Study On Acceleration Techniques Of Matrix MOC And 3-D Neutron Transport Calculation Method

Currently,the third generation numerical methods about the reactor core are studied worldwide,and the direction of these studies is to instead the existing computation mode,namely transport calculation of assembly and diffusion calculation of core,by the transport calculation of core.With the development of computer technology and numerical method,the 3-D transport calculation of core becomes possible.However,the configuration and design of core are becoming more and more complicated,and there still are many challenges associated with the computation efficiency and accuracy.Based on the 2D/1D coupling mode,the current work has developed a numerically efficient 3-D transport code.The details about the current work are listed as following:First of all,the step characteristic,the diamond difference scheme and the linear source approximation are applied in the method of characteristics with matrix form(matrix MOC),respectively.Their numerical efficiency and accuracy are studied and compared with each other,the results show that the matrix MOC with step characteristic scheme is the most efficient with acceptable numerical accuracy.The only limitation of the MOC method is the capability of the geometry processing module,hence the current work developed a powerful geometry processing module based on the CSG formula,and the tracing of characteristic line has been paralleled based on the OpenMP mechanism.In order to provide the evidence for users to judge the correctness of geometry processing and transport calculation,the current work develops a visual module based on the scanning-line-filled pixel visualization technology.The 2-D HTTR problems and C5G7 benchmark problem are simulated,the results demonstrate that the current code has the acceptable numerical accuracy compared with other international famous codes.Secondly,the computation efficiency of code has been improved from three aspects.To improve the efficiency of GMRES(m)which was adopted to solve the resulted sparse linear systems,the following preconditionning technologies have been studied in the current work: the preconditioner based on the typical iteration method,the augmented subspace preconditioner and the ILU preconditioner.In addition,the physical implication of Jacobi preconditioner was given in current paper.The long characteristic line results in the difficulty of matrix MOC to handle the large-scale core,the current work proposes a compensation scheme,namely track truncation method.This method is equivalent theoretically to the original matrix MOC completely,however,it can reduce the memory requirements and solution time dramatically,and it results in an ideal linear scaling relationship between the physical problem size and the computational requirements needed to solve the problem.Benefiting from the development of GPU technology,the solution of sparse linear systems has been paralleled based on the CUDA programming.Starting from the coherent accessing and shared memory technique,two optimal parallel strategies are proposed to perform the accumulation operations of a row vector within a thread and threads block,respectively.The numerical results relevant to the HTTR problem and C5G7 benchmark problem indicate that the optimal parallel strategy can achieve the speedup of more than 5 times.Finally,with the framework of 2D/1D coupling method,the 3-D whole-core transport code has been developed based on the efficient 2-D matrix MOC solver.In the 3-D transport code,the axial solver is based on the discrete ordinate method(Sn)in theory,and the isotropic approximation are made for both the radial leakage and axial leakage terms.With the acceleration of the 3-D CMFD method,the 3-D transport code can achieve an acceptable efficiency.The 3-D C5G7 benchmark problem and its extension versions have been simulated to verify the code,and the results show that the current 3-D transport code has the capability to model the whole core.