高效高精度水动力模型在洪水演进中的应用研究

针对现有的洪水演进及淹没过程二维数值模型中存在的计算精度不高、稳定性差、计算效率低下等问题，本文提出一套基于GPU加速技术的地表水动力数值模型（GAST）。选用英国莫帕斯小镇百年一遇洪灾为研究对象，模拟5 m和2 m分辨率地形下的城区淹没过程及范围，分析GAST模型的模拟精度、效率及稳定性。研究发现：GAST模型模拟淹没面积与实测结果百分误差仅为0.96% ~ 4.36%，同时计算5 m、2 m分辨率地形用时分别为2.17 h、9.00 h，具有较高的模拟精度、效率及计算稳定性；对比相同输入条件下的MIKE21 FM模型，GAST模型计算精度提升1.07 ~ 19.55倍；GAST模型计算效率提升3.4 ~ 11.78倍，且随着计算网格数的增加效率优势更为明显，故其应用前景广阔。

Application of efficient high-resolution hydrodynamic model to simulations of flood propagation

To improve the accuracy, numerical stability and computational efficiency of the existing 2D numerical models of flood propagation and inundation, this paper presents a full hydrodynamic model based on the GPU acceleration techniques (GAST). The model is applied to reproduce a flood event with a return period of 100 years that occurred in Morpeth town, UK. Two grid resolutions of 5 m and 2 m are used in the model and the MIKE21 FM model to investigate different inundation patterns. It is found that the GAST model can achieve an error range of only 0.96% - 4.36% in its calculations of the inundated area and its run-time is 2.17 h and 9.00 h for the 5m and 2m resolutions respectively. It improves computational accuracy and efficiency by 1.07 - 19.55 times and 3.40 - 11.78 times respectively relative to the MIKE21 FM. And this superiority becomes more prominent as the number of grids increases. Thus the GAST model is an ideal tool for modelling flood propagation and inundation.