基于GPU加速的水文模型参数率定

针对水文模型参数率定问题,为显著提升计算效率,选择SCE-UA算法和新安江模型为研究对象,围绕SCE-UA算法并行化与程序化实现、并行SCE-UA算法在图形处理器(GPU)上的加速效果这两个关键科学问题,以GPU硬件平台和通用计算设备架构(CUDA)软件平台为工具,采用时空复杂度分析、算法并行性挖掘、代码深度优化、数值模拟实验等多种手段相结合的方法,进行了水文模型参数率定提速研究。内容包括:①搭建基于CUDA和GPU的并行计算软硬件平台,进行配置与调优;②并行SCE-UA算法及其程序化实现;③并行SCE-UA算法在GPU上的加速效果。研究结果表明:所提出的方法显著提升了参数率定效率,能够促进水文模拟、最优化方法、计算机科学与技术等多学科的交叉、融合与发展,对水文模拟与预报、防洪快速应急响应具有科学意义和实用价值。

Hydrological model parameter calibration based on GPU acceleration

In the light of hydrological model parameter calibration issue, to accelerate hydrological model calculation efficiency significantly, taking SCE-UA algorithm and the Xin'anjiang model as the study objectives, based on the GPU hardware and Compute Unified Device Architecture (CUDA) software platforms, the two key scientific problems, the parallelization of the SCE-UA algorithm and its program implementation, and the performance analysis of the Graphics Processing Unit (GPU)-based parallel SCE-UA algorithm are studied by using integrated approaches of time-space complexity analysis, algorithm parallelism mining, deep code-level performance tuning, and numerical experiment and simulation, etc. The research contents are as follows: (1) to construct CUDA and GPU-based parallel computing software and hardware platforms and to carry on configuration and tuning; (2) to develop the parallel SCE-UA algorithm and its program implementation; (3) to study the performance characteristic of the GPU-based parallel SCE-UA algorithm. The developed method can improve computational efficiency significantly and the expected achievements of the study will rich the knowledge of hydrological simulation, optimization method, computer science and technology, and be of significance for hydrological simulation and forecast and flood control fast response.