Research On CFD Data Reconstruction Technology Of Pwr Reactor Core Flow-field

The reactor core is the most critical equipment in the nuclear power system,and ensuring its safety is one of the core components of core design and reactor operation.As an important means of reactor safety analysis,core thermal-hydraulic calculation based on Computational Fluid Dynamics(CFD)simulates the core to calculate key thermal-hydraulic parameters such as core flow and heat transfer,providing crucial steps and reference objects for designing reactors and developing nuclear power plant operation standards.With the significant improvements in computer performance and model accuracy over recent decades,the accuracy and reliability of CFD analysis technology have progressed considerably,offering researchers high-fidelity coolant flow and heat transfer data,and gradually becoming an essential component in supporting the digitization and intelligent development of nuclear power technologies.However,due to the complex structural characteristics and cross-scale flow phenomena within the core,the enormous number of meshes and complex models required in standard CFD processes need to solve high-dimensional partial differential equations(PDEs),leading to challenges such as high calculation cost and long calculation time,and difficulty in conducting uncertainty analysis in full-size or local size.Data-driven flow field reconstruction methods can use data to obtain the relationship between model inputs and outputs,avoiding PDEs solutions in CFD calculations,and achieving fast calculation response.This efficient computing method provides a very attractive alternative solution for problems that require fast and high-fidelity flow field reconstruction and critical parameter prediction.This paper conducted research on the core CFD data reconstruction technique based on the data-driven approach,and based on this,explore the cross-scale flow field reconstruction,data driven based thermal-hydraulic and neutron coupling calculation,and transient core data reconstruction techniques.The research focus of this paper can be summarized as follows:(1)Aiming for the CFD flow field data reconstruction problem of the reactor core,a flow field data reconstruction model based on feature extraction method and surrogate model is proposed.In the model,the feature extraction algorithm is used to reduce the dimension of CFD data,and the surrogate model is used to establish the mapping relationship between boundary conditions and CFD data in low-dimensional space,so as to realize the rapid and accurate estimation of core flow field.The performance of linear and nonlinear feature extraction and surrogate model techniques in the reconstruction process is studied.The results show that the reconstructed model based on linear method has better reconstruction accuracy when the sample size is small,and the prediction accuracy of the reconstructed model based on nonlinear method is higher with the increase of sample size.The results of core channel flow field reconstruction show that the proposed reconstruction scheme can accurately reconstruct the reactor core flow field,and can achieve ms-level efficiency for flow field prediction.(2)In order to improve the prediction accuracy and flexibility of the flow field reconstruction model of a single data source in high-dimensional and nonlinear problems,a multi-fidelity core CFD flow field data reconstruction framework based on data fusion strategy is proposed.The multi-fidelity flow field data reconstruction framework improves the accuracy and flexibility of the reduced-order model in core flow field reconstruction by integrating multisource data.Combined with the multi-fidelity flow field data reconstruction framework,a feature channel flow field reconstruction scheme is proposed,which can realize the expansion of sample data and the reduction of the target data dimension of reconstruction without increasing the computing cost.(3)Aiming at the issue of super-scale flow field CFD data reconstruction,this paper investigated the flow state characteristics of different rod bundle scale domains,analyzed the rules of flow state evolution with the scale,and clarified the differences and correlations between different fidelity models in super-scale calculations.Based on this,a super-scale flow field data reconstruction scheme based on transfer learning was proposed.In this scheme,the transfer of features and feature expression is realized through the feature channel,and the transferred features are matched by the transfer component analysis method.During the testing process,the proposed super-scale flow field reconstruction scheme is capable of reconstructing large-scale flow field data with a certain degree of accuracy by utilizing small-scale highfidelity data and large-scale low-fidelity models.This is particularly useful when a small number of large-scale watershed samples are missing or when only a limited number of them are available.The reconstructed data approximates high-fidelity CFD data,making the scheme effective in producing accurate results despite the limited amount of data available.(4)A study on the refined Neutronics and Thermal-Hydarulic coupling simulation is carried out.According to the grid distribution difference and boundary parameters of the CFD and neutron transport calculation program,the grid mapping and data transfer scheme were established,and the refined physical thermal coupling calculation model was developed.Additionally,research was conducted on the iterative calculation process between the coupling programs.And an online multi-fidelity model(OMFM-N/TH)was proposed for physicalthermal coupling calculations,which can replace high-fidelity CFD models with data-driven models after a limited number of high-fidelity iterative calculations,thus improving coupling calculation efficiency.Also,research was conducted on the reconstruction approach for temperature fields and proposed a Two-step CFD temperature field data reconstruction scheme(TRI)based on the energy equation.This approach solves the initial negative training issue caused by initial field divergence from the convergent field.(5)A transient segmented CFD calculation scheme with an overlapping region is proposed.Additionally,research on the data reconstruction of CFD for the reactor core is carried out and proposing a transient data reconstruction model based on multi-fidelity models and LSTM networks.Compared with traditional segmented solving schemes,the proposed transient segmented CFD calculation scheme optimizes the data transfer scheme,and the calculation results are more consistent with the overall CFD calculations in both numerical and transient trends.The proposed transient data reconstruction scheme is tested in step changes and continuous operating conditions.The test results show that both the incremental and parallel schemes proposed can accurately reconstruct transient CFD data with a small number of samples.Moreover,because the reconstruction scheme requires only a low-fidelity model for prediction,it has extremely high computational efficiency compared to the high-fidelity CFD calculation process.