摘要
针对海量暂态数据交换通用格式(COMTRADE)数据计算时,由于基础算法和串行编程造成的效率瓶颈,经过试验和改进,优选出了可适应快速计算的递推全波离散傅里叶算法、加窗递推全波离散傅里叶算法和能滤除直流分量的可递推算法。在这3种递推算法的基础上,提出了海量通道数据计算并行化框架,此框架为避免伪共享效应所引起的效率损失,在每个计算线程内对参与计算的每个通道提供了一个独立的"通道高速缓存队列",并提供基于下标映射的虚拟块数据访问机制,形成递推算法可直接访问的窗口数据映射块。试验证明,此递推算法和并行化框架相结合可大幅提升计算效率,并且可随着CPU核心数的增加和计算数据量的加大,获得线性加速比。
In order to improve the calculation efficiency of mass common format for transient data exchange（COMTRADE） data caused by the basic algorithms and serial programming,three fast computation algorithms have preferentially been chosen by tests and improvement：a recursive full-wave discrete Fourier algorithm,a windowed recursive full-wave discrete Fourier algorithm and a recursive algorithm for filtering out the direct current（DC）component.A parallelization framework is proposed for mass COMTRADE data calculation based on the three aforementioned algorithms.To avoid efficiency losses due to the false memory sharing effect,the framework provides a separate channel cache queue for each calculation thread and a data access mechanism based on the virtual block index map,by means of which recursive algorithms can directly access the window data mapping block.Test indicates that high acceleration ratio is achieved and the linear acceleration ratio can be attained with an increase in the core number and COMTRADE data.
出处
《电力系统自动化》
EI
CSCD
北大核心
2014年第1期86-91,共6页
Automation of Electric Power Systems
基金
中央高校基本科研业务费专项资金资助项目(ZYGX2009J089)~~
