基于图形处理器的电力系统稀疏线性方程组求解方法  

A Method for Solving Sparse Linear Equations of Power Systems Based on GPU

作　　者：周挺辉[1] 赵文恺[1] 严正[1] 徐得超[2] 江涵[2] 

机构地区：[1]电力传输与功率变换控制教育部重点实验室(上海交通大学),上海市200240 [2]中国电力科学研究院,北京市100192

出　　处：《电力系统自动化》2015年第2期74-80,共7页Automation of Electric Power Systems

基　　金：国家电网公司大电网重大专项资助项目(SGCC-MPLG018-2012);高等学校博士学科点专项科研基金资助项目(20120073110020)~~

摘　　要：针对电力系统大规模线性方程组的稀疏特点,提出了基于图形处理器（GPU）的直接求解方法。该方法首先利用基于先排序的分块对角加边形式（BBDF）划分方法对方程组系数矩阵进行分割,形成具有粗粒度和细粒度两层并行结构的线性方程组,然后利用GPU的线程块和线程并行特性对其分别予以求解。将上述方法应用到电力系统暂态稳定计算中,并对其加速效果进行了测试。测试结果表明,在目前普及的设备上,所提方法可获得3~4倍的加速比;在高端设备上,能够获得7~8倍的加速比。In view of the sparsity of large-scale system of linear equations of power systems, a direct method for solving the sparse system of equations based on graphics processing units （GPUs） is proposed. In this method, the coefficient matrix of equations are partitioned by the ordering-first block bordered diagonal form （BBDF） partition method so that two levels of parallel structure are formed for solving the sparse linear equations with Blocks and Threads of GPUs in succession. The partition results are applied to the power system transient stability calculation, and the speed-up ratio performance of the method proposed is tested. The test results show that the proposed method is able to achieve a 3 -4-fold speed-up ratio in currently used devices, and a prospective 7~8-fold speed up ratio in the high-end devices.

关 键 词：电力系统 并行计算 暂态稳定计算 图形处理器 稀疏技术 加速比 

分 类 号：TM744[电气工程—电力系统及自动化]