随着球床高温气冷堆的不断发展和建成运行，物理设计对该堆型的中子模拟提出了愈来愈高的要求。三维特征线方法（3-D MOC）是一种适用于复杂几何的确定论方法，在轻水堆中已有成功应用先例，但是针对球床高温气冷堆的随机球床复杂几何，应用三维特征线法仍面临着计算时间和内存的巨大挑战。为了将三维特征线法高效地用于高温气冷堆全堆模拟，本课题深入研究了三维特征线法的加速算法，最终独立自主地开发了具备大规模并行能力的三维特征线程序ARCHER（Advanced Reactor CHaracteristics tracER）。本课题重点研究了加速三维特征线法的粗网有限差分方法（CMFD）。低阶CMFD系统通常基于规则粗网格几何构建，在球床高温气冷堆复杂几何下应用CMFD加速首先需要建立粗网格与细网格相容的嵌套网格。课题使用基于半空间与实体布尔运算的构造实体几何方法，在球床高温气冷堆中成功构建了复杂的粗网格-细网格嵌套几何。另外，课题使用树状网格数据结构降低射线追踪的时间复杂度，同时也便于获得粗网格与其内部细网格的拓扑映射关系。课题使用了多群与单群两级CMFD加速MOC计算，降低多群CMFD的求解代价。此外，本课题还研究了加速特征线法收敛的JFNK（Jacobian Free Newton-Krylov）方法，结合CMFD进一步提出了CMFD-JFNK混合的非线性加速方法。本课题开发了可应用于复杂几何的线性源近似（LSA）方法，降低了特征线法的计算时间与占用内存。并且，课题通过在线计算射线段中心进一步降低了LSA下的射线信息内存。结合优化的输运扫描方案，课题避免了射线段中心相对位置的大量重复计算，大幅减少了输运扫描时间。为了计算真实高温气冷堆的大规模问题，ARCHER程序使用两级MPI-OpenMP混合并行策略。第一级并行采用MPI空间区域分解，每一进程只在各自子域内执行计算任务。课题采用角度、射线和粗网格分组并行，分别实现了射线追踪、射线扫描以及CMFD线性预处理模块的第二级OpenMP并行。课题通过一系列基准题测试验证了程序实施的正确性，和高效的加速效果，以及最终能求解HTR-10和HTR-PM这样真实高温气冷堆模型的能力。
With the continuous development and operation of the pebble-bed High Temperature gas-cooled Reactor (HTR), the physical design puts forward higher requirements for the neutronic simulation of pebble-bed HTRs. The 3-D MOC (Method Of Characteristics) is an efficient deterministic method applicable to complicated geometries, which has been successfully applied in light water reactors. However, the application of the 3-D MOC in the pebble-bed HTRs with randomly distributed pebbles still faces huge challenges in computing time and memory. In order to use the 3-D MOC efficiently in the whole-core simulation of pebble-bed HTRs, this thesis deeply studied the acceleration methods for the 3-D MOC, and finally independently developed the 3-D MOC code ARCHER (Advanced Reactor Characteristics tracER) with large-scale parallel capability.This thesis focuses on the use of the CMFD (Coarse Mesh Finite Difference) to accelerate the 3-D MOC transport calculation. The low-order CMFD system is usually constructed based on the regular coarse mesh geometry. The application of CMFD acceleration in a random pebble-bed HTR requires the construction of a nested grid system compatible with fine meshes and coarse meshes. CSG (Constructive Soild Geometry) is used to describe the complex nested grid structure, which is further based on the Boolean operations of the half-spaces and the solids. In addition, the tree grid data structure is implenmented in the code to reduce the time complexity of the ray tracing, and it is also convenient to obtain the topological mapping relationship between the coarse meshes and its internal fine meshes. In this thesis, the two-level CMFD are used to accelerate the MOC transport sweepings and reduce the cost of solving the multi-group CMFD. Moreover, this thesis also studies the JFNK (Jacobian Free Newton-Krylov) method to accelerate the convergence of MOC, and further proposes the CMFD-JFNK hybrid nonlinear acceleration method combined with CMFD.The LSA (Linear Source Approximation) that can be applied to complex geometries is developed in this thesis in order to reduce the calculation time and memory consumption of the MOC. Moreover, through calculating the ray segment centers online, this thesis further reduces the ray information memory under LSA to an acceptable level. Combined with the optimized transport sweeping algorithm, this thesis avoids a large number of repeated calculations of the ray segment centers, and greatly reduces the transport sweeping time.In order to simulate large-scale problems, the ARCHER code uses a two-level MPI-OpenMP-based hybrid parallel strategy. The first parallel uses MPI spatial domain decomposition, and each processor only performs computing tasks in its own sub-domain. The second level OpenMP parallel of the ray tracing, transport sweeping and CMFD linear preconditioning is realized by using angle, ray and coarse mesh grouping.Through a series benchmark calculation, and real case simulation, the correctness of the code implementation was proven, the efficiency of acceleration on MOC method was also shown, and the capability on the MOC for real reactor such as HTR-10 and HTR-PM was finally demonstrated.
