基于高性能计算的SWAT参数敏感度分析并行框架  

Parallel framework for SWAT sensitivity analysis based on high performance computing

作　　者：李强[1,2,3] 陆忠华[1] 王彦棡[1] 陈曦[2] 罗毅[2] 

机构地区：[1]中国科学院计算机网络信息中心,北京100190 [2]中国科学院新疆生态与地理研究所,乌鲁木齐830011 [3]中国科学院大学,北京100049

出　　处：《计算机应用研究》2015年第1期41-44,70,共5页Application Research of Computers

基　　金：国家"973"计划资助项目(2010CB951002);中国科学院重点部署项目(KZZD-EW-12)

摘　　要：随着大规模水文模拟需求的不断提高,如何解决计算需求问题逐渐成为水文研究的一个热点。SWAT(soil and water assessment tool)模型在进行大规模水文模拟时有着良好的适应性与准确度,但其敏感度分析模块由于计算量过高,计算时长往往长达数月之久。为了加快SWAT敏感度分析的运行速度,针对SWAT敏感度分析模块的特点,基于MPI提出了一种高效的主—从式并行计算框架,并在此框架的基础上,通过将正演过程并行化,在敏感度分析的主—从并行框架中引入通信子空间的操作,将并行化的正演与主—从式的外层并行框架相结合,得到一种混合式的敏感度分析并行框架,大大提高了对参数集合的敏感度分析速度,将SWAT敏感度分析模块使用的处理器数量从原始的单核串行一跃提升到百核的数量级。最后通过天山北坡流域的模拟验证了此并行框架的可行性。With the development of large scale hydrologic modelling, how to resolve the comPuting capability problem be- comes a hot research topic. SWAT （ soil and water assessment tool） model demonstrates good adaptation and precision in hydrologic modelling： But because of its large computation in sensitivity analysis module, the total time of computation usually lasts for months： To accelerate the efficiency of SWAT sensitivity analysis, according to the characteristics of sensitivity analysis module of SWAT, this paper proposed a high performance master-slave parallel computing framework. Then based on this framework, by parallelizing the forward model,it applied communicator management into SWAT, which largely accelerated the sensitivity analysis of input parameters. It increased the number of processor used by SWAT sensitivity analysis from only oneto hundreds. At last, by the hydrologic simulation of basins in the north of Tian Moutain, the resultvalidates the feasibility of the parallel framework.

关 键 词：SWAT LH-OAT算法 参数敏感性分析 并行计算 通信子空间 主—从并行框架 

分 类 号：TP301.5[自动化与计算机技术—计算机系统结构]