中文摘要：

首先,介绍了基于直流潮流模型的故障筛选算法;然后,基于GPU的软硬件架构特点,提出了在任务分配、PCIe总线数据交互、线程分配方式、内存访问模式4个方面GPU算法设计应遵循的通用准则;最后,对GPU加速的故障筛选算法进行了详细设计和优化,提出了一种匹配GPU编程架构的并行直流故障筛选算法。算例分析结果表明,所提算法计算9 241节点算例的绝对执行时间仅为188 ms,相对于6核心CPU算法取得了11倍加速;所提4个设计准则具有通用性,可被应用于其他电力系统算法的GPU加速。

英文摘要：

The CS（Contingency Screening） algorithms based on the DC power-flow model are introduced,the general principles of algorithm design based on the hardware and software architectures of GPU are proposed in four aspects：task assignment,data exchange via PCIe bus,thread allocation strategy and memory access mode. A CS algorithm accelerated by GPU is elaborately designed and optimized,and a parallel DC CS algorithm matching GPU programming architecture is proposed. The case calculation for a 9241-bus system shows that,the absolute computing time of the proposed CS algorithm is only 188 ms,ll times faster than that of the CS algorithm based on a 6-core CPU. The proposed four design principles can be generally applied to the GPU acceleration for other power system algorithms.