城市二维内涝模型的GPU并行方法

针对二维水动力模型应用于城市内涝模拟时，在大尺度区域或精细分辨率情形下运行耗时过长的问题，通过耦合SWMM模型和LISFLOOD-FP模型构建城市内涝模型，采用GPU的并行计算技术加速城市二维内涝模型。以盐城响水县城区的内涝模拟为例，对并行模型的效率进行分析，结果表明，基于GPU的并行计算技术可以显著提升模型运行效率，在5 m分辨率下能够8 min内模拟12 h的内涝事件，可用于突发内涝事件下的快速响应；并行模型的加速效果在更高的空间分辨率下表现更明显，在2 m分辨率下取得最高10.86倍的加速比；要最大化发挥GPU计算效率，首先需要单步长有较大的计算量，其次是要尽量减少与GPU的数据频繁传输导致的额外开销。

GPU parallelized algorithm of urban two-dimensional inundation model

Aiming at the problem that two-dimensional hydrodynamic model is too time-consuming to be applied in a large-scale area or case with fine resolution for the urban flood simulation, an urban inundation model was constructed by coupling the SWMM model and the LISFLOOD-FP model. Then the GPU-based parallel computing technology was adopted to accelerate the constructed inundation model. Taking the flood simulation of urban area in Xiangshui County of Yancheng City as an example, the efficiency of the parallelized algorithm was analyzed. The results show that the GPU-based parallel computing technology can significantly improve the model efficiency, and the parallelized model could simulate a 12-hour flood event in 8 minutes at 5 m resolution, which can be used for quick response to urban flood emergencies. The efficiency of the parallelized algorithm was more obvious at a higher spatial resolution, and the highest speedup of 10. 86 times was achieved at 2 m resolution. In order to maximize the computing efficiency of GPU, a large amount of computation is required in each time step, and the additional time caused by frequent data transmission between host and GPU should be minimized.