论文部分内容阅读
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 view of the ambiguity of the applicability of the parameter optimization method in numerical inversion, the influence of minimized objective function, Markov chain Monte Carlo with adaptive differential evolution Metropolis (MCMC-DREAM) on numerical inversion is studied , To provide reference for exploring efficient parameter optimization methods. The numerical results show that the method of minimizing the objective function is computationally complex but sensitive to the initial values of the parameters, and is suitable for situations where the soil in the target area has a better understanding. MCMC-DREAM is insensitive to initial parameters, But the computational complexity is high, suitable for occasions with limited prior information. Both methods of parameter optimization have the phenomenon of “homogeneous and homogeneous”, priori information and sensitivity analysis can help to overcome this problem and improve the ability of numerical inversion to solve practical problems.