Research On The Theory And Method Of Flood Simulation At Basin Scale Using Heterogeneous Parallel Computing

As be of an important construction of non-project measures for flood control and damage reduction,flood hazard mapping is a key supporting technology to the transformation from flood control to flood management,and it could provide an effective means for reducing flood disaster.China has implemented a project of flood hazard mapping for important areas during 2013-2015 years.This proj ect mainly assumes some typical conditions of hydrologic and breach parameters,and then uses hydrodynamic models for flood routine simulation and flood hazard evaluation.Also those typical conditions could represent the flood risk situation,the typical situation should be more or less different from the real situation since the rainfall-runoff and engineering states are very complex and time-variable,which limits the practical application effect of flood hazard map.Natural hazard risk has a characteristic of dynamic variation,and the mode restriction condition and simulation results might change significantly according to the change of hydrologic and breach parameters.Real-time flood simulation model at basin scale can use the real-time provided hydrologic and breach parameters,and computing fast to achieve flood simulation results within a short-time,and so it is a great supplement of flood hazard mapping project.However,as restricted by the data condition,compute development level,numerical model,and so on,the fast simulation system for flood routine at basin scale has not been developed in China.Fortunately,the state flood hazard mapping project has collected detail data for important areas,which should provide important support to develop the fast simulation system for flood routine at basin scale.At the same time,parallel computing technologies based on GPU have developed rapidly in recently years,so it realize the high performance computing on personal computer or small workstation.In this thesis,a one-dimensional and two-dimensional coupled hydrodynamic model is developed for flood routine simulation at basin scale,and the model parallelism is analyzed.Further,a fast simulation method for flood routine at basin scale is proposed based on the CPU-GPU heterogeneous parallel computing platform.The main work and innovation includes:(1)The one-dimensional hydrodynamic model for complex river network is developed by using the Saint-Venant equations and the traditional four-point implicit Preissmann finite-difference scheme.The three-cascade method is used to reduce the size of the coefficient matrix.To solve the computational instability problem from sluices calculation,a gate-controlled reach method is proposed.This method can improve the performance for complex river network simulation with sluices operation.(2)The two-dimensional hydrodynamic model is proposed by using the two-dimensional shallow water equations and unstructured Godunov type finite-volume method.According to the common model problems,such as wet/dry treatments,mass conservation,and so on,a new water depth reconstruction method is proposed based on the slope grid model,and a flux-limiting method is used to preserve the property of mass conservation.Those methods can improve the model stability and mass conservation significantly,which are the important issues for flood simulation on complex topography with wetting and drying.(3)Based on the one-and two-dimensional hydrodynamic models,a coupled hydrodynamic model is developed by using lateral coupling technology,and then a Riemann solver based method for lateral discharge computation is proposed.This method should resolve the instability problem from using experience formula of weir flows to compute the lateral discharge,and so improve the model computational efficiency by preserving the computational step.(4)The computational parallelism of the two-dimensional model is analyzed,and then the model is parallelized by using OpenACC programming model for heterogeneous parallel computing on CPU-GPU platform.Further,the data management method of unstructured meshes is optimized for GPU based parallel computation.Those methods should greatly improve the computational efficiency of the two-dimensional flood simulation model at large basin scale,and can provide valid support for numerical simulation with fine grids and large scale.(5)Taking a typical flood protection area on large watershed scale as the object of this study,a GIS-based on-line dynamic flood simulation system is developed by using the service-oriented framework and the proposed flood routine simulation models at basin scale with high-performance computation.The proposed system is used for large-scale floods simulation on real topography.Results show the well performance of the developed system with bright prospect for flood emergency decision support.