随机梯度法在新安江模型参数优选中的应用

基于近似梯度法及模式搜索法，提出了复合两种方法的随机优化方法。以Nash确定性系数为目标函数，对新安江模型的参数空间随机搜索后运用梯度法进行了优化，然后采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，使优化方法脱离局部极小解从而达到近似全局最优解集。以杨楼单元流域为应用实例进行了研究，结果表明，随机梯度法可以成功地率定概念性水文模型参数。

Application of stochastic gradient algorithm to parameter optimization of Xinanjiang Model

Based on the approximate gradient-based steepest descent algorithm and the pattern search algorithm, a stochastic optimization algorithm compounded of both the algorithms is proposed herein. By taking Nash deterministic coefficient as the target function, an optimization is made on Xinanjiang Model after the random search of its parameter space, and then the parameter space selection strategy is adopted for getting globally optimal solution set. The above-mentioned method integrates with the algorithms of derivative information and stochastic properties to get the optimization set away from the local minimum solution, thereby reaching the approximate globally optimal solution set. The relevant study is made by taking the representative elementarywatershed controlled by Yangluo Hydrological Station as an application case and then the result shows that the parameters of the conceptualhydrologicalmodel can be successfully calibrated by the stochastic gradient algorithm.