论文部分内容阅读
在解决复杂化工过程优化与模拟问题时,大规模代数差分方程的存在导致大量的计算时间花费在重复求解稀疏大型线性方程组的过程中。随着并行计算和网络技术的发展,为了提高优化或模拟工作的速度,可以通过将非对称矩阵重排为带边块对角形式,从而实现对线性系统的高效并行求解。本文提出一种基于Kernighan-Lin算法的并发式的多层次矩阵重排策略,它以最小化边块为目标,同时保证尽可能小的负荷非平衡性,从而获得好的重排结果。应用该重排策略可以对大型稀疏矩阵进行压缩和并行重排,提高重排算法的效率。在研究过程中应用了基于该技术的并行计算程序对一系列标准矩阵进行了检验,并与一些现有的算法进行了比较,证明了其有效性和可行性。
In solving complex chemical process optimization and simulation problems, the existence of large-scale algebraic difference equations results in a large amount of computational time spent in the process of repeatedly solving sparse large-scale linear equations. With the development of parallel computing and network technology, in order to improve the speed of optimizing or simulating work, an efficient and parallel solution to a linear system can be achieved by rearranging asymmetric matrices into diagonal forms with edge blocks. This paper proposes a concurrency-based multi-level matrix rearrangement strategy based on the Kernighan-Lin algorithm, which aims at minimizing the edge blocks while ensuring the least possible load imbalance, resulting in good rearrangement results. The rearrangement strategy can compress and rearrange the large sparse matrix to improve the efficiency of the rearrangement algorithm. In the process of research, a series of standard matrices were tested by using the parallel computing program based on this technique. Compared with some existing algorithms, it proved its effectiveness and feasibility.