基于人工神经网络的雅砻江流域 水文过程多模型集合模拟

为降低水文模型的不确定性对流域水文过程模拟的影响, 优化模型的实际应用效果, 选取四种常见的水文模型: SWAT 模型、BTOPMC 模型、VIC 模型和 DT VG模型在中国西南的雅砻江流域分别建模, 采用一套统一的模型输入数据与模拟时间范围, 再次运用四个水文模型进行径流计算, 并运用北京师范大学水科学研究院自主开发的基于人工神经网络方法的多模型输出集合系统对四个模型的模拟结果进行集合计算, 得到集合计算的流量过程线及误差水平, 与各水文模型计算结果相比较。研究结果表明, 多模型集合计算的确定性系数和纳什效率系数均达到了0.90, 相比单一水文模型的计算精度有大幅提高, 且计算结果较稳定, 与实际径流过程具有很好的一致性, 说明多模型集合模拟在该流域具有很好的适用性。 In order to reduce the influence of the uncertainty of hydrological models on hydrological simulation and improve the actual application effect of the models, we took the Yalong River basin as an example, and constructed four commonly used hydrological models: SWAT model, BTOPMC model, VIC model, and DTVG model. We conducted independent simulation using these models with the same input data and simulation time range. Then, we calculated the simulation results of the four models using the Multi-model Ensemble Out put System independently developed by Beijing Normal University based on the artificial neural net work method to obtain the flow hydrograph and error, and compared them with the results of the four models. The results indicated that the correlation coefficient and Nash efficiency coefficient of the multi-model ensemble simulation were both above 0.90, which was a great improvement in accuracy than the independent models. The results were stable and consistent with the actual runoff process. These indicated that the multi-model ensemble hydrological simulation had good applicability in this river basin. “十三五”国家重点研发计划项目( 2016YFC 0401308) ; 中央高校基本科研业务费专项资金项目