基于梯度下降法的大房郢水库洪水预报模型参数率定及适用性研究

Paranieter calibration and applicability of flood forecasting model for Dafangying Reservoir based on gradient descent method

大房郢水库长期承担合肥市的防洪、供水任务,准确可靠地开展洪水预报对当地防汛抗旱工作具有重要意义。分别采用新安江模型和前期影响雨量(antecedent precipitation index,API)模型构建大房郢水库入库洪水预报模型,基于梯度下降法对模型参数进行优选,并以方案合格率作为评价指标,对2种模型在该水库的适用性进行分析。结果表明:新安江模型和API模型预报方案合格率均为90.7%,预报精度达到甲等;2种模型在不同预报项目中各有优势,对于洪峰和峰现时差,API模型预报效果更好,预报合格率分别为88.9%和100%,较新安江模型分别提高6.7%和5.9%;对于径流深,新安江模型预报效果更为理想,预报合格率可达94.4%,相对于API模型其提升幅度约13.3%。此外,API模型更适用于单式洪水预报,而新安江模型则更适用于复式洪水预报。研究成果可为提升大房郢水库入库洪水预报精度提供理论依据。

The Dafangying Reservoir has long been responsible for flood control and water supply in Hefei City.The ability to accurately and reliably conduct flood forecasting is crucial for local flood control and drought relief efforts.Both the Xin'anjiang model and the API model were employed to construct inflow flood forecasting models for the Dafangying Reservoir.The model parameters are optimized using the gradient descent method,and the suitability of the two models for the reservoir is analyzed with the qualifica-tion rate of the schemes as the evaluation index.The results show that both the Xin'anjiang model and the API model have a scheme qualification rate of 90.7%,and the forecasting accuracy reaches Grade A.Each model has its advantages in different forecasting aspects.For peak flood and peak occurrence time difference,the API model performs better,with qualification rates of 88.9%and 100%,respectively,which are 6.7%and 5.9%higher than those of the Xin'anjiang model.For runoff depth,the Xin'anjiang model shows a more ideal performance,with a qualification rate of 94.4%,which is about 13.3%higher than that of the API model.Additionally,the API model is more suitable for single flood forecasting,while the Xin'anjiang model is better for compound flood forecasting.The findings of this study provide a theoretical basis for improving the accuracy of inflow flood forecasting in the Dafangying Reservoir.