基于大规模并行的数值反应堆高保真三维耦合输运计算方法研究

Sustainable development of nuclear energy is very essential to the energy security, national safety and diplomacy. In order to satisfy the requirement of safety and efficiency of nuclear usage, numerical reactor technology which applies the state-of-the-art methods to reduce the approximations and adjustments during the calculations, can represent the real physical reactions in the core and realize the accurate analysis and prediction. However, the developing 3D transport methods still need massive computations and the parallel efficiency is still low, which is still a big issue for the numerical reactor technology. This proposal is dedicated to the development of the massive parallel based high-fidelity 3D transport method. A new 3D coupling method will be proposed, which will combine the advantages of both the efficiency of the pin-by-pin method and the accuracy of the 2D/1D coupling method. Then we will study the massive parallel method for the new coupling method; analyze the principal factors and their mathematical relationship to the parallel efficiency, then try to find the optimal parallel method. It is believed that this fundamental cutting-edge investigation will not only has significant value on improving the numerical methods based on the massive parallel for 3D transport calculations, but also provides the instructions applying massive parallel to the numerical reactor technology.

核能的可持续发展对我国的能源保障、国家安全以及大国外交至关重要。为了满足对核能安全高效利用的要求，数值反应堆技术通过最先进的方法理论，最大程序地减少计算中的经验近似和人为修正，还原堆芯实际的物理过程，实现核反应堆的精确分析和预测。但目前研究中的三维输运计算方法计算量大、大规模并行的并行效率依然较低，难以满足数值反应堆高保真计算的要求。本研究拟基于高效的大规模并行算法，开展高保真的全堆芯三维耦合输运计算方法研究，具体包括：结合pin-by-pin输运方法高效率和二维/一维耦合方法高精度的优势，提出新的三维耦合输运方法；研究三维耦合输运计算的大规模并行方法；分析影响并行效率的主要因素，明晰不同因素对并行效率影响的数学关系，建立大规模并行的优化方法和实现方式。本研究不仅对基于大规模并行的三维输运计算方法的理论完善有重要的学术价值，并且对推动超算在数值反应堆技术中的应用具有重要的指导意义。

数值反应堆技术通过最先进的方法理论，还原堆芯实际的物理过程，实现核反应堆的精确分析和预测，为核反应堆的安全高效提供技术基础。本项目首先建立了一种新的基于 MOC 的三维耦合输运方法，构建了以径向 MOC 输运计算、三维 SN 输运计算和粗网有限差分计算相互耦合的2D/3D耦合输运计算模型，并开发了基于 MOC 的三维耦合输运计算程序。针对C5G7 2005 roddedB基准问题， 2D/3D耦合方法采用 4 层轴向划分即可获得较高精度的结果，其中 keff 偏差 50 pcm，最大棒功率偏差 0.01%，组件功率偏差都小于 0.12%。与2D/1D 耦合方法18层计算结果的精度一致。在三维耦合方法的基础上，实现了空间、角度区域分解和特征线并行的三重方法，并分别针对特征线和CMFD计算的并行任务分配、扫描和通信进一步优化，显著提高了计算效率。 针对全堆芯问题可以实现万核以上的并行计算，其中空间并行10000核相对于100核并行效率超过70%，角度并行16核并行效率接近90%，特征线OpenMP并行10线程并行效率接近80%。 将全堆芯输运计算由4000核时降低至150核时，达到国际同类程序的先进水平。为我国数值反应堆技术的多物理耦合奠定了坚实的基础。