基于EFEN-SP_3方法的高性能全堆芯Pin-by-pin计算研究

High-Performance Whole Core Pin-by-Pin Calculation Based on EFEN-SP_3 Method

基于指数函数展开节块3阶简化球谐函数(EFEN-SP3)方法,通过采用基于标准消息传递界面(MPI)的空间并行算法实现高性能全堆芯Pin-by-pin计算,并开发了相应的程序EFEN。该程序通过合理设计区域划分方案以保证负载平衡并使通信次数最小化,充分发挥并行中央处理器(CPU)的计算和存储能力;通过选择红黑Gauss-Seidel节块扫描算法避免区域分解引起的迭代格式退化。参考实际商用堆的堆芯布置,设计2个压水堆(PWR)全堆芯Pin-by-pin算例,相应的数值结果表明:该程序计算结果的精度在可接受范围内;通信周期对计算精度和并行效率的影响都很小;子区域表面体积比较小的区域划分方式具有较高的并行效率;用125个CPU进行一次空间网格数为289×289×218、能群数为4的PWR全堆芯Pin-by-pin计算所需时间约为900 s,并行效率约为90%。

The EFEN code for high-performance PWR whole core pin-by-pin calculation based on the EFEN-SP3 method can be achieved by employing spatial parallelization based on MPI. To take advantage of the advanced computing and storage power, the entire problem spatial domain can be appropriately decomposed into sub-domains and then assigned to parallel CPUs to balance the computing load and minimize communication cost. Meanwhile, Red-Black Gauss-Seidel nodal sweeping scheme is employed to avoid the within-group iteration deterioration due to spatial parallelization. Numerical results based on whole core pin-by-pin problems designed according to commercial PWRs demonstrate the following conclusions： The EFEN code can provide results with acceptable accuracy; Communication period impacts neither the accuracy nor the parallel efficiency; Domain decomposition methods with smaller surface to volume ratio leads to greater parallel efficiency; A PWR whole core pin-by-pin calculation with a spatial mesh 289×289×218 and 4 energy groups could be completed about 900 s by using 125 CPUs, and its parallel efficiency is maintained at about 90%.