半月刊

ISSN 1000-1026

CN 32-1180/TP

+高级检索 English
应用有向图分层的控制系统暂态仿真并行算法及其GPU实现
作者:
作者单位:

(1. 清华大学电机工程与应用电子技术系, 北京市 100084; 2. 国网青海省电力公司电力科学研究院, 青海省西宁市 810000)

摘要:

随着基于变流器的电气接口和交直流混联技术的广泛使用,电网电磁暂态仿真中需对大规模复杂控制系统进行建模。采用细粒度并行方法可加速控制系统计算,提升电网电磁暂态仿真整体效率。文中提出了一种控制系统细粒度并行仿真算法,加速了图形处理器(GPU)计算平台上大规模控制系统仿真。首先,为构造面向GPU的多线程细粒度并行计算,将控制系统建模为由大量基本控制元件构成的有向图。进一步,对控制系统有向图进行分层,生成控制元件求解顺序,以利用GPU的分组细粒度并行实现控制元件的分层计算。最后,结合GPU的三层并行结构,通过优化线程结构和配置共享内存,将计算线程映射到GPU中的计算资源,最大化控制系统仿真的并行度。对分布式电源接入IEEE 13节点系统的仿真结果对比表明,所提出算法在保证电网电磁暂态仿真正确性的同时,可显著提高GPU计算平台上大规模、复杂控制系统的仿真速度,在硬件资源充足时,不存在仿真规模限制。

关键词:

基金项目:

国家电网公司科技项目“太阳能发电多元综合数据平台及应用技术研究”

通信作者:

作者简介:


Layered Directed Acyclic Graph Based Parallel Algorithm for Control System Transient Simulation and Its GPU Realization
Author:
Affiliation:

(1. Department of Electrical Engineering, Tsinghua University, Beijing 100084, China; 2. Electric Power Research Institute of State Grid Qinghai Electric Power Corporation, Xining 810000, China)

Abstract:

Wide utilization of converter-based interfaces in AC-DC hybrid power system requires precise modeling and simulation for complicated and large-scale control system. However, this process can be accelerated by fine-grained parallel algorithm, hence improving the simulation efficiency. A fine-grained parallel algorithm for large-scale control system simulation is proposed, and a large-scale control system simulation on graphics processing unit(GPU)is accelerated. First, in order to generate massive computing threads, the control system is modeled by a directed graph(DG)with preliminary control elements. Second, by introducing a DG layering algorithm, the control elements are divided into different layers, which can be computed by fine-grained parallel strategies on GPU. Finally, through precise adjustment for thread structure and optimal utilization of shared memory, the computing threads are mapped to GPU cores with highest degree of parallelism. In order to validate the accuracy and acceleration of the proposed algorithm, an IEEE 13-bus distribution system with distributed energy resources integration is provided. Simulation results indicate that the proposed method significantly accelerates transient analysis of control system(TACS)on GPU. In addition, no scale limits are added to the proposed algorithm with adequate computing resources. This work is supported by State Grid Corporation of China.

Keywords:

Foundation:
引用本文
[1]宋炎侃,黄少伟,陈颖,等.应用有向图分层的控制系统暂态仿真并行算法及其GPU实现[J].电力系统自动化,2016,40(12):137-143. DOI:10.7500/AEPS20151127002.
SONG Yankan, HUANG Shaowei, CHEN Ying, et al. Layered Directed Acyclic Graph Based Parallel Algorithm for Control System Transient Simulation and Its GPU Realization[J]. Automation of Electric Power Systems, 2016, 40(12):137-143. DOI:10.7500/AEPS20151127002.
复制
分享
历史
  • 收稿日期:2015-11-27
  • 最后修改日期:2016-04-28
  • 录用日期:2016-03-11
  • 在线发布日期: 2016-04-18
  • 出版日期:
相关附件