面向GPU的电力系统电磁暂态并行计算方法研究

项目名称： 面向GPU的电力系统电磁暂态并行计算方法研究

项目编号： No.51207076

项目类型： 青年科学基金项目

立项/批准年度： 2013

项目学科： 电气科学与工程学科

项目作者： 陈来军

作者单位： 清华大学

项目金额： 24万元

中文摘要： 现代电力系统正面临动态过程日益复杂和仿真规模迅速扩大的严峻挑战。GPU（Graphic Process Unit）自进入科学计算领域以来，展示出了巨大的计算潜力。为此，研究面向GPU的电磁暂态并行计算方法，在理论和工程两方面均具有重要意义。 本项目拟结合GPU并行计算的特点，将并行计算理论加以发展并应用于电力系统电磁暂态仿真分析。首先，研究影响GPU并行计算效率的关键因素，在此基础上建立面向GPU的并行计算效率分析模型。进一步，从并行网络切分，分解协调计算和计算负载均衡三个方面研究适合于GPU体系架构的多层次细粒度并行算法。最后，以模型和算法研究成果为依据，构建基于GPU的高性能电磁暂态并行仿真平台。 开展本项目研究，有望系统地建立面向GPU的并行计算研究框架，对电力系统并行计算的理论和实践带来重大变革，为满足未来电力系统日益增长的仿真计算需求提供理论基础和技术支撑。

中文关键词： 电力系统；电磁暂态仿真；并行计算；效率分析模型；图形处理器

英文摘要： Modern power systems are facing severe challenges with the increasing complexity in system dynamics and dimension. GPU (Graphic Process Unit) shows a great computation capability since it is adpoted for the general scientific computing. In this case, developing a systematic methodology for GPU oriented electromagnetic transient parallel computing is of great importance in both theory and engineering aspects. The objective of this project is to extend the parallel computing theory for exploiting the huge computing capability of GPU to meet the challenges in power system electromagnetic transient simulation.First, an efficiency analysis model for the GPU parallel computing will be established based on the analysis of the key factors that affect the simulation performance. Further, a fine-grained multi-level parallel algorithms for the GPU architecture will be developed with consideration of network partition, decomposition-coordination and load balancing. Finally, a GPU-based high-performance electromagnetic transient simulation platform will be build according to the efficiency model and parallel algorithm. The project is expected to establish a framework for the GPU oriented parallel computing, make significant changes on the parallel computing theory and provide a theoretical basis and technical support to me

英文关键词： power system；electromagnetic transient simulation；parallel computing；efficiency analysis model；graphics processing unit