Research On2D/1D Coupled3D Whole-core Transport Calculations Based On Parallel Computation

Traditional reactor physics analysis theory based on generalized equivalencehomogenization and nodal methods adopts much approximation, and it’s incompetentfor design of complex reactors. With the continuous advance of computing technology,the direct3D whole core transport calculations involving no a priori spacialhomogenization has been much attractive these years. This thesis focuses on theresearch of2D/1D coupled direct whole-core transport calculations based on large-scaleparallel computation. The research contents are shown below.Firstly, Matrix MOC based on modular ray tracing was studied. Modular raytracing reduces memory demand of geometric information significantly. With suitablearrangement of characteristics lines in modular units introduced by symmetry, methodfor symmetric cores was proposed. The numeric properties of the coefficient matrixwere dug out by analysis of the construction process. Taking advantage of the numericproperties, constructing computation and memory demand were reduced significantlyand efficiency of solving the Matric MOC equations was improved.Secondly, multi-group GMRES was studied based on Matrix MOC and its numericproperties. Compared with traditional Gauss-Seidel iterations, multi-group GMRES ismuch faster in case of upscatter. Combined with multi-group GMRES, Wielandtiteration and IRAM were studied for solving the critical problems, and better efficiencywas obtained than traditional power iteration.Thirdly, spacial domain-decomposed parallel Matrix MOC and relevant coarsemesh finite difference (CMFD) method were studied. Multi-domain coupled PGMRESbased on PETSc library was proposed for directly solving the domain-decomposedMatrix MOC equations. Convergence of iterations of angular flux of inner boundarieswas improved. CMFD acceleration method was adopted for domain-decomposedMatrix MOC, and better efficiency was obtained.Finally,2D/1D coupled direct whole-core transport method based on parallelcomputation was studied within the framework of3D CMFD. In the radial direction,spacial domain-decomposed parallel Matrix MOC was selected. In the axial direction,finite-difference diffusion was used. Large-scale parallel computation was realized by combining parallel feature of both directions, with MPI communicator to handle themessage passing.According to the theory above, the code Tiger-3D for whole-core transportcalculations was developed. Large-scale parallel computation was achieved in Tiger-3D,in which axial planes were solved independently and each radial plane was solved bydomain-decomposed Matrix MOC code Tiger-2D. Numerical results demonstrated thatTiger-3D has good accuracy, high efficiency and great scalability, thus importantengineering practicability.