Multi-GPU-Based High-Performance Numerical Simulation Method For The Rainstorm Flood Simulation In Cities And Watersheds

The extreme rainstorms are difficult to be forecast and frequently cause floods with huge economic loss.The response capacities of flood forecasting,flood control and drainage systems in urban areas can be improved by strengthening urban flood numerical model computation ability.The majority of rainstorm flood simulation methods focus on the accuracy of simulation and lack of the acceleration performance of high-performance numerical simulation methods.The insufficient calculation performance that exists in present numerical simulation methods of hydrodynamics and its accompanying transportation process,cannot meet the needs of solving more practical problems.In addition,the acceleration calculation performance needs to be further improved.Therefore,through exploring a high-efficiency and high-performance numerical simulation method,the rapid simulation of large-scale rainstorm flood process can be realized.It is also helpful to improve the time and efficiency of flood disaster emergency decision-making.This dissertation studies the rainstorm flood process in cities and watersheds,based on the two-dimensional hydrodynamic numerical model coupled with the hydro-logical process.Applying the multi-GPU parallel computing technology,this work creatively put forward a multi-GPU-based high-performance numerical simulation method for the rainstorm flood simulation in cities and watersheds.The several test cases were used to verify the proposed high-performance simulation method firstly,then the high-performance simulation of actual regional the rainstorm flood simulation was carried out.The acceleration computation performance of multi-GPU-based high-performance numerical simulation method for the rainstorm flood simulation was discussed from multiple perspectives and scenarios.The main results are as follows:(1)A multi-GPU-based high-performance numerical simulation method for the rainstorm flood simulation in cities and watersheds was proposed.The multi-GPU-based high-performance numerical model for the rainstorm flood simulation in cities and watersheds was constructed based on the established two-dimensional hydrodynamic numerical model coupled with the hydro-logical process.CUDA/C++language programming was used to realize the multi-GPU accelerated parallel computing process,and the multi-GPU parallel computing methods of rainfall,infiltration,flow generation and other processes were developed.A domain decomposition method based on structured grid was designed to reasonably allocate the memory and data of the computation unit to different GPUs.The CUDA Stream method was utilized to control the computation of each GPU,which solved the problems of data communication and exchange in the multi-GPU calculation process,and realized multi-GPU-based rainstorm flood high-performance numerical simulation calculation.(2)The computational accuracy and acceleration performance of a multi-GPU-based high-performance numerical model for the rainstorm flood simulation in cities and watersheds were assessed,and the model was tested using the classical ideal cases and experimental simulation method.The good numerical harmony of the model was demonstrated by simulation results based on the hydrostatic case.The rainfall runoff process of a typical V-channel watershed was also simulated.The NSE values of analytical solution and simulation results in slope and gully flow were 0.941 and 0.987,respectively,indicating that the model had the higher calculation accuracy.The relative acceleration ratio compared with multi-GPU and single GPU under two different grid numbers was calculated.The values were 1.27 and 1.57 times respectively,which proved that the proposed method had an obvious effect of accelerating calculation.The experimental runoff-producing and flow confluence process simulation was carried out.Through the comparison of the results of the LID scenario and the traditional development scenario for the typical test case simulation,indicating that the model had a high simulation accuracy and good acceleration calculation performance.(3)The multi-GPU-based high-performance simulation of the rainstorm flood simulation in cities and watersheds was carried out on the practical application,and the accurate simulation of large-scale rainstorm flood process was realized.The acceleration computation performance of the multi-GPU rainstorm flood model applied in the actual area was discussed in this work.Through simulating the process of rainstorm flood in a certain area of Xingxi New City,the established numerical model can realize the rapid assessment of urban flood risk with a significant acceleration calculation performance.The computation efficiency of the simulation results with a grid number of 8.59×105(resolution of 5m)and 5.37×106(resolution of 2m)was quantitatively analyzed,and the relative acceleration ratio and absolute acceleration ratio(considering all scenarios)of applying multi-GPU calculation can reach 10.8～12.6 times and 1.32～1.68 times as much as that of CPU and single GPU.The large-scale rainstorm flood simulation in this area is realized,the result showed that the simulated water depths were consistent with the measured water depth,indicating the the simulation accuracy of the model was high.The rainstorm flood simulation of Wangmaogou catchment was carried out,and the visualization of rainstorm flood dynamic process is realized.When the number of grids of this catchment were 4.52 × 105(resolution of 5m)and 2.83 × 106(resolution of 2m),the relative acceleration ratio and absolute acceleration ratio(considering all scenarios)of applying multi-GPU calculation can reach 4.0～4.3 times and 1.09～1.60 times as much as that of CPU and single GPU.Based on simulation results analysis showed that the more the number of grids in the computational region,the better the acceleration performance of the proposed model.At the same time,the large-scale rainstorm flood simulation in Hulu River basin using the multi-GPU-based rainstorm flood model was realized,and the difficulty of the large-scale regional high-performance simulation was solved.The proposed numerical model can be applied to simulate the effect of water conservation measures on surface runoff control in Hulu River basin,and the effect of vegetation coverage on surface runoff control was analyzed.