摘要
提出了一种改进的自适应重要抽样方法,以广义极值分布为例,引入L-矩法,建立样本统计特性与分布参数的联系,估算极限事件的发生概率.以浙江省云港流域的24h设计暴雨为例,计算金竹岭和仙人潭两个站点降雨量分别大于213mm和200mm的概率.计算结果表明改进的自适应重要抽样方法能很好地模拟水文极限事件,叠代次数随着抽样个数的增加逐渐减小.与常规的MC法比较,重要抽样的效率有显著提高.另外,此改进的自适应重要抽样方法还能推广到其他的分布函数.
This paper proposes an improved adaptive importance sampling method that can be used for variables following a GEV distribution. The L-moment method is employed to build the relationship between sample statistics and distribution parameters. In the case study, the 24-hour design storm of Yungang Basin in Zhejiang Province is considered, and the probability of extreme precipitation is calculated for two rainfall gauge stations, Jinzhuling and Xianrentan. The results show that the proposed method performs well in simulating extreme rainfall, and the iterative number decreases as the number of samples increases. Compared with traditional MC simulation, the improved adaptive importance sampling method has better efficiency. Besides, this method has huge potential to be used for other distributions.
出处
《系统工程理论与实践》
EI
CSSCI
CSCD
北大核心
2012年第10期2345-2350,共6页
Systems Engineering-Theory & Practice
基金
国家自然科学基金(50809058)
科技部国际科技合作计划(2010DFA24320)
教育部博士点基金项目(200803351029)
关键词
改进自适应重要抽样法
广义极值分布
水文极限事件
L-矩法
improved adaptive importance sampling
GEV distribution
extreme hydrological events
L-moment method
