中国洪水预报系统方案参数优选加速方法研究

Acceleration methods for parameter optimization functions of national flood forecasting system(NFFS)

为有效缩短中国洪水预报系统方案参数优选耗时,选择东辽河支流十屋流域建立三水源新安江模型预报方案,挑选历史最大10场次洪用于参数率定研究。中洪系统利用…\\Temp目录下的文件作为模块之间、模块与系统之间的通讯,参数优选时频繁的文件读写消耗了大量的时间,本文据此设计了三套解决方案:换用固态硬盘和使用虚拟内存盘方案旨在提高中洪系统的文件读写效率,借用河海大学HMCE系统用于测试整个数据流优化之后的效果。当用全局参数优选的基因法重复运行20次进行参数率定时,上述四种方法的平均耗时依次为230 min、55 min、19 min、4.4 min;当用局部参数优选的单纯形法时,平均耗时依次为185 min、45 min、17 min、4.1 min。研究成果可以有效缩短预报方案的参数优选时间,对提高中洪系统的应用效率、研发相关软件等具有重要的应用价值。

In order to effectively shorten the time-consuming optimization of scheme parameters of National Flood Forecasting System(NFFS),this paper chooses Shiwu watershed,a tributary of the East Liaohe River,to establish a three-source Xin′anjiang model for forecasting,and chooses the 10 largest secondary floods in history for parameter calibration.NFFS uses files in the directory named“…\Temp”as communication between modules and between modules and systems,frequent reading and writing of files in parameter optimization consumes a lot of time.On these grounds,this paper designs three solutions:In order to improve the file reading and writing efficiency of NFFS,we switch to Solid State Disk and use virtual memory.Then we use the HMCE system of Hohai University to test the effect of the whole data flow optimization.The average time consumed by the four methods was 230 min,55 min,19 min and 4.4 min respectively when the gene method for global parameter optimization is repeated 20 times for parameter calibration;when the simplex method of local parameter optimization is used,the average time consumed is 185 min,45 min,17 min and 4.1 min respectively.The research results can effectively shorten the time of parameter optimization of forecasting schemes,and have important application value for improving the application efficiency of NFFS and related software development.