引用本文:吴志勇,王晞阳,陈继林.一种基于FPGA并行加速的稀疏矩阵求解方法[J].电力系统保护与控制,2021,49(11):155-162.
WU Zhiyong,WANG Xiyang,CHEN Jilin.A method for solving a sparse matrix based on FPGA parallel acceleration[J].Power System Protection and Control,2021,49(11):155-162
【打印本页】   【下载PDF全文】   查看/发表评论  【EndNote】   【RefMan】   【BibTex】
←前一篇|后一篇→ 过刊浏览    高级检索
本文已被:浏览 7058次   下载 2773 本文二维码信息
码上扫一扫!
分享到: 微信 更多
一种基于FPGA并行加速的稀疏矩阵求解方法
吴志勇,王晞阳,陈继林
(1.国家超级计算无锡中心,江苏 无锡 214072;2.中国电力科学研究院有限公司,北京 100085)
摘要:
研究了电力系统电磁暂态仿真中最耗时的稀疏矩阵快速求解问题。采用了算法定义架构的设计思想,提出了一种DAG静态并行调度算法,并设计了与之相适配的硬件并行加速阵列架构。在设计实现中,针对电磁暂态仿真运算中稀疏矩阵求解的特性,采用了精确的节拍级硬件资源调度,实现了高度融合的软硬件协同加速。在此基础上,进行了该设计的测试及性能分析。实验结果表明,该结构和方法在电力系统稀疏矩阵运算中的性能优于通用CPU和GPU。
关键词:  稀疏矩阵  DAG  数据流机  静态调度  并行算法  现场可编程门阵列(FPGA)
DOI:DOI: 10.19783/j.cnki.pspc.200948
分类号:
基金项目:国家电网有限公司总部科技项目“适应于电力系统应用的高性能计算技术研究与开发”(XTB17201900305)
A method for solving a sparse matrix based on FPGA parallel acceleration
WU Zhiyong, WANG Xiyang, CHEN Jilin
(1. National Supercomputing Center in Wuxi, Wuxi 214072, China; 2. China Electric Power Research Institute Co., Ltd., Beijing 100085, China)
Abstract:
This paper studies the most time-consuming sparse matrix fast solving problem in the electromagnetic transient simulation of a power system. It adopts the design idea of algorithm definition architecture, proposes a DAG static parallel scheduling algorithm, and designs a suitable hardware parallel to accelerate the array architecture. In the design and implementation, in order to solve the sparse matrix in the electromagnetic transient simulation operation, the precise beat-level hardware resource scheduling is adopted to realize the highly integrated software and hardware coordination acceleration. A design test and performance analysis is carried out. The results show that the structure and method performance is better than general-purpose CPU and GPU in the sparse matrix operation of a power system. This work is supported by the Science and Technology Project of State Grid Corporation of China “Research and Development of High-performance Computing Technology Suitable for Power System Applications” (No. XTB17201900305).
Key words:  sparse matrix  DAG  data-flow computer  static scheduling  parallel algorithm  field programmable gate array (FPGA)
  • 1
X关闭
  • 1
X关闭