摘要
为解决大规模地下水流数值模拟参数估值计算时间过长的问题,基于粗粒度任务分解方法,采用消息传递接口（message passing interface,MPI）实现地下水流模拟程序MODFLOW中参数估值灵敏度分析部分的并行。利用多个分区规模的模型验证并行程序的正确性和有效性,对不同参数规模下的加速比情况进行分析,分析结果表明,对16个参数问题进行参数估值,并行程序在1~16个进程下得到了1.74~8.32的加速比,在8进程下取得了5.91的超线性加速,有效提高了地下水流数值模拟参数估值计算速度。
To solve the problem of high computational time cost in numerical simulation of groundwater flow with large scale,and based on decomposition tasks method of coarse-grain parallelism,message passing interface（MPI）was used to realize the parallel sensitivity analysis of parameters of MODFLOW.The accuracy and efficiency of this method was demonstrated through the examples with different parameter scales.The speedup rates from different parameter scales were completely analyzed.The results,which were estimated from 16 parameters of scales,show that the speedup rate ranged from 1.74 to 8.32 can be achieved from 1-16 processes.The superlinear speedup of 5.91 is also got from 8processes.It effectively improves the speed of parameter estimation in numerical simulation of groundwater.
出处
《计算机工程与设计》
北大核心
2016年第1期189-194,共6页
Computer Engineering and Design
基金
国土资源部公益性行业科研专项经费基金项目(201211079-4)
