中山大学学报自然科学版 ›› 2011, Vol. 50 ›› Issue (2): 120-126.

• 研究论文 • 上一篇    下一篇

基于SCEM-UA算法和全局敏感性分析的水文模型参数优选不确定性研究

曹飞凤1,2,张世强3,许月萍2,楼章华2   

  1. (1.浙江省水利水电工程局,浙江 杭州 310020;2浙江大学 建筑工程学院水文与水资源工程研究所,浙江 杭州 310058;3中国科学院寒区旱区环境与工程研究所,甘肃 兰州 730000)
  • 收稿日期:2010-03-29 修回日期:1900-01-01 出版日期:2011-03-25 发布日期:2011-03-25
  • 通讯作者: 张世强

Parameter Optimization of Hydrologic Model Parameters based on Regional Sensitivity Analysis and SCEMUA Algorithm

CAO Feifeng1,2,ZHANG Shiqiang3,XU Yueping2 ,LOU Zhanghua2   

  1. (1.Bureau of Water Resources & Hydropower Engineering of Zhejiang Province,Hangzhou 310020,China;2Institute of Hydrology and Water Resources, Department of Civil Engineering, Zhejiang University, Hangzhou 310058, China;3Cold and Arid Regions Environmental and Engineering Research Institute, Chinese Academy of Science, Lanzhou 730000,China)
  • Received:2010-03-29 Revised:1900-01-01 Online:2011-03-25 Published:2011-03-25

摘要: 针对复杂非线性水文模型参数识别及不确定性分析问题,引入基于马尔可夫链蒙特卡罗思想的SCEM-UA算法,以岷江流域为研究实例,对降雨径流模型的参数优选问题进行了分析,并探讨了该算法在推求参数后验分布的搜索性能和效率。结果发现,SCEM-UA算法能快速有效地推求出参数后验概率分布。同时,开展基于SCEM-UA算法取样的参数全局敏感性分析,对比参数敏感性和后验分布,表明两者密切相关,敏感性强的参数其边缘后验概率密度分布存在明显峰值,相反,敏感性弱的参数其后验概率分布较为平坦且无规律可循,从而导致模型参数的不确定性大大增强。

关键词: Markov链蒙特卡罗法, 参数优选, SCEM-UA, 敏感性分析, 不确定性分析, 降雨径流概念模型

Abstract: Shuffled Complex Evolution Metropolis Algorithm(SCEM-UA) is an adaptive Markov Chain Monte Carlo sampler, which can be applied to parameter optimization of nonlinear hydrologic model and uncertainty analysis The efficiency and effectiveness of SCEMUA for sampling the posterior distribution of model parameters are discussed based on the case study of the Min River catchment The results show that SCEMUA algorithm is consistent, effective and efficient in inferring the parameter posterior distribution Moreover, the results of regional sensitivity analysis using samples from SCEMUA algorithm sampler show that sensitivity and posterior distribution of parameters are highly interdependent High sensitive parameters correspond with distinct peak in posterior distribution, while low sensitive parameters correspond with flat posterior distribution which could highlight the uncertainty of model parameters.

Key words: Markov Chain Monte Carlo, parameter optimization, Shuffled Complex Evolution Metropolis Algorithm, sensitivity analysis, uncertainty analysis, conceptual rainfallrunoff model

中图分类号: