基于GPU加速的大电网N-1故障扫描批量计算方法

Batch computing method for N-1 fault scanning of large power grid based on GPU acceleration

随着电网规模的不断扩大,从各种可能的设备开断情况中筛选出严重故障集成为N-1安全校验的重要耗时部分。为了加速大电网N-1安全校验的故障筛选,提出了一种基于中央处理器-图形处理器(CPU-GPU)异构计算框架的实时N-1故障扫描批量计算方法。考虑到不同工况下的计算存在粗粒度并行性,进一步挖掘计算中的细粒度并行性是提高计算效率的有效途径。提出了同时考虑断线故障和发电机开断故障这2种预想事故下的细粒度并行计算方法,并设计了关键计算步骤的核函数。增加考虑了网络拓扑中的移相器,使得计算精度更高,通过与IEEE标准算例和欧洲真实电网算例对比,验证了各工况下批量计算方法的正确性,并取得了显著的加速效果。

With the increasing scale of power grid,the selection of serious contingency from various possible equipment interruptions are becoming an important time-consuming part of N-1 safety verification.In order to accelerate the speed of fault selection of N-1 safety verification of large power grid,a real-time N-1 fault scanning batch calculation method based on CPU-GPU(Central Processing Unit-Graphics Processing Unit)heterogeneous computing framework is proposed.Considering that coarse-grained parallelism exists in computation under different working conditions,it is an effective way to improve computational efficiency to further exploit fine-grained parallelism in computation.A fine-grained parallel calculation method considering both branch break fault and generator break fault is proposed,and the kernel function of key calculation steps is designed.What’s more,the phase shifter is considered in the network topology to improve the calculation accuracy.Compared with the IEEE standard cases and European actual grid cases,the correctness of the batch calculation method under various working conditions is verified,and a remarkable acceleration effect is achieved.