于慧莉, 李勤新, 宫春明. 一种基于OpenMP和MPI的非序贯蒙塔卡罗暂态稳定评估的动态混合并行化方法[J]. 现代电力, 2014, 31(2): 31-36.
引用本文: 于慧莉, 李勤新, 宫春明. 一种基于OpenMP和MPI的非序贯蒙塔卡罗暂态稳定评估的动态混合并行化方法[J]. 现代电力, 2014, 31(2): 31-36.
YU Huili, LI Qinxin, GONG Chunming. A Dynamic Hybrid Parallel Algorithm of Assessing the Non-sequential Monte Carlo Transient Stability Based on OpenMP and MPI[J]. Modern Electric Power, 2014, 31(2): 31-36.
Citation: YU Huili, LI Qinxin, GONG Chunming. A Dynamic Hybrid Parallel Algorithm of Assessing the Non-sequential Monte Carlo Transient Stability Based on OpenMP and MPI[J]. Modern Electric Power, 2014, 31(2): 31-36.

一种基于OpenMP和MPI的非序贯蒙塔卡罗暂态稳定评估的动态混合并行化方法

A Dynamic Hybrid Parallel Algorithm of Assessing the Non-sequential Monte Carlo Transient Stability Based on OpenMP and MPI

  • 摘要: 为了提高暂态稳定性分析中非序贯蒙塔卡罗串行算法的计算效率,提出并实现了一种基于OpenMP和MPI的风险评估混合动态并行算法。利用动态规划规则将蒙塔卡罗抽样样本点以MPI方法分配给计算机集群上的多个PC进程,再在各进程内采用多线程的并行计算模式对系统故障的分析处理的循环部分进行OpenMP并行分解。大量实验模型的测试和数据分析表明,在保证准确度的前提下,算法取得了理想的并行效率,普通集群最高达到了3.28倍的加速比,混合集群最高达到了5.06倍的加速比。

     

    Abstract: A kind of hybrid parallel algorithm of assessing risks of power system based on OpenMP and MPI is presented in this paper to improve the efficiency of the large-scale numerical calculation for non-sequential Monte Carlo Simulation in the analysis of transient stability. The Monte Carlo sampling points are allocated in multiple processes of PC cluster based on MPI method by using dynamic programming, and the loop of analyzing and processing system failures is decomposed based on OpenMP by using of multi-thread parallel computing model in each process. The models testing and data analysis show that this algorithm has expectant parallel calculation efficiency with good accuracy. The normal PC cluster can achieve up to 3.28 times speedup, and hybrid cluster can achieve up to 5.06 times speedup by using of proposed algorithm, which verifies its effectiveness.

     

/

返回文章
返回