Research On GPU Parallelization Method For Watershed Hydrology Simulation

Basin hydrological simulation is the primary method for studying hydrological laws and solving water conservancy engineering problems.It plays an essential supporting role in many vital areas such as water conservancy project planning and design,hydrological forecasting,water environment protection,and water resources allocation.However,with the gradual increase in hydrological data,more complex hydrological models,more refined simulation objects,and increasing requirements for water conservancy informatization applications,the computational efficiency of the model has become an important factor restricting the development of hydrological simulation.The hydrological model calibration is a necessary process for the basin hydrological simulation;distributed hydrological models are the future development trend,and urban hydrological models are essential tools to solve current urban water problems.Therefore,this thesis focused on the computational efficiency problems faced by basin hydrological simulation in the above three aspects,introduced GPU parallel computing technology,studied advanced theories and methods for improving the computational efficiency of basin hydrological simulation,and designed and implemented a basin hydrological simulation information service platform.The research results of this thesis are of great theoretical value and practical significance to improve the efficiency of basin hydrological simulation,improve the capacity and level of hydrological services,and speed up the construction of water conservancy information.The main conclusions and innovative results of this thesis are as follows:(1)This thesis proposed an improved hydrological model calibration method,which solved the rainfall-runoff event screening and parameter reuse in traditional hydrological model calibration.The improved method uses the dynamic time warping algorithm to screen the rainfall-runoff events for hydrological model calibration and selects different hydrological parameters for different forecast rainfall events.Case studies showed that the improved method is more adaptable and accurate in hydrological forecasting.GPU parallelization methods of rainfall-runoff event screening and the optimization algorithm are proposed based on the improved method.The GPU parallelization of single-objective and multi-objective particle swarm algorithms is realized.Case studies showed that both Open MP parallelism and GPU parallelism improved the calculation efficiency of hydrological model optimization.Still,GPU parallelism has improved the calculation efficiency more significantly.The acceleration ratio of GPU parallel calculation increased with the increase in the number of algorithm populations.This thesis further proposes a GPU parallelization method for hydrological model calibration under multi-watershed scenarios to improve the calculation efficiency of large-scale hydrological model calibration.Simulation studies showed that the GPU parallel method significantly improved the calculation efficiency of large-scale hydrological model calibration,and the larger the scale of hydrological model calibration,the more it can reflect the advantages of GPU parallel computing.The research results have laid the foundation for large-scale and fast calculation of hydrological model calibration.(2)A distributed hydrological model(Grid-Xinanjiang model)for the Xunhe Basin located in the Han River was constructed,and the hydrological parameters of the Grid-Xinanjiang model were determined using the GPU-based parallel hydrological model calibration method.The GPU parallelization of the runoff yield calculation in the Grid-Xinanjiang model is realized,and the grid layering algorithm that can identify the parallel part of the grid concentration is proposed.Then the GPU parallelization of the grid concentration calculation is realized.Case studies showed that the computational efficiency of GPU parallelization of runoff yield calculation and concentration calculation is significantly improved.The GPU parallel acceleration ratio with the advantage of the number of threads is significantly higher than Open MP parallelism.Based on the above research,the basin layering algorithm under the multi-basin scenario is further proposed.The combined grid layering and basin layering algorithm can identify the parallel parts under multi-basin scenarios,thus realizing the GPU parallelization of multi-basin hydrological simulation under different water systems.The simulation results showed that the GPU parallel method effectively improves the computational efficiency of distributed hydrological models in hydrological simulations in large basins.The research results laid the foundation for rapid hydrological simulation of large-scale basins.(3)Taking the urban area of Suqian City,Jiangsu Province,the SWMM project was constructed as an example.The hydrological parameters of the SWMM model were determined by the GPU-based parallel hydrological model calibration method,and the water movement of the surface and pipeline network in the area was simulated.The basic principle and solution method of pipe network sink calculation with the largest time overhead in the SWMM model are analyzed in-depth,and the parallel part of pipe network sink calculation is excavated.Then the GPU parallel and Open MP parallel of the pipe network convergence calculation are realized.Case studies showed that the speed-up ratio of GPU parallel computing is significantly higher than that of Open MP parallel,and a large number of threads is still the advantage of GPU parallel computing.The research results provide new ideas for improving the calculation efficiency of urban hydrological models.(4)Based on the GPU parallel SWMM model,a model computing server(WEB-SWMM)that can provide online computing services for water conservancy information systems is designed and implemented.WEB-SWMM is based on the B/S structure model,uses Spring MVC and Mybatis as the system framework,and connects to the database in real-time.Currently,WEB-SWMM has been integrated,deployed,and applied in the Suqian Smart Water Project.Application results show that WEB-SWMM can quickly,stably,and accurately provided online computing services for urban water management systems.Compared with the traditional SWMM model,the GPU-based parallel SWMM model improves model calculation efficiency,reduces service response time,and relieves the external access pressure of WEB-SWMM.WEB-SWMM is a typical demonstration of the informatization and servicing of GPU parallel hydrological models.It provides a reference for the parallelization,informatization,and servicing of traditional hydrological models.