应用图形处理器实现无功优化并行计算  

Implementation of Parallel Reactive-Power Optimization Computing Using Graphics Processing Unit

作　　者：黄玉龙[1] 刘明波[1] 

机构地区：[1]华南理工大学电力学院广东省绿色能源技术重点实验室,广州510640

出　　处：《电工技术学报》2011年第11期182-190,共9页Transactions of China Electrotechnical Society

基　　金：国家自然科学基金资助项目(50777021)

摘　　要：以求解无功优化问题的内嵌离散惩罚非线性原对偶内点法为基础,利用高性能图形处理器实现了线性修正方程的并行求解。将计算密集部分在图形处理器上实现,其余部分在CPU上执行,并且采用单精度和双精度两种模式进行对照。该算法充分利用了图形处理器强大的并行处理能力和极高的存储器带宽,可获得显著的加速效果。在IEEE 118节点系统和实际538、1133和2212节点系统的计算表明,采用单精度浮点运算的无功优化计算速度最快,加速效果最好,在2212节点系统上的加速比达到近30倍。Based on the nonlinear primal-dual interior-point algorithm with discretization penalty embedded for solving reactive power optimization problems,parallel computing of linear correction equation occurred in optimization process is implemented using a high performance graphics processing unit（GPU）.Computationally intensive parts in full optimization process are executed in GPU and the other ones are done in CPU.Also,single and double precision computing modes are compared.The proposed algorithm makes a good use of GPU＇s powerful parallel processing capability and high memory bandwidth and remarkable speedup benefit is gained.Computational results on IEEE 118-bus system and three real systems including 538-bus,1133-bus and 2212-bus systems demonstrate that reactive power optimization in single precision mode has the fastest speed and the best speedup and indeed speedup reaches almost 30 times on 2212-bus system.

关 键 词：无功优化 非线性原对偶内点法 离散惩罚 并行计算 图形处理器 

分 类 号：TM76[电气工程—电力系统及自动化]