involving assimilation are extracted, and the convergence characteristics of pressure drop values ​​and peak positions under different sample sizes are analyzed. The engineering stability and cross-condition generalization ability of the optimal parameters are comprehensively evaluated.

[0014] In some optional implementations, the step of iteratively updating the set of turbulent parameter samples to be assimilated further includes convergence determination and sample traversal determination, including: Determining whether the iteration has converged specifically includes: during the iteration process, judging whether the change magnitude of the RMSE cost function of the consecutive iteration steps and the average deviation between the simulation results and the experimental data meet the preset convergence threshold requirements. If the convergence requirement is met, the iteration is terminated and the optimal turbulence parameters are obtained; if the convergence requirement is not met, the iteration continues to update the turbulence parameter sample set. Determine whether the entire parameter calibration process has ended, specifically including: determining whether the current sample size has reached all preset levels; If not all samples are traversed, the sample size is changed, and a new complete calibration cycle is executed. If all sample sizes have been assimilated, validated, and statistically analyzed, the overall calibration process ends, and a complete set of calibration parameters, error curves, and engineering sample selection recommendations are output.

[0015] According to a second aspect of the embodiments of this specification, a device for calibrating parameters of a CFD turbulence model of a nuclear reactor rod beam flow field is provided, including: The CFD simulation modeling module is used to establish a three-dimensional geometric model of the fuel rod bundle with a mixing wing and complete the computational mesh generation. The SST k-ω turbulence model is selected, the set of multi-dimensional turbulence empirical parameter vectors to be assimilated and their value ranges are determined, boundary conditions consistent with the experimental conditions are set, the dimensionless conversion and physical quantity conversion standards of simulation output and experimental data are unified, and the CFD numerical simulation model of the fuel rod bundle is built. The parameter sampling and batch calculation module is used to perform hierarchical sampling of the parameter space using the Latin hypercube sampling method, generate multiple sets of turbulent parameter sample sets to be assimilated, batch call the built CFD simulation model to carry out parallel steady-state calculations, obtain the rod bundle flow field simulation results corresponding to each set of parameter samples, extract the simulation data at the experimental measurement location, and project the CFD grid data to the LDA experimental measurement points through one-dimensional linear interpolation to complete spatial alignment. The ensemble Kalman filter assimilation module is used to introduce LDA experimental measurement data of rod bundle flow field. Based on the ensemble Kalman filter algorithm, it establishes the error propagation and correction relationship between CFD simulation results and experimental observation data, calculates the ensemble mean, prediction error covariance matrix and Kalman gain, performs posterior update of parameter sample set, and constructs the root mean square error (RMSE) cost function to quantify prediction bias. The iterative optimization module is used to minimize the RMSE cost function, resubmit the updated posterior turbulence parameter sample set into the batch CFD simulation process, and iteratively update the turbulence parameter sample set to be assimilated in a loop. The calibration and verification output module is used to output the optimal combination of turbulent parameters after iterative optimization convergence, and to complete the parameter calibration of the fuel rod bundle turbulence model.

[0016] According to specific embodiments provided by the present invention, the present invention discloses the following technical effects: 1. In the technical solution of this application, when optimizing parameters in a multi-dimensional turbulence empirical parameter space, Latin hypercube sampling is first used to hierarchically cover the parameter space to generate a sample set, and then CFD simulation models are called in batches to carry out parallel steady-state calculations. Since this hierarchical sampling method can achieve uniform coverage of the parameter space with a small number of samples, it is equivalent to efficiently extracting the response information of different parameter combinations to the predicted results of the rod bundle flow field under limited simulation computing resources. This can reduce the dependence of multi-parameter joint optimization on a large number of CFD trials to a certain extent and improve the computational efficiency of the parameter calibration process.

[0017] 2. In the technical solution of this application, when quantitatively calibrating the model's empirical parameters, experimental measurement data of the rod bundle flow field LDA are introduced as observation constraints for Bayesian optimization. The ensemble mean, prediction error covariance matrix, and Kalman gain are calculated by constructing the EnKF system's fundamental equations. Because this ensemble Kalman filter algorithm can dynamically allocate the weight of experimental observations for parameter updates based on the correlation between the prediction covariance and the observation error covariance, a quantitative error propagation and correction channel is established between simulation predictions and experimental observations during parameter optimization. To a certain extent, this can utilize a limited amount of experimental data to guide multiple turbulence parameters towards a direction that more closely matches the real physical state, effectively improving the calibration accuracy of turbulence empirical parameters.

[0018] 3. In the technical solution of this application, before calculating the error between the CFD simulation results and the LDA experimental data, one-dimensional linear interpolation is first performed to align the discrete grid data and discrete measurement points, using the abscissa of the experimental measurement points as the reference to obtain the simulation values ​​at the same location. Since this operation eliminates the comparison deviation caused by the spatial non-coincidence of the grid nodes and experimental measurement points, it can provide a unified and unbiased data benchmark for subsequent cost function calculation and parameter iterative updates. This, to a certain extent, ensures that the parameter calibration process is based on reliable and standardized data pairs, improving the data reliability of the entire calibration process.

[0019] 4. In the technical solution of this application, after completing one EnKF update, the updated posterior parameter set is re-substituted into the batch CFD simulation process, and the parameter sample set is iteratively updated with the goal of minimizing the RMSE cost function. Since this closed-loop iterative mechanism can continuously compare the predicted results after parameter correction with the experimental data in a new round, and continuously reduce the prediction deviation, this is equivalent to realizing an adaptive parameter optimization loop. Thus, to a certain extent, the parameter combination can gradually approach the global optimal solution through automated repeated iterations, and finally output the optimal turbulence parameter combination with the highest fitting degree to the LDA experimental data, completing the parameter calibration of the fuel rod bundle SST k-ω turbulence model. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 A flowchart illustrating a method for calibrating parameters of a CFD turbulence model of a nuclear reactor rod beam flow field, provided as an embodiment of this specification. Figure 2 This is a complete data assimilation closed-loop flowchart of the present invention, which includes "Latin hypercube sampling - CFD batch calculation - EnKF update"; Figure 3 This is a schematic diagram of the three-dimensional geometric model of the 5×5 mixed-wing rod bundle of the present invention; Figure 4 This is a schematic diagram showing the mesh division of the rod bundle and the arrangement of the LDA measurement sections at z=0.03m and z=0.09m in this invention; Figure 5 This is a comparison curve of the radial distribution of axial normalized velocity before and after assimilation under low Reynolds conditions in this invention. Figure 6 This is a comparison curve of the radial distribution of turbulence intensity before and after assimilation under high and low Reynolds conditions in this invention; Figure 7 The curves showing the variation trend of low Reynolds turbulence intensity RMSE under different sample sizes in this invention; Figure 8 The curves showing the variation trend of high Reynolds turbulence intensity RMSE under different sample sizes in this invention; Figure 9 This is a graph showing the maximum axial pressure drop as a function of sample size under high and low Reynolds conditions according to the present invention. Figure 10 This is a comparison diagram of the static pressure cloud map and the location of the maximum pressure drop under two sets of samples with high Reynolds conditions N=40 and N=100 in this invention; Figure 11 For corresponding Figure 1 A schematic diagram of a calibration device for CFD turbulence model parameters of a nuclear reactor rod beam flow field. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] As stated in the background section, traditional CFD simulations with fixed turbulence constants suffer from drawbacks such as large prediction errors across operating conditions and the lack of unified standards for manual parameter tuning. Given the limitations of existing calibration schemes in terms of generalization ability, computational cost, and comprehensive evaluation of multiple physical quantities, this invention proposes a turbulence parameter calibration method based on ensemble Kalman filtering for mixed-winged rod bundles (e.g., 5×5 reactor cores). This method employs low Reynolds LDA observation constraints and high Reynolds independent verification, quantifies sample sensitivity, and balances simulation accuracy with engineering reliability. The aim is to address the technical problems of poor generalization ability, high computational cost, and one-sided evaluation of single physical quantities in traditional turbulence parameter calibration methods.

[0024] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0025] like Figure 1 As shown, the method for calibrating parameters of a CFD turbulence model of a nuclear reactor rod beam flow field provided by the present invention includes the following steps: Step 102: Establish a 3D geometric model of the fuel rod bundle with a mixing wing and generate a computational mesh, selecting SST k- A turbulence model is used to determine the set of multi-dimensional parameter vectors to be assimilated and their value ranges. Boundary conditions consistent with experimental conditions are set, and the dimensionless conversion and physical quantity conversion standards of simulation output and experimental data are unified to build a CFD numerical simulation model of fuel rod bundles containing the computational grid.

[0026] In the embodiments of this specification, the three-dimensional geometric model of the fuel rod bundle with mixing wings can refer to the typical square-arranged fuel rod bundle structure with positioning grids and mixing wings in a pressurized water reactor fuel assembly. For example, taking a common 5×5 fuel rod bundle with mixing wings in a pressurized water reactor core as an example, for this simulated object, the fluid inside the bundle is strongly disturbed by the grid mixing wings, resulting in complex turbulent effects such as lateral flow and vortices. Therefore, conducting three-dimensional computational fluid dynamics (CFD) simulation is a necessary means to analyze the refined flow field of the fuel rod bundle. The two-equation turbulence model is a core prediction model within the RANS (Reynolds-averaged Navier-Stokes) system. It utilizes a mixing function to apply k-... in the near-wall region. The model uses k- in the far field region. The model is suitable for handling complex conditions such as strong shear and flow separation within rod bundles with mixing winglets. In this embodiment, the geometry, mesh generation, and boundary conditions of the simulation computational domain are all built with reference to publicly available standard experimental examples of rod bundle LDA. This allows the simulation accuracy to fall between that of simple single-parameter empirical correlations and high-precision large eddy simulations, ensuring the fidelity of the rod bundle flow field prediction while also taking into account the engineering requirements of parameter calibration for computational efficiency.

[0027] In the technical solution of this application, the generation of the computational grid is a fundamental step in numerical simulation, such as... Figure 3 As shown in this embodiment, when building a complete fuel rod bundle CFD numerical simulation model, it is necessary to complete the global hybrid structured mesh generation and pre-verify the mesh size independence to ensure computational accuracy. For example, two characteristic sections downstream of the mixing rotor, z=0.03m and z=0.09m, can be selected as the reference positions for LDA experimental data extraction and subsequent comparison. Simultaneously, for example, the CFD mesh node coordinates are transformed into a dimensionless coordinate system based on the rod bundle pitch, and the turbulent kinetic energy output from the simulation is converted into experimentally measurable turbulent fluctuation root-mean-square velocity, achieving a dual alignment of physical quantities and geometric coordinates. These preprocessing steps provide a unified model basis for subsequent batch sample calculations and experimental comparisons.

[0028] Step 104: The parameter space corresponding to the set of parameters to be assimilated is sampled hierarchically using the Latin hypercube sampling method to generate multiple sets of turbulent parameter samples to be assimilated. The completed CFD simulation model is called in batches to carry out parallel steady-state calculations, obtain the rod bundle flow field simulation results corresponding to each set of turbulent parameter samples to be assimilated, and extract the simulation data at the experimental measurement location. The CFD grid data is projected onto the LDA experimental measurement points through one-dimensional linear interpolation to complete the spatial alignment.

[0029] In the embodiments of this specification, the multidimensional turbulence empirical parameter vector set can contain multiple key parameters to be calibrated, involving turbulent dissipation, eddy viscosity coefficient, and inlet / outlet boundary conditions, etc. Since these parameters constitute a high-dimensional space, the computational load would increase exponentially if manual trial calculations or simple grid search methods were used, which is highly impractical. Latin hypercube sampling (LHS) is an effective stratified sampling method. It divides the range of each parameter into N equally probable sub-intervals and randomly selects sample points within each sub-interval, thereby uniformly covering the entire parameter space with a smaller sample size, avoiding the spatial clustering problem of traditional Monte Carlo sampling. Its overall sampling and data assimilation process is as follows: Figure 2 As shown, specifically, this may include: first, generating multiple sets (even up to 100 sets or more) of model constant vectors (i.e., multidimensional parameter vectors to be assimilated) through Latin hypercube sampling, and then inputting these vectors in parallel into SST k- The model performs flow field predictions and obtains the corresponding predicted values.

[0030] After generating the sample set, this embodiment introduces a batch processing parallel computing mechanism, that is, calling the CFD simulation model at once and simultaneously completing the steady-state flow field calculation of multiple sets of samples under stable geometric grid and boundary condition settings. To ensure that the simulation results are comparable with the experimental data, simulation data at the experimental measurement locations can be extracted. Since the coordinates of the CFD grid nodes and the coordinates of the discrete measurement points in the LDA (Laser Doppler Velocimetry) experiment usually cannot completely coincide, this embodiment uses a one-dimensional linear interpolation method, using the dimensionless abscissa of the experimental measurement points as a reference, to project the simulated physical quantities on the CFD grid onto the corresponding experimental measurement points. This can eliminate the data comparison bias caused by the spatial mismatch and obtain standardized, uniformly calculable assimilated data pairs.

[0031] Step 106: Introduce the LDA experimental measurement data of the rod bundle flow field, establish the error propagation and correction relationship between the CFD simulation results and the experimental observation data based on the ensemble Kalman filter algorithm, calculate the ensemble mean, prediction error covariance matrix and Kalman gain of the sample set of turbulent parameters to be assimilated, perform the posterior update of the sample set of turbulent parameters to be assimilated, and construct the root mean square error (RMSE) cost function to quantify the prediction deviation between the CFD simulation results and the LDA experimental observation data.

[0032] In the embodiments of this specification, LDA experimental measurement data can provide true observations of the flow field and serve as a benchmark for calibrating turbulence model parameters. This invention employs the Ensemble Kalman Filter (EnKF) algorithm to achieve data assimilation. This algorithm, based on Bayesian inference principles, can effectively integrate the uncertainties of the observation data into the prediction of model parameters. For example... Figure 2 As shown, the observed values ​​obtained from the experiment. The predicted values ​​from the aforementioned model are input together into a Kalman filter algorithm for assimilation analysis. First, the predicted states of the sample set are statistically analyzed, and the ensemble mean of the flow field and parameter states for all samples is calculated. Then, the prediction error covariance matrix is ​​solved by calculating the sum of squared sample biases. The covariance matrix can reflect the degree of correlation between turbulence parameters and rod bundle flow field observations.

[0033] The key to the EnKF algorithm lies in calculating the Kalman gain matrix. This matrix is ​​a weighting matrix obtained by jointly solving the prediction error covariance matrix and the observation error covariance matrix. It determines the extent to which the observed data can correct the current parameter prediction bias. In this scheme, the LDA experimental observation vectors are used... By introducing and combining Kalman gain with posterior updates, the corrected sample set will more closely approximate the actual flow field state. To quantitatively evaluate the correction effect, this embodiment introduces the root mean square error (RMSE) as a cost function, which is the square root of the average of the sum of squared residuals between the simulation interpolation results and the LDA experimental data corresponding to the current parameter sample set. The RMSE value can be used to intuitively quantify the overall deviation between the simulation prediction and physical reality under the current operating conditions.

[0034] Step 108: With minimizing the RMSE cost function as the optimization objective, the updated posterior set of turbulent parameters to be assimilated is re-substituted into the batch CFD simulation process. The set of turbulent parameters to be assimilated is iteratively updated until the optimal combination of turbulent parameters after iterative optimization converges is output, thus completing the parameter calibration of the fuel rod bundle turbulence model.

[0035] In the embodiments described in this specification, the parameter calibration process is a closed-loop iterative optimization process. After the initial posterior update of the parameter sample set using the EnKF algorithm, this method does not stop directly, but instead re-imports the updated parameter vector set into the batch CFD simulation flow. At this time, the new parameter vectors will be substituted into SST k- The model is recalculated to generate new flow field predictions. Then, the interpolation alignment in step 104 and the EnKF update and RMSE error calculation in step 106 are performed again, and this process is repeated.

[0036] In each iteration, the RMSE cost function theoretically converges towards a smaller value. This embodiment sets a preset convergence threshold during the iteration process. Once the RMSE change between consecutive iterations is less than this threshold (i.e., the model prediction accuracy no longer significantly improves), or the average deviation between the simulation results and experimental data meets specified requirements, the iteration process is considered complete, and the final converged optimal turbulence parameter combination is output. This optimal parameter combination enables the CFD simulation results to fit the LDA experimental measurement data to the maximum extent, thereby completing the simulation of SST k- with mixed airfoil bundles under the target operating condition. Precise parameter calibration of the turbulence model.

[0037] Based on the technical solutions described above, this specification also provides some specific implementation schemes, which are described below.

[0038] In an optional embodiment, the step of establishing a three-dimensional geometric model of the fuel rod bundle with a mixing wing and generating a computational mesh involves selecting SST k- The steps for constructing a CFD numerical simulation model of a fuel rod bundle containing the computational grid include: defining the set of multi-dimensional parameter vectors to be assimilated and their value ranges, setting boundary conditions consistent with experimental conditions, and unifying the dimensionless conversion and physical quantity conversion standards between simulation output and experimental data. Construct the geometric computational domain of the rod bundle, and select SST k- under the RANS time-averaged system. The two-equation turbulence model serves as the core prediction model, and its turbulent kinetic energy transport equation and specific dissipation rate transport equation are as follows: , ; In the formula For fluid density, For dynamic viscosity, Where is the eddy viscosity coefficient. The viscosity coefficient of the eddy current is . Let Reynolds stress tensor be the stress tensor. For velocity components, For coordinate components, For turbulent kinetic energy, This represents the value of the strain rate. This is the vorticity value. For model constants, the mixture function The interpolation rule is ,in , , , For SST k- Empirical constants for turbulence models, used to achieve near-wall k- With far-field k- The smooth transition of the model is achieved by solving for its eddy viscosity coefficient using the following formula: and In the formula The absolute value of the vorticity in the flow field. For the second mixing function, These are the model's empirical constants. The determined vector of parameters to be assimilated is... ,in , , , For SSTk- Model core constants , These are the boundary condition parameters for the inlet and outlet turbulent kinetic energy. , These are the boundary condition parameters for the inlet and outlet specific dissipation rates, and each parameter is used to solve for turbulent dissipation, eddy viscosity coefficient, and inlet and outlet turbulent boundary conditions.

[0039] The technical solution in this embodiment uses mathematical equations to analyze SST k- The closure of the turbulence model is physically described. Among these, turbulent kinetic energy... Sum of dissipation The transport equations together determine the evolution of turbulent characteristics in the flow field. The equations in... The Reynolds stress, multiplied by the velocity gradient, represents the generation term of turbulent kinetic energy; while and This is the turbulent dissipation term, which determines the decay rate of turbulent kinetic energy. The smooth switching mechanism used by the SST model for the near-wall and far-field regions is highly dependent on the mixing function. It automatically determines the k-value to use based on the distance from the flow field to the wall. Or k- The model's feature parameters constitute the fusion formula. The physical meaning of eddy viscosity coefficient. and eddy viscosity The solution formula introduces a second mixture function. With absolute value of vorticity It can calculate the viscous effect of turbulence on the flow.

[0040] Due to SST k- The model's standard constants are calibrated based on flat-plate flow, and directly applying them to complex, strongly swirling flow fields with turbulent blade bundles often results in significant deviations. Therefore, this embodiment will directly affect four core constants of the model's physical mechanism ( , , , ) and four inlet and outlet turbulence boundary parameters ( , , , Together they form the parameter vector to be assimilated. The introduction of this parameter vector can provide a clear physical parameter dimension for the subsequent EnKF data assimilation process. By pre-setting a uniform prior interval with a certain percentage of fluctuation above and below the standard value, for example, a fluctuation range of 20% to 50% can be selected as the prior interval, thereby ensuring that the search space has both physical rationality and sufficient optimization coverage.

[0041] In an optional embodiment, the step of standardizing the dimensionless conversion and physical quantity conversion of the unified simulation output and experimental data may specifically include: Extract the geometric coordinates along the measurement line and perform dimensionless grid pitch conversion. The coordinate transformation formula is as follows: ; In the formula, Y represents the original spanwise coordinate of the cross section, and Pitch represents the bar strand pitch. This formula is used to align the CFD mesh nodes with the abscissa system of the LDA test measurement points. The definition of the average axial velocity during the LDA test is as follows: ; In the formula The axial velocity during LDA testing, Instantaneous velocity Where N is the sampling time interval and N is the number of sampling points; the root mean square velocity of turbulent fluctuations is derived from turbulent kinetic energy, and the conversion formula is: ; In the formula Let k be the root mean square velocity of the turbulent fluctuations, and k be the turbulent kinetic energy. Then, normalize this to the cross-sectional average velocity to obtain the dimensionless physical quantity of the assimilated observation target. as well as In the formula, W is the axial velocity. The average flow velocity across the cross section is given.

[0042] In this embodiment, the data obtained from CFD simulation are the raw physical quantities at discrete grid nodes, while LDA experimental measurements are performed at a series of fixed physical spatial points. The two cannot directly correspond in spatial location. Therefore, this embodiment first defines... The dimensionless coordinate transformation formula divides the spanwise coordinates of the rod bundle section by the rod bundle pitch, allowing the geometry of rod bundles of different scales and the positions of measurement points to be mapped to a unified coordinate range. This eliminates the differences in absolute spatial dimensions for subsequent interpolation and error calculation. Meanwhile, the LDA system measures instantaneous velocity signals, and its actual output is the average axial velocity during the test. By measuring instantaneous velocity The CFD output is obtained by time-weighted averaging, and this embodiment follows the definition to match the output. Furthermore, the CFD calculation yields turbulent kinetic energy. LDA equipment can only directly measure the root mean square value of velocity pulsation. This embodiment establishes a transformation formula based on the isotropic assumption. This converts the original turbulent kinetic energy into measurable fluctuating velocity, and further modulates it with the cross-sectional average velocity. Divide to obtain the dimensionless assimilation target quantity. and .

[0043] In an optional embodiment, the step of using the Latin hypercube sampling method to perform stratified sampling of the parameter space corresponding to the set of parameter vectors to be assimilated, and generating multiple sets of turbulent parameter sample sets to be assimilated, may specifically include: The parameter space corresponding to the set of parameter vectors to be assimilated is divided into N equally probable sub-intervals. For each sample group, a single parameter component is randomly selected from each sub-interval and then randomly paired. Here, N is a preset positive integer used to set different sample size levels. The parameter vectors to be assimilated obtained from each sampling group are input into SSTk- The turbulence model is used to perform batch steady-state CFD calculations to obtain the flow field state matrix corresponding to all samples in the set.

[0044] In this embodiment, Latin hypercube sampling (LHS) can address the high-dimensional hierarchical sampling problem in multidimensional parameter spaces. Traditional Monte Carlo sampling is prone to sample point clustering when the sample size is small, failing to cover the boundary regions of all parameters. This scheme defines prior floating intervals for each of the multiple parameters involved in assimilation and equally divides each interval into N sub-intervals. A sample component is randomly selected within each sub-interval, and then the sample components of each dimension are randomly paired, ensuring that even with a small sample size N, the sample points are uniformly distributed in the multidimensional space without clustering. As an example, taking a 5×5 mixed rod bundle as an example, when N=20, only a small number of samples are needed to completely cover 20%~50% of the floating intervals of each parameter, significantly reducing the number of CFD batch calculations. This sampling strategy can extract the influence information of the parameter space on the rod bundle flow field response to the maximum extent with very few CFD calculations, providing a high-quality initial sample set for the EnKF algorithm.

[0045] In an optional embodiment, the step of extracting simulation data at the experimental measurement location and projecting the CFD mesh data onto the LDA experimental measurement points using one-dimensional linear interpolation to complete spatial alignment may specifically include: One-dimensional linear interpolation is performed to align the CFD discrete mesh data with the experimental discrete measurement points. The interpolation is based on the abscissa of the experimental measurement points to obtain the simulation values ​​at the same location, thus eliminating the comparison deviation caused by the non-coincidence of the mesh and the measurement points. The interpolation calculation formula is as follows: ; In the formula For the simulated value at the i-th interpolation point, The dimensionless abscissa of the test measurement point, , The discrete x-coordinates of adjacent nodes in the CFD mesh. , This represents the flow field value at the corresponding node location.

[0046] This embodiment addresses the practical engineering problem of misalignment between CFD mesh node coordinates and LDA fixed measurement point coordinates by employing piecewise linear interpolation mapping technology. The CFD mesh is unstructured or hybrid within the rod bundle computational domain, and its abscissa distribution depends on the flow field solution accuracy and mesh quality. In contrast, the abscissa of the LDA experimental measurement points is a pre-defined fixed coordinate. The interpolation formula assumes a linear distribution of flow field physical quantities (such as axial velocity and turbulent kinetic energy) between adjacent nodes in the CFD mesh. A scaling factor is used to accurately interpolate and extract the physical quantities at the corresponding locations of the experimental measurement points from the CFD mesh nodes. For measurement points outside the simulation range, a boundary assignment strategy is adopted to ensure reliable alignment between simulation results and experimental data. As an example, taking a 30mm section downstream of the mixer rotor as an example, the experiment has 45 measurement points. The interpolation formula yields simulation prediction values ​​corresponding to each measurement point, eliminating comparison errors caused by mesh misalignment and providing a unified standardized dataset for the cost function calculation in step 106.

[0047] In an optional embodiment, the steps of introducing LDA experimental measurement data of the rod bundle flow field, establishing the error propagation and correction relationship between CFD simulation results and experimental observation data based on the ensemble Kalman filter algorithm, and calculating the ensemble mean, prediction error covariance matrix, and Kalman gain of the turbulence parameter sample set to be assimilated may specifically include: The fundamental equations of the EnKF system are constructed, and the ensemble prediction process expression is as follows: ; In the formula For the first Prior predicted state of the group of samples, For the first The initial parameter set of the group, F(·), is SST k- CFD prediction model Represents the system noise term; the expression for the set state matrix is ​​as follows: ; In the formula For the first The flow field prediction vector corresponding to the group of samples For the first Given a set of parameter vectors to be assimilated; solve for the ensemble mean and the prediction error covariance matrix. The formula for calculating the ensemble mean is as follows: ; In the formula The average flow field and parameter states for all samples. The total number of samples in the current sample set; the expression for the prediction covariance matrix is ​​as follows: ; In the formula, P is the prediction error covariance matrix, which is used to quantify the degree of correlation between turbulence parameters and flow field observation variables.

[0048] This embodiment describes the prediction and statistical calculation process of ensemble Kalman filtering. Expressing SST k- The CFD simulation model transforms initial parameter samples into prior predicted states. (Set state matrix) The first row in the two-dimensional structure is composed of The first row contains the flow field prediction vectors (such as velocity and turbulence intensity fields) corresponding to each group of samples; the second row contains the parameter vectors to be assimilated for each group. This enables the synchronous evolution and correlation between the flow field state and model parameters. (Set mean) This represents the average predicted state of the system under the current prior parameter distribution. Prediction error covariance matrix. By taking the outer product of the deviations of all sample states from the mean and calculating the statistical average, the amplitude and direction characteristics of the flow field state changes caused by parameter variations can be captured.

[0049] In an optional embodiment, the step of performing a posteriori update of the set of turbulent parameter samples to be assimilated may specifically include: Construct a diagonal form of the observation error covariance matrix. The single-point observation error constraint is as follows: ; In the formula For the first Standard deviation of observations at each measuring point The observation error covariance matrix for the LDA test observations at the measuring point is as follows: ; In the formula, R is the observation error covariance matrix; the Kalman gain matrix is ​​calculated to complete the error weight allocation, and the gain solution formula is as follows: ; In the formula, K is the Kalman gain matrix, used to assign correction weights to the turbulence parameters based on experimental observation data; B is the observation operator matrix; and T is the matrix transpose sign. After performing ensemble sample posterior update, the parameter and flow field state correction formulas are as follows: ; In the formula For the updated version Group posterior state To standardize the LDA experiment observation vector, To observe the disturbance noise vector.

[0050] This embodiment incorporates the inherent error of the experimental data into the parameter assimilation process. For example, an LDA laser Doppler velocimetry system typically has a measurement uncertainty of approximately 5% and a single-point observation standard deviation. In the calculation formula, it is set to 0.05 times the maximum value of the observed absolute value and The lower bound of the measurement error covariance matrix can reflect the true error level of physical measurements and effectively prevent matrix singularities caused by excessively small values ​​at a few individual measurement points. Constructed as a diagonal matrix, its diagonal elements are the squares of the errors at each measurement point. Assuming that the measurement points are independent simplifies the calculation. After obtaining... and After that, Kalman gain The solution formula essentially seeks the optimal weight allocation between the reliability of the prediction (represented by the covariance P and the observation operator B) and the reliability of the observation (represented by the covariance R). Ultimately, it is achieved through... Perform a post-hoc update. The larger the value, the greater the weight of the observation on the state correction. This process uses real experimental data to calibrate the Bayesian optimal estimate of the model's prior state, which can effectively bring the flow field prediction closer to the experimental observation.

[0051] In an optional embodiment, after the step of calibrating the fuel rod bundle turbulence model parameters by outputting the optimal combination of turbulent parameters after iterative optimization convergence, the method further includes a generalization capability verification and comprehensive evaluation step, which may specifically include: The optimal parameter vector obtained from the output is re-imported into the CFD simulation model, and the flow field is recalculated under both the assimilation condition involved in the calibration process and the generalized condition independent of the calibration process. The simulation output data is processed uniformly according to the coordinate dimensionless formula and the linear interpolation formula. Based on the RMSE calculation formula, the prediction errors of the axial normalized velocity and turbulence intensity under the two types of conditions are calculated respectively. The error reduction of the initially set multi-dimensional parameter vector set before assimilation is compared with that of the optimal parameters after assimilation. The radial and axial maximum pressure drop data that did not participate in the assimilation are extracted, and the convergence characteristics of the pressure drop value and the location of the maximum pressure drop under different sample sizes are analyzed. The engineering stability and cross-condition generalization ability of the optimal parameters are comprehensively evaluated.

[0052] This embodiment introduces a cross-Reynolds number dual-condition verification mechanism after parameter calibration to avoid overfitting the model under a single condition. Specifically, the optimal parameter vector obtained from calibration can be re-imported into the CFD simulation model. First, a recalculation is performed under the low Reynolds condition that is involved in assimilation to verify its fit to the training data. Subsequently, the same flow field recalculation is performed under the independent generalized high Reynolds condition that has not been involved in assimilation. Since the flow state (such as turbulence intensity and Reynolds stress distribution) of this condition is significantly different from that of the assimilated condition, its prediction error can truly reflect the physical adaptability of the calibration parameters to different flow characteristics.

[0053] This embodiment not only utilizes RMSE to evaluate the accuracy of assimilated physical quantities such as velocity and turbulence intensity, but also extracts the flow pressure drop, which was not involved in the data assimilation process, for comparative analysis. Combined with... Figure 5 , Figure 6 The flow field characteristic comparison curves shown indicate that, after adopting the optimal parameters (SST-optimal) specified in this invention, the agreement between the predicted axial normalized velocity and the experimental data is improved to varying degrees compared to the default parameters (SST), especially in terms of turbulence intensity (i.e., In terms of prediction, the peak-valley characteristics and overall amplitude accuracy of the parameter prediction curve are improved after calibration, which demonstrates the selective optimization effect of parameter calibration on different physical quantities. Since pressure drop is a key engineering parameter in reactor thermal-hydraulic analysis, this technical solution also quantitatively reveals the numerical stability of parameter calibration for key engineering parameters such as pressure drop by comparing the convergence of pressure drop values ​​and the peak position of maximum pressure drop under different sample sizes, thereby comprehensively evaluating the generalization ability and engineering reliability of the optimal parameters.

[0054] In an optional embodiment, the step of iteratively updating the set of turbulent parameter samples to be assimilated further includes convergence determination and sample traversal determination, which may specifically include: Determining whether an iteration has converged can specifically include: during the iteration process, judging whether the change magnitude of the RMSE cost function of consecutive iteration steps and the average deviation between simulation results and experimental data meet the preset convergence threshold requirements. If the convergence requirement is met, the iteration is terminated and the optimal turbulence parameters are obtained; if the convergence requirement is not met, the iteration continues to update the turbulence parameter sample set. Determining whether the entire parameter calibration process has ended can specifically include: determining whether the current sample size has reached all preset levels; If not all samples are traversed, the sample size is changed, and a new complete calibration cycle is executed. If all sample sizes have been assimilated, validated, and statistically analyzed, the overall calibration process ends, and a complete set of calibration parameters, error curves, and engineering sample selection recommendations are output.

[0055] In this embodiment, the core of convergence determination lies in monitoring the decay trend of the cost function during iterative optimization. When the change magnitude of the RMSE cost function between two consecutive iterations, as well as the error decay ratio of high and low Reynolds number turbulence intensity, are both stably lower than the preset convergence threshold (i.e., the model's predictive ability no longer significantly improves), the EnKF iteration process of the current sample size can be considered to have reached the optimal state. At this point, the parameter update loop is exited, and the optimal turbulence parameter combination under the current sample size is locked.

[0056] Furthermore, considering the sensitivity of calibration results to sample size, this embodiment also designs a systematic sample traversal process, the overall sensitivity results of which are as follows: Figure 7 , Figure 8 and Figure 9 As shown. Combined with Figure 8 As can be seen, for example, when the number of samples increases from the initial default state to 20, the turbulence intensity RMSE decreases rapidly. As the number of samples further increases (40, 60, 80, 100), the RMSE continues to converge slowly and tends to stabilize. Similarly, combined with... Figure 9 Under high Reynolds conditions, the maximum axial pressure drop corresponding to different sample numbers converged well to a stable range after experiencing fluctuations within a certain range, indicating that the calibration parameters possess engineering stability. As an example, Figure 10 The comparison of static pressure contour maps of the downstream surface under high Reynolds conditions (N=40 and N=100) shows that the location of the maximum pressure drop is consistently located in the downstream characteristic region of the mixing vane under different sample sizes, with consistent numerical distribution characteristics. This further confirms the stability and reliability of the optimal parameters under different sampling statistical conditions. The technical solution in this application requires the completion of complete calibration and verification calculations for multiple sample sizes (N=20, 40, 60, 80, and 100) as examples. The entire calibration solution process can only be completed after all preset sizes have undergone full-process assimilation and statistical analysis. The purpose of this design is to provide engineers with recommendations for selecting the minimum sample size that balances computational cost and accuracy by comparing the calibration error curves and pressure drop stability characteristics under different sample sizes.

[0057] It should be understood that in the methods described in one or more embodiments of this specification, the order of some steps may be adjusted according to actual needs, or some steps may be omitted.

[0058] Based on the foregoing technical solutions, this invention also provides a calibration device for the parameters of a CFD turbulence model of a nuclear reactor rod beam flow field, such as... Figure 11 As shown, the device, from a macroscopic perspective, may include the following modules: CFD simulation modeling module 1102 is used to establish a three-dimensional geometric model of fuel rod bundle with mixing wing and complete the computational mesh generation, select the SST k-ω turbulence model, determine the set of multi-dimensional turbulence empirical parameter vectors to be assimilated and their value ranges, set boundary conditions consistent with the experimental conditions, unify the dimensionless conversion and physical quantity conversion standards between simulation output and experimental data, and build a CFD numerical simulation model of fuel rod bundle. The parameter sampling and batch calculation module 1104 is used to perform hierarchical sampling of the parameter space using the Latin hypercube sampling method, generate multiple sets of turbulent parameter sample sets to be assimilated, batch call the completed CFD simulation model to carry out parallel steady-state calculations, obtain the rod bundle flow field simulation results corresponding to each set of parameter samples, extract the simulation data at the experimental measurement location, and project the CFD grid data to the LDA experimental measurement points through one-dimensional linear interpolation to complete spatial alignment. The ensemble Kalman filter assimilation module 1106 is used to introduce LDA experimental measurement data of rod bundle flow field, establish the error propagation and correction relationship between CFD simulation results and experimental observation data based on the ensemble Kalman filter algorithm, calculate the ensemble mean, prediction error covariance matrix and Kalman gain, perform posterior update of parameter sample set, and construct root mean square error (RMSE) cost function to quantify prediction bias. The iterative optimization module 1108 is used to re-substitute the updated posterior turbulence parameter sample set into the batch CFD simulation process with the RMSE cost function minimization as the optimization objective, and iteratively update the turbulence parameter sample set to be assimilated. The calibration and verification output module 1110 is used to output the optimal combination of turbulent parameters after iterative optimization convergence, and to complete the parameter calibration of the fuel rod bundle turbulence model.

[0059] Those skilled in the art will understand that the accompanying drawings are merely schematic diagrams of one embodiment, and the modules or processes shown in the drawings are not necessarily essential for implementing the present invention.

[0060] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for calibrating parameters of a CFD turbulence model of a nuclear reactor rod beam flow field, characterized in that, Including the following steps: A three-dimensional geometric model of the fuel rod bundle with a mixing wing was established and the computational mesh was generated. SST k- was selected. Turbulence model, determine the set of multi-dimensional turbulence empirical parameter vectors to be assimilated and their value ranges, set boundary conditions consistent with experimental conditions, unify the dimensionless conversion and physical quantity conversion standards between simulation output and experimental data, and build a CFD numerical simulation model of fuel rod bundle. The parameter space was hierarchically sampled using the Latin hypercube sampling method to generate multiple sets of turbulent parameter samples to be assimilated. The completed CFD simulation model was called in batches to carry out parallel steady-state calculations, obtain the rod bundle flow field simulation results corresponding to each set of parameter samples, and extract the simulation data at the experimental measurement location. The CFD grid data was projected onto the LDA experimental measurement points through one-dimensional linear interpolation to complete the spatial alignment. LDA experimental measurement data of rod bundle flow field is introduced. Based on the ensemble Kalman filter algorithm, the error propagation and correction relationship between CFD simulation results and experimental observation data is established. The ensemble mean, prediction error covariance matrix and Kalman gain are calculated. The posterior update of the parameter sample set is performed. The root mean square error (RMSE) cost function is constructed to quantify the prediction bias. With minimizing the RMSE cost function as the optimization objective, the updated posterior turbulence parameter sample set is re-substituted into the batch CFD simulation process, and the turbulence parameter sample set to be assimilated is updated iteratively. Output the optimal combination of turbulence parameters after iterative optimization convergence to complete the parameter calibration of the fuel rod bundle turbulence model.

2. The method for calibrating the parameters of the CFD turbulence model of the nuclear reactor rod beam flow field as described in claim 1, characterized in that, The process involves establishing a three-dimensional geometric model of the fuel rod bundle with a mixing wing and completing the computational mesh generation, selecting SST k- The steps for building a CFD numerical simulation model of a fuel rod bundle include: determining the set of multi-dimensional turbulence empirical parameter vectors to be assimilated and their value ranges; setting boundary conditions consistent with experimental conditions; and unifying the dimensionless conversion and physical quantity conversion standards between simulation output and experimental data. Construct the geometric computational domain of the rod bundle, and select SST k- under the RANS time-averaged system. The two-equation turbulence model serves as the core prediction model, and its turbulent kinetic energy transport equation and specific dissipation rate transport equation are as follows: , ; In the formula For fluid density, For dynamic viscosity, Where is the eddy viscosity coefficient. The viscosity coefficient of the eddy current is . Let Reynolds stress tensor be the stress tensor. For velocity components, For coordinate components, This represents the value of the strain rate. This is the vorticity value. For model constants, the mixture function The interpolation rule is Used to achieve near-wall k- With far-field k- The model exhibits a smooth transition; its eddy viscosity coefficient is calculated using the following formula: and In the formula The absolute value of the vorticity in the flow field. The second mixing function is defined as follows: The determined parameter vector to be assimilated is... Each parameter is used to solve for turbulent dissipation, eddy viscosity coefficient, and inlet / outlet turbulent boundary conditions.

3. The method for calibrating the parameters of the CFD turbulence model of the nuclear reactor rod beam flow field as described in claim 1, characterized in that, The steps for standardizing the dimensionless conversion and physical quantity conversion of the unified simulation output and experimental data include: Extract the geometric coordinates along the measurement line and perform dimensionless conversion of the grid pitch. The coordinate transformation formula is as follows: ; In the formula, Y represents the original spanwise coordinate of the cross section, and Pitch represents the bar strand pitch. This formula is used to align the CFD mesh nodes with the abscissa system of the LDA test measurement points. The definition of the average axial velocity during the LDA test is as follows: ; In the formula The axial velocity during LDA testing, Instantaneous velocity Where N is the sampling time interval and N is the number of sampling points; the root mean square velocity of turbulent fluctuations is derived from turbulent kinetic energy, and the conversion formula is: ; In the formula, The root mean square velocity of the turbulent fluctuations. The turbulent kinetic energy is then normalized to the cross-sectional average velocity to obtain the dimensionless physical quantity of the assimilated observation target. as well as In the formula, W is the axial velocity. The average flow velocity across the cross section is given.

4. The method for calibrating the parameters of the CFD turbulence model of the nuclear reactor rod beam flow field as described in claim 1, characterized in that, The step of using the Latin hypercube sampling method to perform stratified sampling of the parameter space and generate multiple sets of turbulent parameter sample sets to be assimilated includes: The parameter value range in the parameter space to be assimilated is equally divided into N equally probable sub-intervals. For each sample group, a single parameter component is randomly selected from each sub-interval and then randomly paired. Here, N is a preset positive integer used to set different sample size levels. The multi-dimensional parameter vector obtained from each sampling is input into SST k- The turbulence model is used to perform batch steady-state CFD calculations to obtain the flow field state matrix corresponding to all samples in the set.

5. The method for calibrating the parameters of the CFD turbulence model of the nuclear reactor rod beam flow field as described in claim 1, characterized in that, The step of extracting simulation data at the experimental measurement location and projecting the CFD mesh data onto the LDA experimental measurement points through one-dimensional linear interpolation to complete spatial alignment includes: One-dimensional linear interpolation is performed to align the CFD discrete mesh data with the experimental discrete measurement points. The interpolation is based on the abscissa of the experimental measurement points to obtain the simulation values ​​at the same location, thus eliminating the comparison deviation caused by the non-coincidence of the mesh and the measurement points. The interpolation calculation formula is as follows: ; In the formula For the first Simulated values ​​at each interpolation point The dimensionless abscissa of the test measurement point, , The discrete x-coordinates of adjacent nodes in the CFD mesh. , This represents the flow field value at the corresponding node location.

6. The method for calibrating the parameters of the CFD turbulence model of the nuclear reactor rod beam flow field as described in claim 1, characterized in that, The steps of introducing LDA experimental measurement data of the rod bundle flow field, establishing the error propagation and correction relationship between CFD simulation results and experimental observation data based on the ensemble Kalman filter algorithm, and calculating the ensemble mean, prediction error covariance matrix, and Kalman gain include: The fundamental equations of the EnKF system are constructed, and the ensemble prediction process expression is as follows: ; In the formula For the first Prior predicted state of the group of samples, For the first The initial parameter set of the group, F(·), is SST k- CFD prediction model Represents the system noise term; the expression for the set state matrix is ​​as follows: ; In the formula For the first The flow field prediction vector corresponding to the group of samples For the first A set of multidimensional turbulent parameter vectors to be assimilated is obtained; the ensemble mean and prediction error covariance matrix are solved. The formula for calculating the ensemble mean is as follows: ; In the formula The average flow field and parameter states for all samples. The total number of samples in the current sample set; the expression for the prediction covariance matrix is ​​as follows: ; In the formula, This is the prediction error covariance matrix, used to quantify the correlation between turbulence parameters and observed flow field variables.

7. The method for calibrating the parameters of the CFD turbulence model of the nuclear reactor rod beam flow field as described in claim 6, characterized in that, The step of posteriorly updating the execution parameter sample set includes: Construct a diagonal form of the observation error covariance matrix. The single-point observation error constraint is as follows: ; In the formula, For the first Standard deviation of observations at each measuring point The observation error covariance matrix for the LDA test observations at the measuring point is as follows: ; In the formula, R is the observation error covariance matrix; the Kalman gain matrix is ​​calculated to complete the error weight allocation, and the gain solution formula is as follows: ; In the formula, K is the Kalman gain matrix, used to assign correction weights to the turbulence parameters based on experimental observation data. To observe the operator matrix, Here is the matrix transpose notation; after performing ensemble sample posterior update, the parameter and flow field state correction formulas are as follows: ; In the formula, For the updated version Group posterior state To standardize the LDA experiment observation vector, To observe the disturbance noise vector.

8. The method for calibrating the parameters of the CFD turbulence model of the nuclear reactor rod beam flow field as described in claim 1, characterized in that, After completing the parameter calibration step for the fuel rod bundle turbulence model, the output iterative optimization converges to obtain the optimal combination of turbulence parameters. The process also includes generalization capability verification and comprehensive evaluation steps, including: The optimal parameter vector obtained from the output is re-imported into the CFD simulation model, and the flow field is recalculated under the first Reynolds number condition involving assimilation and the second Reynolds number condition not involving assimilation. The simulation output data is processed uniformly according to the coordinate dimensionless formula and the linear interpolation formula. Based on the RMSE calculation formula, the prediction errors of axial normalized velocity and turbulence intensity are calculated for the two types of conditions, and the error reduction of the default parameters before assimilation and the optimal parameters after assimilation is compared. The maximum radial and axial pressure drop data not involving assimilation are extracted, and the convergence characteristics of pressure drop values ​​and peak positions under different sample sizes are analyzed. The engineering stability and cross-condition generalization ability of the optimal parameters are comprehensively evaluated.

9. The method for calibrating the parameters of the CFD turbulence model of the nuclear reactor rod beam flow field as described in claim 1, characterized in that, The step of iteratively updating the set of turbulent parameter samples to be assimilated further includes convergence determination and sample traversal determination, including: Determining whether the iteration has converged specifically includes: during the iteration process, judging whether the change magnitude of the RMSE cost function of the consecutive iteration steps and the average deviation between the simulation results and the experimental data meet the preset convergence threshold requirements. If the convergence requirement is met, the iteration is terminated and the optimal turbulence parameters are obtained; if the convergence requirement is not met, the iteration continues to update the turbulence parameter sample set. Determine whether the entire parameter calibration process has ended, specifically including: determining whether the current sample size has reached all preset levels; If not all samples are traversed, the sample size is changed, and a new complete calibration cycle is executed. If all sample sizes have been assimilated, validated, and statistically analyzed, the overall calibration process ends, and a complete set of calibration parameters, error curves, and engineering sample selection recommendations are output.

10. A calibration device for CFD turbulence model parameters of a nuclear reactor rod beam flow field, characterized in that, include: The CFD simulation modeling module is used to establish a three-dimensional geometric model of the fuel rod bundle with a mixing wing and complete the computational mesh generation. The SST k-ω turbulence model is selected, the set of multi-dimensional turbulence empirical parameter vectors to be assimilated and their value ranges are determined, boundary conditions consistent with the experimental conditions are set, the dimensionless conversion and physical quantity conversion standards of simulation output and experimental data are unified, and the CFD numerical simulation model of the fuel rod bundle is built. The parameter sampling and batch calculation module is used to perform hierarchical sampling of the parameter space using the Latin hypercube sampling method, generate multiple sets of turbulent parameter sample sets to be assimilated, batch call the built CFD simulation model to carry out parallel steady-state calculations, obtain the rod bundle flow field simulation results corresponding to each set of parameter samples, extract the simulation data at the experimental measurement location, and project the CFD grid data to the LDA experimental measurement points through one-dimensional linear interpolation to complete spatial alignment. The ensemble Kalman filter assimilation module is used to introduce LDA experimental measurement data of rod bundle flow field. Based on the ensemble Kalman filter algorithm, it establishes the error propagation and correction relationship between CFD simulation results and experimental observation data, calculates the ensemble mean, prediction error covariance matrix and Kalman gain, performs posterior update of parameter sample set, and constructs the root mean square error (RMSE) cost function to quantify prediction bias. The iterative optimization module is used to minimize the RMSE cost function, resubmit the updated posterior turbulence parameter sample set into the batch CFD simulation process, and iteratively update the turbulence parameter sample set to be assimilated in a loop. The calibration and verification output module is used to output the optimal combination of turbulent parameters after iterative optimization convergence, and to complete the parameter calibration of the fuel rod bundle turbulence model.