基于矩阵分块递归求逆的电力系统机电暂态并行算法

A parallel algorithm for electromechanical transient simulation by recursive block matrix inversion

在机电暂态仿真计算中,常用的稀疏三角分解存在分解效率不高、并行度低等问题。基于隐式梯形积分法,首先根据雅克比矩阵的特点,将雅可比矩阵分裂,然后利用扩展的Sherrman-Morrison求逆公式递推求解,从而避免了稀疏三角分解,得到了一类新的机电暂态仿真算法。利用高性能的图形处理器(Graphics Processing Unit, GPU)对大规模合成系统进行仿真测试。结果分析表明,所提出的并行算法可以获得较好的加速比和良好的扩展性。

In electromechanical transient simulation, there are some problems in sparse triangular decomposition, such as low decomposition efficiency and low parallelism. Based on the implicit trapezoidal integration method, this paper first splits Jacobian matrix according to the characteristics of the Jacobian matrix, and then uses the extended Sherrman-Morrison inverse formula to solve it recursively, so as to avoid sparse triangular decomposition, finally a novel kind of electromechanical transient simulation algorithm is presented. The large-scale synthesis system is tested and analyzed by using high-performance Graphics Processing Unit(GPU). The results show that the parallel algorithm proposed in this paper can achieve better acceleration ratio and good scalability.