面向指数积分方法的电磁暂态仿真GPU并行算法

为满足对大规模可再生能源接入的电力系统进行快速电磁暂态仿真的需求,提出了一种面向指数积分方法的电力系统电磁暂态仿真图形处理器(GPU)并行算法。首先,分析了矩阵指数积分算法求解过程所具有的高度数据并行性,进而将该特性与GPU计算资源相结合;利用GPU处理指数积分方法求解时所需的大规模矩阵运算,而将较为复杂的系统状态判别与更新保留在CPU中完成,有效提升了仿真计算速度。最后,分别针对17台和100台风机的风电场算例进行了测试,验证了所提并行算法的正确性和有效性,同时也说明了算法的加速效果会随着系统规模的增加而愈发明显。

GPU Based Parallel Algorithm Oriented to Exponential Integration Method for Electromagnetic Transient Simulation

In order to realize fast electromagnetic transient simulation for the electric power system with large-scale renewable energy integration, this paper proposes a graphics processing unit(GPU)based parallel algorithm oriented to the exponential integration method for electromagnetic transient simulation. Firstly, the high degree of data parallelism in the solving process of matrix exponential integration algorithm is analyzed. Then, this feature is combined with GPU computing resources. Large-scale matrix operations required by the exponential integration method are processed by GPU while the complex parts such as system status calculation and update remain in CPU. The simulation speed is effectively improved through these measures. Finally, the parallel algorithm is tested in the wind farm with 17 and 100 wind turbines, respectively. The correctness and efficiency of the proposed parallel algorithm are verified. It also shows that the acceleration of the algorithm will be more obvious for large-scale systems.