SCE-UA算法优化土壤湿度方程中参数的性能研究

借助于一维土壤湿度模型,分别将土壤成份和土壤性质相关参数作为待优化的参数,通过观测系统模拟试验的方式,评估SCE-UA (Shuffled Complex Evolution Algorithm) 优化算法对这些参数的优化效果。结果表明:优化的效果不仅依赖于参数的取值范围,还依赖于参数的敏感性,敏感的参数通过优化算法易得到最优值;不敏感的参数存在“不敏感区间”,在“不敏感区间”中易陷入次优,通过缩小参数优化分布区间和增加优化的次数可以部分提高优化的效果。此外,模型的超定性也可能导致参数次优值的出现,而通过恰当地给出参数之间的约束条件和优化判据,可以提高参数优化的效果。

Calibration of Parameters in Soil Moisture Equation with Shuffled Complex Evolution Algorithm

In this study, by using an observing system simulation experiment, a shuffled complex evolution algorithm (SCE-UA) is evaluated in terms of the effectiveness and efficiency of calibrating parameters in one-dimensional Richards equation, including soil components and all physical parameters. The result shows that the ability of calibrating parameters with SCE-UA depends on not only the uncertainty ranges of parameters but also their sensitivity degrees. For sensitive parameters, unique optima parameter estimates can be easily obtained. However, for insensitive ones, there exists an "insensitive range", and only suboptimal parameters are obtained in this range. By increasing training times and reducing the parameters' range, the performance of parameter calibration in the insensitive range can beimproved. Moreover, the overdetermination of the model parameters may result in suboptimal parameter estimates, and a good calibration effectiveness can be archived by appropriately setting parameters and adding constrains and criterions.