随机GP法在概念性水文模型参数优选中的应用

Application of Stochastic GP Algorithms Optimization to Conceptual Hydrologic Model Parameters

在基于近似梯度及模式搜索法的基础上,提出了复合上述两种方法的GP局部优化方法。以Nash确定性系数为目标函数,对水文模型的参数空间随机搜索后采用GP方法优化,运用参数空间筛选策略,以获得全局最优解集。上述方法结合导数信息和随机性质的算法,使优化过程脱离局部极小解从而达到近似全局最优解集。杨楼单元流域应用新安江模型的实例研究结果表明,随机的GP优化方法可以成功的率定概念性水文模型参数。

Combining the approximate gradient-based steepest descent algorithm and the pattern search algorithm, the GP algorithm, a new local optimization algorithm for conceptual hydrologic model parameters is presented. With Nash facticity coefficient as the target function the random search techniques is used for searching parameter space, then optimize the selected parameter set using GP algorithm. The global optimization parameter is achieved by filtering parameter space strategy. The above-mentioned method comprise the derivative information and stochastic properties, make the optimization set escaping the local maximum to the global set. The practical efficiency is verified by using a case in YandLou unite drainage basin. It is shown that parameters of hydrologic model can be automatically calibrated successfully.