单目标与多目标的全局优化算法在新安江模型参数率定中的耦合应用研究

Application of coupling global optimization of single-objective algorithm with multi-objective algorithm to calibration of Xinanjiang model parameters

基于已有的研究成果,对利用SCE-UA算法优化新安江模型参数做了一些相关研究,发现目标函数的选取、参数的优化范围以及资料的长度均会对算法最终的寻优结果产生一定的影响,并且异参同效等原因使得最优解并不能稳定唯一。本文选择了呈村,胡乐,东湾等流域作为研究流域,旨在通过"降维"的思想经验性地固定一些参数,以期能使寻优结果稳定,得到唯一的最优解;此外,将多目标优化算法NSGAII应用到新安江模型参数优化中,并对其优化结果做了一些必要的讨论与分析。最后,在利用SCE-UA优化全部参数得到多组参数解的基础上,将两种算法耦合,提出以Pareto前沿确定的径流流量预报区间作为实测流量过程线之外衡量SCE-UA参数解可靠性的又一标准,提出以"不可靠度"量化这一标准,以期能够从中选取一组参数解供实际使用。

This paper presents a study on application of SCE-UA algorithm to optimization of the Xinanjiang model parameters based on the previous studies. We reveal that the optimized solution depends on objective function, parameters range, and data series length, and that the existence of equifinality makes the solution unstable and hence the optimized parameter values not unique. This study selected the basins of Chencun, HuLe and Dongwan as the study area and tried to fix some parameters by reducing dimension and using empirical knowledge, so as to stablize the optimal results and obtain a unique solution. Then, a multi- objective optimization algorithm NSGAII was applied to calibration of the Xinanjiang model. The optimization results are analyzed and discussed in this paper. Finally, we combined these two algorithms based on the SCE- UA solutions obtained and put forward another standard for measuring the reliability of SCE-UA parameter solutions. As an alternative to the prediction interval determined by Pareto front, this new standard, called ＇ unreliable criterion＇ , should provide a group of parameters applicable to practical modeling.