非饱和土壤水运动模型参数优选方法研究  

Parametric optimization of unsaturated soil hydraulic movement model

作　　者：饶元[1,2] 许文俊[1] 江朝晖[1] Naftali Lazarovitch 李绍稳[1] 

机构地区：[1]安徽农业大学信息与计算机学院,安徽合肥230036 [2]本古里安大学旱地农业系,以色列斯代博克84990

出　　处：《浙江农业学报》2016年第12期2082-2089,共8页Acta Agriculturae Zhejiangensis

基　　金：国家自然科学基金资助项目(61203217;61402013);农业部引进国际先进科学技术948项目(2015-Z44;2016-X34);安徽省自然科学基金项目(1608085QF126);国家留学基金(201308340014)

摘　　要：针对数值反演过程中参数优选方法适用性不明确的问题,研究最小化目标函数、带有自适应差分演化Metropolis的马尔可夫链蒙特卡罗方法(MCMC-DREAM)对数值反演效果的影响,为探索高效的参数优选方法提供参考。数值算例反演结果表明:最小化目标函数方法计算复杂度较低,但对参数初值敏感,适合于对目标区域土壤有较为深入了解的场合使用;MCMC-DREAM对参数初值不敏感,但计算复杂度较高,适合于先验信息有限的场合使用。两种参数优选方法都存在"异参同效"现象,先验信息与敏感性分析有助于克服该问题,提高数值反演解决实际问题的能力。In order to explore the suitability of parametric optimization methods,the influence of 2 typical parametric optimization methods,namely Minimizing the Objective Function（MOF）,and Markov Chain Monte Carlo with DiffeRential Evolution Adaptive Metropolis algorithm（MCMC-DREAM）,on numerical inversion performance was evaluated to offer suggestions for further exploring more efficient parametric optimization methods. Numerical case study showed that MOF had lower computation complexity,however,higher sensitivity to the initial solution. Therefore,MOF was suitable for conducting parametric optimization in the case of enough prior information. In contrast,MCMC-DREAM was insensitive to the initial solution,but took longer time to complete computation. As a result,MCMC-DREAM was suitable for conducting parametric optimization in the case of limited prior information. Both optimization methods suffered equifinality. However,both methods＇ ability to solve the practical problems could be improved by overcoming the equifinality drawback with adequate prior information and sensitivity analysis.

关 键 词：非饱和土壤 水运动模型 数值反演 参数优选 

分 类 号：S24[农业科学—农业电气化与自动化] TP391[农业科学—农业工程]