探讨了一种适用于我国自主研发的数值天气预报模式软件GRAPES的不完全LU(ILU)分解预条件子。针对GRAPES模式所特有的具有对角优势结构的赫姆霍兹方程系数矩阵,提出了一种有效的ILU分解方案,并将分解得到的预条件子应用到模式核心的动力积分计算迭代算法中,从而达到加速算法收敛,提高模式软件整体性能的目的。
How to design the incomplete LU (ILU) decomposition preconditioner for the domestic weather prediction model GRAPES (global/regional assimilation and prediction enhanced system) is discussed. According to the special no-zeros structure matrix derived from the Helmholtz equation, an effective ILU decomposition method is presented. The resulting preconditioner is applied to the iterative algorithm in GRAPES, and thus speeds up the convergence of the algorithm and enhances the overall performance of the model software .