Computation of Power System Dynamics using Parallel Computers

Parallel algorithms for power system dynamics have attracted enormous attention of researchers, but experience with and the comparison of different algorithmic implementations are still limited. In this paper, the implementation of three typical classes of algorithms, the Gauss-Jacobi, the dishonest Newton and the relaxed Newton (SOR-Newton) on both shared and local memory multiprocessors, are described and cross-compared with the solution timings. The relaxed Newton is shown to be most advantageous. Different implementation strategies, together with the suitability of the algorithms to the machine architectures, are discussed. The implementation on the hypercube requires significant manual programming to schedule the processors and their communication whereas the compilers in the shared memory machines (Alliant, Symmetry) recognize parallel steps but only if software is properly coded. The paper presents these various considerations with detailed test results.